Investigations in Universal Grammar: A Guide to Experiments on the Acquisition of Syntax and Semantics (Language, Speech, and Communication)

Acknowledgments

The experimentalmethods and the model of the languageapparatusdescribedin this book were developed over many years, so there are many people to thank. We extend our gratitude first to students and collaborators at the University of Connecticut: Sergey A vrutin , Eva Bar-Shalom, Carole Boster, Paul Gorrell, Anne Halbert, Paul Macaruso, Kazumi Matsuoka, Yoichi Miyamoto , Weijia Ni, Carrie O'Leary, Jaya Sarma, Ana Varela, and especially Cecile McKee, Mineharu Nakayama, and Laura Conway. We also thank more recent students and other collaborators at the University of Maryland at College Park: Peggy Antoinesse, SharonArmon-Lotem, John Drury , Julien Musolino, Susan Powers, Fred Savarese , Caro Struijke, Spyridoula Varlokosta, and especiallyLenaGavruseva. Weare grateful to many colleagues, whose advice we frequently sought and whose encouragementwe always received: Gennaro Chierchia, Peter Culicover, Janet Fodor, Teresa Guasti, Henry Hamburger, Norbert Hornstein, Howard Lasnik, Al Liberman, David Lightfoot, Diane Lillo-Martin , Don Shankweiler, Juan Uriagereka, Amy Weinberg, Jurgen Weissenborn, and Edwin Williams. We profited most of all from extensive discussionswith Ken Wexler, who deservesmuch of the credit for developing the Modularity Matching Model. The true heroes of the piece are the administrators, teachers, and children at the day care centerswhere we conductedour research. We want to mention in particular the Director, Charlotte Madison, and two of our longtime friends, the teachersSue Spencerand Margaret Lima, at the Child Development Laboratoriesat the University of Connecticut. Among our new family at the Center for Young Children at the University of Maryland at College Park, we want to give specialrecognition to the Director, Fran Favretto, and to the Director of Research , Anne Daniel. For helping to take the photos, we are indebted to Julien Musolino. An extraspecialthanks goes to Laura Conway for drawing many of the illustrations. There are four people we cannot thank enough, for providing insightful comments on the entire manuscript: Teresa Guasti, William Snyder, Brian Byrne, and Roger Wales. We end by expressingour heartfelt appreciation to our parents, Garth and Judith Thornton and Bill and Kay Crain, as well as to Stephen's sons, Willis and Austin. Most of this book was written in the year before and the year after the birth of our own LAD, Aurora LuisaThornton Crain. Not only is Aurora free of responsibility for any mistakescontainedherein- she has made the past two years the happiest of our Iives.

Chapter 1 In traduction

This book is an introductory guide to researchon child languageacquisition. It is intended for advancedundergraduates , graduate students, and researchersin cognitive science, especially ones with interest in the psychology of language. Discussion of research methods is couched within the framework of generative grammar , or what is known as the theory of Universal Grammar. In part , we chose this framework because

of its emphasison explaining how children acquirenatural language. A distinguishing feature of the theory of Universal Grammar is that it postulates principles that are specificto grammarformation, rather than attempting to cnaracterizelanguageacquisition using general principles of learning or cognitive growth (e.g., Chomsky 1971, 1975). Taken together, the linguistic principles of Universal Grammar constitute a theory of the organization of the initial state of the mindjbrain of the language leamer- that is, a theory of the human faculty for language. In addition to linguistic competence,performancefactors also contribute to the linguistic behavior of both children and adults. In our view, investigations of language acquisitionmust be framedwithin a specificmodel of performance. To assesschildren's underlying linguistic competence, it is necessaryto understandthe role of the competencegrammar within the performancesystem. Therefore, we will not focus solely on researchon child languageacquisition, but will also discussaspectsof the performance system within which the competence grammar resides. 1.1 Experimental Methods: Two Tasks

Our main motivation for writing this book is to help prepare students of language acquisition to conduct experimentalinvestigations of children's linguistic knowledge. We have not chosen to survey a variety of experimental techniques, however.1 Instead, we discussresearchon child languagedevelopmentby explaining the design featuresof two experimentaltechniquesfor assessingchildren's linguistic competence: one production task and one comprehensiontask. The production technique is known as the elicitedproductiontask. Experiments using this task aredesignedto evokeparticular sentencestructuresfrom children. These structuresare elicited by devising situations that are uniquely felicitous for a specific kind of sentence. For example, an experimentermight want to elicit declarative sentencesthat contain a restrictive relative clause. The experimental situations are constructed with the meaning of the target sentence in mind ; the result of the experiment

is a sentencethat correspondsto that meaning in the child's grammar. In some cases , the sentencesproduced by children will be identical to those produced by adults, but in other cases, children will produce nonadult structures. Children's nonadult

4

Chapter I

productions provide important insight into their grammars, and into the nature of the acquisition processitself. The elicited production task is also used as a tool for ongoing, in-depth exploration of individual children's grammars. Suppose, for instance, that a linguistic principle is thought to have an effect on five different structures. These five structures could be elicited from the same child subjects, to find out if the principle has the predicted effect on children's grammars. It would be insightful to discover whether or not the principle is in effect in all five structures, as predicted. Just as importantly, from a methodological point of view, the properties of each structure can be further investigated using the elicited production task. In this way, a great deal of evidencecan be gathered and a relatively Jlcomplete " picture can be assembledof children's emerging grammaticalprinciples. The comprehensiontechniqueis known as the truth valuejudgmenttask. This task is usedto investigate children's understandingof the meaningsof sentences . It is often of theoretical interest to know whether or not children assign the samerange of interpretations to sentencesthat adults do. The truth value judgment task can be used to tell if sentencesare ambiguous or unambiguousfor children and adults. The distinction between unambiguoussentencesand ambiguoussentencesproves to be crucial in demonstrating children's. adherenceto certain linguistic principles, known as constraints . The demonstration that children know a constraint involves showing that they judge sentencesgoverned by the constraint to be nonambiguous. Children who lack a constraint should find the sentencesthat are governed by it to be ambiguous; the interpretation ruled out by the constraint should be available to these children. In assessingknowledge of linguistic constraints, therefore, it is essentialto understand how children respond to ambiguous sentences . Children's responsesto ambiguous sentencesare used as a yardstick for measuring their performancein responding to sentenceswithin the jurisdiction of the constraint. In addition, it is sometimesworth knowing if children assignfewer interpretations to certain constructions than adults do. Becausechildren are language learners, it is conceivable that they initially hypothesize only a subsetof the adult meaningsand later extend their interpretive options to include ones that were previously absent. The truth value judgment task is a useful tool for this purpose as well. There are several reasonsfor examining child language using these two research methods. First, these two tasks have proven to be especially revealing about children's underlying linguistic competence.Thesetasksareparticularly useful in assessing children's knowledge of principles of Universal Grammar, that is, linguistic principles that are likely to be innately specified. A secondreasonfor discussingthesetwo tasks in particular is that, when used properly, they are relatively free from the influenceof performancefactors that have been found to mask children's linguistic knowledge in other tasks. Properly designedstudiesadopting the truth value judgment task and the elicited production task have resulted in extremely consistent and reliable performance by children- indeed, performancethat is often on a par with that of adults. Even when children's behavior differs from that of adults, the pattern of children's responsesis so consistent that the responsescan be confidently attributed to linguistic knowledge and not to performancefactors. A final reasonfor focusing on these two tasks is that they are, in large part, our own inventions; we know them well, and we have refined them over the years to make them better tools for evaluating children's

Introduction

5

growing linguistic competence. We continue to use these tasks almost exclusively in our own research .

By looking at the methodological prescriptions that have been administeredusing these tasks in our own prior research, we hope to convince a greater number of students and researchersin child language of the basic correctnessof the Innateness Hypothesis and the theory of Universal Grammar. No one will be convinced, or should be, by researchthat is improperly designedor poorly executed, or by findings that are open to alternative interpretations. The methodological formulas presented in this book are designed to overcome many of these obstaclesto the empirical investigation of children's linguistic competence. Researchthat follows these methodological formulas will therefore have a better chanceof being taken seriously, even by critics of the theoretical assumptionsthat underlie the particular linguistic bent of the researcher. Of course, no single experiment can control every potentially confounding factor or rule out all but one explanation of the findings. However, this should not deter researchersfrom attempting to conduct sound and tightly controlled empirical investigations of child language. At the very least, the findings from carefully designed studies can be replicated and cannot be dismissedas artifacts of improper proceduresor unnatural experimentaltasks. 1.2 Structure and Process

The book has three parts. Part I, encompassingchapters 1- 16, is an extended discussionof the fundamentalassumptionsthat guide researchin child languageacquisition. Researchdoes not take place in a vacuum. Both the research questions of interest and the methods that are employed to answer these questionsdepend on a theoretical framework . This framework includes not only a theory of grammar , but

also a set of assumptionsabout how grammar is embeddedin a performancesystem. As indicated earlier, we will assumethat children's linguistic competenceis as described by the theory of Universal Grammar. In part I, we outline certain linguistic phenomenaas they are viewed within this theoretical framework. The phenomenawe describeare characteristicof the researchtopics in child language that we return to later in the book. We also explain why studies of children are especiallypertinent to the investigation of theselinguistic phenomena. In addition to an explicit theory of linguistic competence,part I introducesa model of languageprocessing. In our view, this is essentialto the investigation of children's knowledge of linRuistic principles . Research findings from studies of child language

cannot be properly understood without an appreciationfor the languageperformance system within which linguistic principles operatefor children and adults. It is beyond the scopeof this book to delve deeply into all aspectsof the performancesystem, but we will present many of its relevant aspects, on the basis of our own conception of

the language apparatusand with support from the findings of experimental investigations of both children and adults .

A description of the generaloperating characteristicsof the performancesystem is crucial to any seriousinvestigation into children's knowledge of linguistic principles. It is frequently lamented that there is an inherent tension between linguistic theory and the kind of performance data that are obtained in experimental researchwith children (e.g., Cook and Newson 1996, 310). It is commonly held that linguistic theory is "pure" and that performance data are "messy." As a theory of an ideal

6

Chapter 1

speaker

-

hearer

itatively

'

linguistic

-

it

cate

,

"

the

competence

is

.

"

data

(

factors

.

One

of

discuss

a

Competing

advocate

a

linguistic

performance

forthcoming

)

processing

.

-

.

Finally

and

adults

tralinguistic

sentence

is

for

ambiguous

sentences

,

to

the

task

ambiguous

and

diction

as

language

.

,

behave

in

adults

If

as

the

than

adults

do

,

' s

larity

must

be

,

the

ways

according

to

cal

findings

findings

and

The

the

chapters

model

I

first

of

glance

.

possible

in

linguistic

un

But

-

,

to

question

.

of

,

and

to

It

are

children

consistent

and

possible

sources

the

children

,

and

and

Modu

and

important

the

,

can

reexamine

to

differ

empiri

.

with

-

adults

therefore

adults

model

-

differently

of

review

with

that

is

pre

perform

the

children

both

typically

both

perspective

of

for

extraneous

should

,

because

,

factors

adults

course

the

inconsistent

differences

Of

in

adults

empirical

by

performance

Model

of

further

found

from

into

be

a

been

limited

to

-

a

performance

that

and

.

performance

,

ex

understanding

linguistic

marred

quantitatively

brought

. ,

used

with

as

children

the

s

makes

have

quite

linguistic

. g

When

The

the

-

relevant

model

are

.

part

of

at

both

children

Matching

,

range

part

then

'

typically

in

is

the

,

both

( e

.

sometimes

or

same

model

not

differences

~ Modularity

in

views

the

which

seem

the

in

or

in

that

discussed

and

are

any

children

knowledge

and

,

,

1984

of

competition

producing

the

The

linguistic

,

in

Pinker

ambiguity

are

more

.

Model

.

see

factors

information

,

,

processing

apparatus

is

in

further

same

(

,

-

knowledge

a

language

not

Wexler

language

the

exception

essentially

,

their

responses

explained

describe

)

and

common

make

to

the

one

however

14

be

maintains

Model

We

optimal

Crain

nonlinguistic

The

are

is

qualitatively

Matching

-

to

mistakes

both

,

13

sentences

nonadult

because

have

to

appeal

of

performance

make

children

aduJts

a

on

preempts

.

factors

expected

with

sometimes

of

.

representations

(

understanding

sources

or

model

,

children

goes

of

Matching

conformity

it

language

Nonlinguistic

linguistic

that

but

adults

instances

unambiguous

well

,

and

chapters

is

-

.

share

assume

do

and

Modularity

experimental

exam

language

linguistic

adults

architecture

most

see

,

certain

interference

subjects

Model

and

model

children

children

(

in

performance

child

research

between

children

this

nonlinguistic

.

-

model

carefully

the

of

by

-

advo

this

and

Matching

adults

modular

in

we

how

experimental

knowledge

)

either

According

any

,

as

grammatical

ambiguity

knowledge

ternative

,

ambiguous

the

just

the

knowledge

resolving

model

that

to

,

. studies

Modularity

production

owing

model

explain

in

in

we

On

. only

knowledge

interrelations

does

,

that

.

factors

experimental

is

system

children

this

to

there

processing

Assuming

of

that

performance

is

the

only

language

,

children

of

Not

notion

with

-

the

performance

language

performance

the

this

:

of

)

.

qual

abandoning

the

grammatical

optimal

called

to

assumption

8

)

in

of

,

principles

mechanisms

14

competence

Acquisition

model

.

-

discussion

in

According

grammatical

contentious

the

Language

specific

system

of

12

' s

Without

the

studies

by

desideratum

of

with

of

in

affected

reduced

result

Models

and

,

of

fundamental

We

y

.

and

model

even

person

task

issue

between

be

that

is

. 3

to

goals -

strategies

performance

the

versel

friction

can

a

any

grammar

1965

the

that

take

,

ad

Chomsky

factors

we

anticipated

of

assumed

in

to

not

sites

is

competence

is

see

particular

A performance

it

performance

According

are

grammar

the

s

,

the

embedded

circumstances

ine

'

distinction

between

messy

,

person

performance

tension

which

7

behavior

that

competence

herent

1

s

outstrips

the

I

also

interrelation

compare

the

between

Modularity

Matching

linguistic

knowledge

Model

with

and

other

two

factors

al

-

Introduction

7

that contribute to linguistic behavior: environmental input, extralinguistic knowledge, verbal short-term memory, and so on. One viewpoint that contrastswith the Modu larity Matching Model is called the Input Matching Model. This model assumesthat children's grammars develop largely in responseto their linguistic input. Empirical reasonsfor questioning the Input Matching Model are presentedin chapter 5. That chapter presentsevidenceof both children's nonadult productions and their nonadult interpretations of sentences . Another alternative view of language and mind is called the CompetingFactors Model. This model is pervasivein the literature on languageacquisition. It is assumed by most researchersworking within the generative framework, even by researchers who share with us many fundamental assumptions about the nature of linguistic knowledge, including the InnatenessHypothesis. Theseresearchersdo not shareour assumptionsabout the relation between the.competencegrammar and linguistic performance, however. Researcherswho adopt the Competing Factors Model view linguistic knowledge as only one among several componentscontributing to linguistic behavior in any experimentaltask. Therefore, on this model, less than perfect performanceis expectedin studieswith children. This viewpoint has becomeso deeply ingrained that most researchersseemto assumeit without argument. On the Modularity Matching Model, by contrast, accessto and application of linguistic principles preempts the influenceof other factors. On this model, therefore, children's performance should parallel their linguistic competencein many instances. The Competing Factors Model is presented in detail in chapter 6. Chapter 6 also discussesboth areaswhere the Competing Factors Model and the Modularity Matching Model converge, and areaswhere they diverge. That chapter explains our reasonsfor questioning the utility of researchdesignsassociatedwith the Competing Factors Model for the study of child language. Chapters 8 and 9 review many empirical findings that seem to bear out the predictions of the Competing Factors Model; these chaptersand the two that follow them, chapters10 and 11, outline the variety of research designs that have been employed by the model's advocates. Chapters 12 and 13 present our responseto the Competing FactorsModel. There we argueagainstthe basicassumptionsof the model: that memory limitations (chapter 12) and extralinguistic knowledge (chapter 13) interfere with children's accessto linguistic principles. Chapter 14 focuseson one differencebetween the languageapparatusof the child and that of the adult: namely, that children have accessto a languageacquisitiondevice (LAD ).2 This differenceleadsto someinteresting empirical consequencesand explains certain apparent differencesbetween child and adult linguistic performance. Finally, part I concludeswith a consideration of what can go wrong in assessingchildren's linguistic knowledge (chapter 15) and a myriad of practical details in administering experimentswith children (chapter 16). We partition models of language and mind into three broad classeslargely for purposesof exposition. Making this partition permits us to comment on certain tasks and researchstrategiesthat are commonly used in the study of child language, but ones that we argue are inappropriate for studies of child language within the generative framework. In our view, the use of these tasks and researchstrategiesimplies the (implicit) acceptanceof the Competing FactorsModel. Becausedifferent research strategies are suggestive of different models, we have chosen to paint in broad strokesthose researchstrategiesthat are associatedwith the different models.

B

Chapter 1

We hope to convince researcherswho would adopt the Modularity Matching Model that it is also necessaryto adopt its methodological assumptionsto assess children's knowledge of the principles and parameters of Universal Grammar. It seems to us that many researchersadopt the methodological assumptionsof the Competing FactorsModel simply becausethey are not aware of alternativesand becausethey do not realizethat the assumptionsof that model are inconsistentwith the theoretical framework they embrace. This book is our attempt to clarify the relation betweenmodels of child languageand researchmethodology. We argue that many of the researchstrategies of the past should be abandonedin favor of new research strategiesthat are more in keeping with a model of languagecompetenceand performance that is based on the theory of Universal Grammar. One such model is the Modularity Matching Model. In making the casefor a new set of researchstrategies, we inevitably take issue with researcherswho shareour commitment to incorporating insights from the syntax and semanticsof generativegrammar. We do so preciselybecausewe believe that this commitment demandsdifferent researchstrategies. Through a critical examination of researchby those of like concern, we hope to convince investigators "within the family'! of the need to adopt alternative researchstrategies. Part II of the book (chapters 17- 24) covers the elicited production paradigm, and part III (chapters25- 40) covers the truth value judgment task. Becausethese are dynamic tasks and must be modified to addressspecificresearchquestions, we provide detailed discussionof a number of experimentsusing eachparadigm. We include discussion of past and present researchto illustrate particular design features of these tasks. A brief disclaimeris called for. Although we review the theoretical backgroundfor the syntactic, semantic, and discourseprinciples that are the subjectmatter of the experimental investigations we describe, this book is not intended to replacea solid introduction to linguistic theory.3 Readerswithout a sufficient theoretical background in linguistics may find that we presupposea firmer grasp of theory than they have. We do not believe, however, that lack of familiarity with any of the theoretical machinery we employ will interfere with proper understanding of the important points of methodology, and this, after all, is what the book is about.

Chapter 2 Constraints and Universal Grammar

Universal Grammar is a theory of the human biological endowment for language, that is, those aspectsof linguistic knowledge that are innately specified. Severalempirical hallmarks indicate that a biological property is innately specified. We use the term hallmarksbecausea linguistic property could be innately specifiedwithout manifesting certain of the relevant characteristics . One hallmark of innateness is that a property appears in the absence of decisive

evidencefrom the environment. If children adhereto -a linguistic principle for which there is no correspondingevidencein the environment, then the principle is likely to be innately specified. In caseswhere evidencefor a linguistic principle is available in the primary linguistic input, the evidence could actually lead learnersastray, should they try to avail themselvesof it . We will discusstwo such casesin this chapter. A related hallmark of innate specificationis that children acquirea linguistic principle despite considerablelatitude in the primary linguistic data they encounter. If all (or at least all normal) children in a linguistic community adhereto a particular principle, despite being exposedto different input, then innate specificationis suggested, particularly if the principle is highly complex (from a pretheoretic vantage point). Another hallmark of innatenessis universality. If a linguistic principle is part of the human biological blueprint for languagegrowth, then it is expectedto be manifested by children learning any natural language- hence the name Universal Grammar. There are caveatsto this expectation, however. In certain instances, an innate linguistic principle will not be expected to be exhibited in all languages. For example, the parametersof natural language(e.g., the null subject parameter) are consideredto be innately specified. Although the various settings of a parameter may be innately given, one setting may be manifestedby one classof languages, and another setting by another classof languages. Moreover, it is not necessaryfor every option to be manifestedduring the course of development in a single language. If the initial value of a parameteris consistentwith the target language, then other values will never be adopted; they will simply atrophy, or go wherever unused parameter values go. However, linguistic principles that are not parameterizedare expectedto appearin all languages, and in all children.1 A final hallmark of innatenessis sometimessuggested: early emergence. It is not logically necessaryfor innate principles and parametersto emergeearly in the course of development. Just as some properties of physical development are biologically timed to appear months , even years, after birth (e.g., the secondary sexual character-

istics), so certain aspectsof linguistic knowledge could become operative only at a certain stage of development. In fact, two specificmaturation accountsare currently at the forefront of child language research: Borer and Wexler (1987, 1992) propose

10

Chapter 2

that the capacity to form A -chainsmaturesin children, and Radford (1990) and others propose that certain functional projections mature. Regardlessof the outcome of the debates on these -proposals , one should bear in mind that it is consistent with the InnatenessHypothesis for even (seemingly) highly complex linguistic principles to be part of the young child's language faculty. Early emergenceof linguistic principles could be construedas additional evidenceof innate specification . ... The alternative to innatenessis learning; yet the early emergenceof seemingly complex linguistic principles castsdoubt on many learning-theoretic scenarios. This makes child language a good testing ground for innateness: it provides an opportunity to see whether or not particular linguistic principles meet the IIearly emergence" hallmark of innate specification. 2.1 The InnatenessHypothesis

These joint expectations- that innate principles (a) emerge early, (b) are universal, and (c) appearwithout decisive evidencefrom the environment- will be referred to as the Innateness Hypothesis . Much of the researchreported in this book is concerned with this hypothesis. Becausethe focus is the study of child languageacquisition, the researchwe discuss examinestwo of the hallmarks of innateness: early emergence and the mastery of linguistic principles by children in the absenceof corresponding experience. We interpret the results of this researchas support for the InnatenessHypothesis, generally, and as support for the theory of Universal Grammarin particular. The remainder of this chapter outlines the kinds of linguistic phenomenathat fall within the confines of the theory of Universal Grammar and that are consequently explored in this book. 2.2 Constraints

Unlike much current research, ours will not focus on what children say or on what interpretations they assignto sentences . Instead, the emphasiswill be on what children do not say and on what meanings they do not assign to sentences , when there are grounds for thinking that they might say these things and assign these meanings. The InnatenessHypothesis is designed to solve this puzzle- why children do not produce sentencesthat they might be expectedto produce, and why they do not assign meaningsthat they might be expected to assign. The relevant linguistic knowledge comesin the form of constraints , of which this chapter will introduce two kinds. Constraints of both kinds have been investigated using the experimental methods describedin subsequentchapters. Constraints are subject to the argument from the poverty of the stimulus, which concludesthat knowledge of constraints cannot be learned and therefore is likely to be innately specified. The argument from the poverty of the stimulus is the topic of the next chapter. The remainderof this chapter will be devoted to constraints. 2.2.1 Sentences and Meanings To begin the discussion of constraints, we introduce some terminology. Following Aristotle, we view languageas pairings of sound and meaning. In more current terms, language is conceived of as a (psychological) mapping between sentencesand their

11

Constraints and Universal Grammar

associated

meanings

ciated

with

some

on

sentence

(1 )

Clearly

an , can

a critique

of of

experience more

;

they

a

By

the

as

which follows

sentences

are

asso

-

:

and

meanings

( e . g ., parental

could

input

be

) . For

evident

to

example

, con

-

likely a

couch

the

basis

often

assumed on

assumption

meaning

;

is be

is

between

one third

to

the

encountered

particular

are

acquired

,

meaning on

. It

the

of

and 1990

for

of

could

be

through

that

some

attained

certain

that

experience

have

sentences are

unlikely

other

sentences

sentences

seems

basis

Gleitman knowledge

that certain

where poorly

sentences

see

, that

meanings

knowledge

situations

admittedly

between ( but

knowledge

although

in is

mappings

, however

and the

the

learners

process

experience clear

is is

"

such less

that

a

by the

" positive of

mapping

second

.

. Although

even

knowledge

given

be

couch

sentences

this a

the

that

) . It

of

turn

is on

mastered

mapping

on

. We

well

these will

are

ill

formed

,

aspects

consider

of each

.

Ambiguity

the

time

speakers

adult

of

English

grammar ( the

have

) , they

alternative

are

reached able

to

the

final

judge

interpretations

state

that

are

of

language

sentences

indicated

by

like the

development ( 2a ) and

paraphrases

( 3a ) are in

( 2b

)

( 3b ) ) .

(2 )

(3 )

can

a.

We

fed

her

chicken

b .

We

fed

her

some

a.

They

seem

b .

They

enjoy

represent

( sentence This

learning

this

data

blanket

this

aspect

ambiguous

We

Indian

. One

in

( i . e ., the

and

an

knowledge

them

2 .2 .2

is be

cannot

linguistic

, is

sentences

linguistic

blanket

this

than

formed

between

primary

like

it

meanings

of

is

Indian

understood

aspects

, then

abbreviate

(1 ).

There

is

mapping

the

, a sentence

there

can

)

, the

based

sider

. We

, meaning

instances

learners

a grammar

meanings

( sentence

In

. Learning

which

is

representations

read

II A

sentence

vs .

enjoy it .

of

vs .

ambiguity

' meaning2 is

We

boiling

doing

knowledge

, { meaningl ,

to

McNuggets .

associated

. fed

it

some

champagne They

.

enjoy

using

.

drinking

the

it .

following

notation

:

} ) with

a

set

of

( in

this

case

,

two

)

semantic

."

2.2.3 Constraints on Form Extending the terminology , we can represent the knowledge that certain sentences are ill formed . For example, the verbal elements want and to may be contracted to form wanna in many linguistic contexts , but they cannot be contracted in certain other contexts . In (4)- (7), we provide examples of constructions in which want and to may be contracted . This is followed by example (8). In (8a), contraction of these elements makes the sentence ill formed ; only (8b) is well formed . Notice that much of the evidence available to someone learning English runs counter to the constraint exemplified in (8a). Contraction of want and to is tolerated in general- (8a) is the exception .

12 Chapter 2 Therefore , if learnerswereto adoptthe traditionalprinciplesof induction, they would be temptedto violate the constraint . The relevantempiricalfacts about children learningEnglishwill be reportedbriefly in chapter3, andin detailin chapter21. (4) a. Who doesArnold wannamakebreakfastfor? b. Who doesArnold want to makebreakfastfor? (5) a. DoesArnold wannamakebreakfastfor Maria? b. DoesArnold want to makebreakfastfor Maria? (6) a. Why doesArnold wannamakebreakfast ? b. Why doesArnold want to makebreakfast ? (7) a. I don't wannamakebreakfastfor Arnold or Maria. b. I don't want to makebreakfastfor Arnold or Maria. But: (8) a. *Who doesArnold wannamakebreakfast ? b. Who doesArnold want to makebreakfast ? Knowledgeof factslike that in (8a) (i.e., aboutdeviantsentence forms) canbe representedasfollows: *sentence The "star" (*) indicatesthat the sentence is not well formed- it is deemedto be ungrammatical by somecomponentof grammar . The knowledgethat certainsentences areill formedis represented in the grammarby a constraint . A statementof the constrainton contractionof wantandto is presentedin chapter21. Because constraintsloom largein our discussionof experimental methodology , we will discussthe notion of a constraintin further detail. First, it shouldbe notedthat a constraintis a linguisticprinciplethat governsa rangeof linguisticphenomena , not just a smallset of sentences . Any sentencethat is similarto (8a) in the relevant respectswill be subjectto the sameconstraint . Second , a constraintis a prohibition againstcertainsentence /meaningpairs. The addition of a constraintto a grammar resultsin an overall reductionin the language(sentence /meaningpairs) that the grammargenerates (seeFodorandCrain1987). To seethat constraintsaresanctions againstcertainsentence /meaningpairs, considerwhat would happenif the constraint on contractionof wantandto wasremovedfrom the granunarof English. Let us call the languagethat results, Shmenglish . Lackingthe constrainton contractionof want and to, Shmenglishwould includemore sentences , with their associated meanings , than English. Shmenglishwould generatesentences like (8a), whereasEnglishdoes not.2 English (sentencel ) Shmenglish (sentencel , sentence2 ) Childrenwho lackedthe constraintwould be expectedto allow contractionin questions like (8a) roughly asoften asthey do in questionslike (4a), for example . To test the constraintunderdiscussion , the elicitedproductiontaskhasbeenusedto encouragechildrento producequestionslike (8a) aswell asoneslike (4a). Thismakesit possible to performthe neededcomparisonbetweenthe proportion of contractionsin children's questionslike (8a) andthe proportionof contractionsin questionslike (4a).

Constraints and Universal Grammar The findings

are briefly

summarized

13

in chapter 3, and fuller details of the study are

described in chapter 21 . Before we proceed , there is a small wrinkle to iron out . It may be possible in many instances to understand certain sentences that are not well formed , if these were actually

produced

by a speaker in some conversational

context . A speaker who pro -

duced (8a), for example , would probably be understood perfectly well , although the utterance itself would sound odd . According to the theory of Universal Grammar , however , such sentences could be produced

only by mistake ; that is , they would

be a

product of the performance system , not the grammar . However , grammatical knowl edge is embedded within a performance system with a specific architecture , such that there are severe limits on the range of performance mistakes . Whether or not people mistakenly contract want and to, for example , is an empirical question . Our own re search suggests that speakers do not make such performance mistakes (see chapters 13 and 14 ). The relation between the competence grammar and the performance sys tem will be discussed further as we proceed . 2,2,4 Constraints on Meaning There

are also constraints

formed

on the meanings

sentences . In an interesting

some sentences may not be interpreted

that can be assigned

to certain

set of cases, speakers of a language in a particular

know

well that

way . For example , note that the

pronoun he and the name the Ninja Turtle can pick out a single individual in (9) and (10 ). By contrast , these same linguistic expressions cannot pick out the same individ ual in sentence (11 ). In the examples , if one NP is underlined and another NP is not , they refer to different individuals ; if both NPs are underlined , they refer to the same individual . (9)

While he danced the Ninja Turtle ate pizza . a. While ~ danced the Ninj a Turtle ate pizza . b.

(10 )

While

~

danced the Ninja Turtle

The Ninja Turtle a. b.

ate pizza .

danced while he ate pizza .

The Ninja Turtle The Ninja Turtle

danced while ~ danced while ~

ate pizza . ate pizza .

But : (11 )

Knowledge

He danced while the Ninja Turtle ate pizza . a. .y.& danced while the Ninja Turtle ate pizza . b. & danced while the Ninja Turtle ate pizza . of what a sentence cannot mean is represented

(sentence , { meaningliy

as follows :

.meaning2 } )

Since the sentence in question is well formed , we do not attach the asterisk to the sentence portion of the sentence / meaning pair . Rather , we are noting that some par ticular meaning cannot be assigned to a well -formed sentence ; in (11 ), the pronoun he cannot designate the Ninja Turtle - it must pick out some individual who is not mentioned in the sentence . Although ( 11 ) is a grammatically well formed sentence , it has only one of the meanings associated with (9) and (10 ). As the asterisk (~) in dicates , the interpretation according to which the pronoun he and the name the Ninja Turtle have the same referent is ruled out . That is, (9) and (10 ) are ambiguous ; in each

14

Chapter 2

case , the pronounmay, but neednot, havethe samereferentasthe name.In (10), by contrast , theseelementsmusthavedisjointreference ; the sentence is unambiguous . It is worth noting that muchof the evidenceavailableto learnersrunscounterto disj oint reference ; coreference betweenpronounsand namesis toleratedmuch of the time, as attestedby examples(9) and (10). Example(11) is the exception . Children who adoptedthe traditionalprinciplesof induction, therefore , couldeasilybe misled aboutthe rangeof interpretationsavailablefor sentences like (11). Noncoreference factssuchasthe oneexhibitedin (11) areattributedto a principle known as PrincipleC. This principleprohibits coreference betweena pronounand a namewhenthey arein a certainstructuralrelationship(seechapter26). PrincipleC is a constraint . Recallthe criterionfor decidingwhetheror not a linguisticprinciple countsasa constraint : addinga constraintto a grammarresultsin an overallreduction in the language(sentence /meaningpairs) that it generates . PrincipleC meetsthis criterion. To seethis, supposethat a dialectof English(i.e., Shmenglish ) lackedPrinciple C. If so, Shmenglishwould includemore sentence /meaningpairsthan English does. Specifically , Shmenglishwould generate(lIa ). Shmenglishwould differ from Englishasfollows: English (sentence , {meaning } , "'meaning2 }) Shmenglish (sentence , {meaning } , meaning2 }) In discussingthe constrainton wannacontraction , we noted that the utteranceof a sentence that violatesthe constraintwould soundodd to the ear; never :theless , the sentencecouldbe understood . What aboutviolationsof PrincipleC, the constraint on coreference ? Clearly, the utteranceof a sentencelike the one in (11) would not soundodd, because the constraintgovernsthe interpretationof (11), not its form. The constraintprohibitsinterpretation(a). As we have shown, childrenwho lack the constrainton coreference shouldfind sentences like (12) to be ambiguous ; both readings , (a) and (b), shouldbe availableto thesechildren. (

12

)

He

danced

a

.

b

Therefore

,

.

)

the

and

27

. 3

The

The

be

one

.

&

danced

while

the

Ninja

Turtle

ate

pizza

.

would

the

adults

12

.

(

)

12

The

to

)

( i

the

as

(

. e

9

. ,

should

)

( b

)

be

in

10

)

.

should

of

By

be

unambiguous

the

12

)

In

the

fact

coreference

children

(

only

to

be

be

coreferential

)

who

know

just

as

reported

in

presented

in

the

it

is

un

of

of

3

chapters

26

(

12

Context

sentence

(

in

)

to

describe

12

a

a

)

could

be

abnormal

conversational

picture

in

context

in

which

another

.

there

For

is

sense

example

only

:

,

one

individual

-

chapter

.

Importance

-

accept

noncoreferential

children

will

will

the

children

these

in

should

corresponding

available

itself

the

,

,

findings

study

.

contrast

for

empirical

details

accept

(

accept

(

)

incorrectly

NP

they

and

relevant

methodological

inappropriately

uses

often

like

(

expected

and

as

of

utterance

used

be

pronoun

sentences

Sentence

for

and

2

pizza

interpretation

ambiguous

.

ate

roughly

.

. pizza

Turtle

in

( deictic

ate

Ninja

interpretation

constraint

Turtle

the

children

interpretation

Ninja

while

involving

this

the

danced

these

terpretation

while

&

namely

/

suppose

that

,

the

it

could

some

Ninja

,

Constraints and Universal Grammar

15

Turtle, who is dancing while he is eating pizza. Although it would be infelicitous for anyone to use (12) to describe such a picture, this kind of infelicity is not uncharacteristic of experimental investigations assessinghow children interpret sentences . In such experiments , children would be asked to indicate whether or not sentence

(12) correctly describesthe picture. On the basis of our own experience, we believe that many children would answer affirmatively, indicating that (12) is an accuratedescription of such a picture (seechapter28). Is such a finding evidence that these children lack Principle C7 Not necessarily. It may simply be evidence that these children ignore or override their knowledge of the constraint on coreferencein order to comply with the experimenter's request. The constraint on coreference pertains to sentences that are encountered in conversational contexts with (at least) two individuals , the Ninja Turtle and another male character; in such contexts , the pronoun cannot refer to the Ninja Turtle . Notice also that the

use of the pronoun he implies that the speakerpresupposesthat the other male individual has previously been introduced into the conversationor is highly salient in the context. Therefore, the application of the constraint in the context of the picture under discussion would require children to accommodatea failed presupposition. Specifically, they would need to modify their mental model of the conversational context by adding the secondindividual. What if children are less able than adults to accommodatepresuppositionalfailures (see chapters 12 and 15)7 If so, they would be compelled to construct a semantic representation that does not require accommodation - one that is consistent with the context . In the picture we discussed, which was shown in association with (12), the only consistent semantic representation is one in which the pronoun and the name refer to the same individual could

be forced

to violate

. This line of reasoning the constraint

leads to the conclusion

on coreference

, not

because

that children

it is absent

from

their grammars, but because of the processing difficulty associatedwith accommodating failed presuppositions.3 This hypothetical experiment is no doubt an extreme example of improper experimentation, but it allows us to make a point: that children's failuresto comply with a linguistic constraint could arise even if they had mastery of the constraint. Failures could arise because children are not as well versed as adults in recovering from prag matic infelicities . This should hardly be surprising , given children 's more limited experience in such matters , not to mention the fact that both children 's and adults'

experiencelargely consistsof sentencespresentedin felicitous contexts. In any event, the observed

differences

between

children

and adults would

not reside within

the

languagefaculty. These observations underscorethe importance of proper experimental design in assessingchildren's grammaticalknowledge.4 If experimentalsentencesare presented in inappropriate circumstances , the wrong conclusionsmight be drawn. In the present example, the researchercould erroneously conclude, from their IIYes" responses , that many children lack Principle C. Justthe opposite conclusionmight be drawn if children were tested in felicitous circumstances . In an experiment using the truth value judgment task, for example, a sentencelike (12) would be presenteain circumstancesthat makeit felicitous on both of the readingsat issue: (a) the reading that conforms to the constraint on coreference, and (b) the reading that would result if the constraint were absent from a child 's grammar . Among the felicitous contexts is one in which someone other

than the Ninja

Turtle -

say , Grover -

refused

to dance while

the Ninja

16

Chapter 2

Turtle

ate

and

pizza

eating

.

pizza

incorrect

,

the

the

is

.

In

should

( 12

ter

in

had

just

of

a

on

final

of

one

kinds

1995

) .

( 13

a

contain

( 13

)

No

)

However

In

same

addition

sentence

as

that

( NPs

negative

to

.

For

example

be

anaphorically

)

in

a

( 14

)

every

Simba

' s

that

seem

major

-

con

-

felicity

of

of

parts

II

party

.

He

was

Simba

' s

party

.

* &

No

mouse

came

to

~

' s

party

.

&

to

Simba

' s

party

.

He

b

.

Every

mouse

carne

to

~

' s

party

.

&

)

sequence

is

which

.

~

:_ Q ~

b

.

A

bear

~ r

worth

.

)

This

into

,

are

applies

to

.

.

upset

.

upset

.

pronoun

can

sentence

contrasts

Genie

' s

into

Genie

sleepwalked

into

g

illustrates

mouse

.

was

was

singular

preceding

sleepwalked

and ( 16

a

a

no

certain

Chierchia

;

with

the

that

be

is

related

,

discourse

the

to

an

following

sequences

in

.

noting

negatively as

in

ambiguous

sleepwalked

a

is

not

! ! ! ! : Q~

that

appears

ambiguous

are

bear

demonstrates

that

as

to

pronoun

to

constraint

upset

* &

,

such

upset

was

.

( 15

singular

( see

upset

was

party

bear

limited

.

was

' s

a

,

same

upset

to

! ! ! ! Q ~

the

not

.

Simba

,

in

to

such

related

the

came

came

,

NPs

shows

mouse

mouse

it

role

refusing

focus

sentence

to

A

the

-

character

the

are

preceding

carne

)

not

that

constraints

mouse

) ,

charac

this

alter

are

quantificational

Example

quantifier

came

well

cannot

phrases

.

that

Every

( 14

,

to

discourse

note

sequence

universal

example

.e

and

to

.

,

-

judgment

have

changes

a

NP

( IS

related

.

wish

No

Every

,

indefinite

to

shows

mouse

.

b

( 14

)

we

discourse

constraint

the

a

)

apply

a

noun

such

that

discours

but

of

Example

to

but

found

design

con

sentences

active

if

,

have

these

experimental

to

an

contexts

because

used

second

What

pizza

We

was

value

the

played

.

ate

?

truth

,

pizza

experimental

largely

the

above

)

.

instance

Turtle

of

of

,

quantificational

subject

NPs

,

sentence

of

For

ate

responses

structure

clarification

grammar

in

of

.

-

part

the

"

is

,

this

( 12

" No

it

non

context

know

presupposition

properties

Ninja

' s

features

of

the

which

;

the

that

who

respond -

failure

On

noting

in

should

Turtle

.

false

Discourse

point

sentence

These

worth

dancing

is

of

children

described

the

,

is

,

omitted

Ninja

children

responses

It

)

pizza

description

.

-

basic

were

while

ate

be

( 12

.

Constraints

As

( 13

.

they

context

the

correct

pizza

any

the

affect

in

book

-

could

,

experiment

the

while

Turtle

constraint

Turtle

however

accommodate

dancing

' s

ate

,

details

this

children

the

he

a

hypothetical

experimental

not

sentences

III

the

Ninja

the

Ninja

be

judgment -

dance

would

on

. 4

,

would

describe

differences

target

he

by

it

the

the

experiment

to

to

-

sequences

the

in

correct

)

refusing

minor

and

as

,

with

while

while

important

( appropriate

,

either

ingly

judgment

;

dance

paragraphs

stood

dance

same

the

many

the

story

the

moreover

complies

dance

true

did

having

although

the

not

be

the

preceding

,

would

without

,

that

did

make

) -

The

task

)

context

reading

Turtle

context

like

2

( 12

Ninja

infelicitously

straint

same

the

Grover

reading

because

the

On

because

adult

of

In

.

that

universally

the

quantified .

~

house

.

' s

k

He

ate

house

' shouse

singular

spaghetti

ate

the

spaghetti

.

.

&

ate

the

spaghetti

.

he

if

.

&

pronoun

NPs

the

.

these

can

elements

be

anaphorically

appear

in

the

Constraints and Universal Grammar

17

(16) No/ Everymouseat Simba 's party saidthat ~ wasupset. a. No/ Everymouseat Simba 's party said~ wasupset. b. No/ Everymouseat ~!!!!~~'s party said~ wasupset. MoreoverI a plural pronoun can be related to precedingquantificationalNPs in differentsentences in a discourse . (17) No/Everymousecameto ErnieandBert's party. Theywereupset. a. No/ Everymousecameto ErnieandBert's party. They wereupset. b. No/ Everymousecameto ErnieandBert's party. ~ wereupset. On the basisof theselast three examples , childrenwho lackedthe constraintin evidencein (13) and (14) might be expectedto permit anaphoricrelationsbetween the singularpronounheandthe quantificational NPseverymouse andnomouse . Lacking the constraint , childrenmight find discoursesequences like (13) and (14) to be ambiguous , whereasthey areunambiguous for adults. The situationcanbe depicted asfollows: Child (discoursesequence , {meaning1 ' meaningz }) Adult (discoursesequence , meaning1 ) Furthermore , childrenwho lackedthe constrainton discoursewould be expectedto allow anaphoriclinks in discoursesequences like (13) and (14) roughly as often as they do in sentences like (15)- (17). To test children's knowledgeof the constrainton discoursebinding, the truth valuejudgmenttaskwasusedto comparethe proportion of anaphorachildrenassignedto discoursesequences like (13) and(14) with the proportion of anaphorathey assignedto sentences like (15)- (17). The resultsare presentedin chapter34. 2.5 Conclusion This chapter introduced two kinds of linguistic constraints . One kind of constraint encompasses knowledge that certain sentences are ill formed ; the other encompasses knowledge that certain well -formed sentences (or discourse sequences) cannot be interpreted in a particular way . Both kinds of constraints can be the exception , rather than the rule, in the linguistic input to children . Children should be expected to vio late some linguistic constraints , then, if they learn constraints on the basis of the input . In the next chapter, we will follow up on this observation in order to argue that constraints are part of Universal Grammar . As noted at the beginning of this chapter, Universal Grammar is a theory of the initial state of the human language faculty . The conclusion of the argument presented in the next chapter, therefore , is that constraints are part of the language faculty from its inception .

Chapter3 The Poverty of the Stimulus

How

do

guists

speakers

have

acquire that

Grammar

. This

conclusion

the

of

present

topic Stated

two

the

very

All

premises

of

and

the

is

not

to

of

poverty

the

aspect

have

conclude

P must

be

been

of

-

Universal

the

stimulus

,

stimulus

begins

with

of

their

language

, call

it

learned

on

the

basis

of

the

C .

innately

- of - the - stimulus

premise

specified

(i .e ., part

of

Universal

, which

argument

has

often

been

to

linguistic

constraints

challenged

in

the

. It

literature

.

knowledge

for

knowledge

data

such

many

of

which

stimulus

first

, or

claims theories

' s linguistic

the

data

of

can the

be

applied

following

consists and

the

certain

and

which

to

a range

quotations

to

child

their are

of

from

relations

the

in of

acceptable

fact

language

linguist

but

are

not

sentence

are

- types

as

linguistic Hornstein

lie

evi -

complex

outside

the

systematically , that

never

no

. Crucial and

ungrammatical

I sen -

primary

lin -

informed a given

paraphrases may

which

children

concerning , and

ungrammatical

sentences

for

exposed

judgments

. Children

are sets

of

of they

ambiguity

available

to

that

structure

to

sentences

experience

these

are

available

legitimate

guistic

the

, paraphrase

hypothetical

ambiguous

the

the

of

in

' , all

guistic

of in

:

sentences

tences

poverty

available

evidence

child

psycholin

of

.

invited

poverty

the

attain

dence

ductive

poverty

particular

P could

are

is illustrated

1981

People

If

the

some

property

second

from

Lightfoot

some

know

data

of

the

. This

rare

the

part

Coverage

phenomena and

from

and

, as

).

applies on

argument

argument

from

property

B , we

Knowledge

chapter

The

argument

linguistic

A

concentrates

Empirical

the

linguists

specified

.

speakers

Grammar

3 .1

on

? Many

innately

P.

Knowledge

(C )

This

is based

, the

native

primary From

constraints are

.

property (B )

of

constraints

chapter

generally

premises (A )

knowledge

concluded

of

each

occur

in

that

sentence other a child

is

, and ' s lin -

. (pp . 9 - 10 ) are

correct

must

be

experience

, then

it

follows

abandoned does

not

as

a simple

, because

there

provide

the

is no basis

point

of

inductive for

logic

that

in -

base . If

establishing

the

some

20

Chapter 3 particular aspect of our linguistic knowledge , there will have to be some other source for that knowledge . That aspect of our linguistic knowledge will have to be a priori in the sense that it arises independently of experience. (p. 12) Given these facts, it is reasonable to look for a priori knowledge available to the organism , which permits language acquisition to circumvent the environmental deficiencies and thus to take place. (p. 13)

The phenomenon of interest in this chapter is what Hornstein and Lightfoot describe as "judgments concerning . . . ungrammatical 'sentences.' " In our terms, these are judgments about linguistic constraints . Let us turn , then , to the application of the poverty -of-the-stimulus argument to linguistic constraints , invoking the two constraints introduced in chapter 2. 3.2 The Innatenessof Constraints The first premise of the argument from the poverty of the stimulus is that native speakers have knowledge that is encoded by constraints . This fact is not contested, as far as we are aware.l The second premise is that linguistic constraints could not be learned on the basis of the primary linguistic data. This premise hinges on the claim that there is no information in the environment corresponding to linguistic constraints . This claim merits further discussion. As stated earlier, constraints are concerned with the ungrammaticality of sentences. They are sanctions against certain ways of putting a message, or sanctions against assigning certain meanings to sentences that are, themselves, well formed .2 It is conceivable that constraints could be learned by children , assuming the usual mechanisms of induction , only if the relevant kind of evidence is available . This evidence is called negative evidence(or negative data). Negative evidence is the presentation of un grammatical sentences, marked as such. If negative evidence were available to children , they could learn constraints on the basis of their experience . When Hornstein and Lightfoot claim that IIchildren are not systematically informed that some hypothetical sentences are in fact ungrammatical ," they are claiming that negative evidence of this kind is not systematically available to learners. If Hornstein and Lightfoot are correct in asserting that children lack access to negative evidence, then it follows that children 's knowledge about the ungrammaticality of sentences (i .e., constraints ) is not learned . Hence, this knowledge is known independently of experience; presumably , it is innately specified. 3.3 Negative Evidence Other acquisition scenarios are often suggested, however . According to this line of thinking , negative evidence is available, but in more subtle forms than the kind of negative evidence mentioned by Hornstein and Lightfoot - overt negative judg ments about hypothetical sentences. A more realistic acquisition scenario is often suggested, which goes as follows : First , children violate a constraint on form , such as the constraint on wanna contraction , producing nonadult utterances (e.g., IWho does Arnold wanna make breakfast?" ). In response to these errant forms , parents would provide corrective feedback: I'No , say, ' Who does Arnold want to make breakfast.' "

Poverty of the Stimulus

21

Negative evidence could be even more subtle. Parentscould provide negative feedback simply by failing to understandtheir children's nonadult utterances; they could supply negative evidence by expanding their children's nonadult utterances; and so

on

.3

Basedon these observations, a good deal of researchhas been conducted, and a great deal of ink has been spilt, on the question of whether some sourceof negative evidenceis availablein the primary linguistic input. This question is potentially misleading, however. It is one thing to determine whether or not negative evidence of some relevant kind exists; it is quite another thing to determine whether sufficient negative evidence is available at the relevant time (s) to ensure that all children converge on the target grammar .

Unquestionably, all children master the kinds of linguistic constraintswe have described, such as the constraint on contraction , as claimed in the first premise of the argument from the poverty of the stimulus . As far as we know ; this premise has not been contested . For negative evidence to guarantee that all children learn all of the

linguistic constraints found in the adult grammar, however, it would have to be abundantly available in the primary linguistic data. If negative evidencewere scarce, then somelearnerswould not encounterenough of it and would not converge on the target grammar . Since this is contrary to fact (as stated in the first premise), all of the

ingredients necessaryfor convergencemust be available in sufficient quantity (see Lasnik and Crain 1985 ).

As far as we can ascertain, if negative evidence is available to children at all, it is

not available in sufficient quantity or at the right times to guaranteethat every child converges on the adult grammatical system. Several researchershave reviewed the literature on the availability of negative evidence and have reached the same conclusions that we have : first , that no source of negative evidence is systematically

available to all learners; and, second, that the potential substitutes for negative evidence that have been identified (e.g., expansions) are not available throughout the courseof development (they may even occur lessfrequently at those stagesof development at which they would be most useful). Here is how Pinker (1990) summarizes

the findings: [WJhen parents are sensitive to the grammaticality of children's speechat all, the contingency between their behavior and that of their children is noisy , indiscriminate , and inconsistent For other reviews

that reach similar

from child to child and age to age . (p . 217 ) conclusions , see Bowerman

1987 , 1988 , Brown

and Hanlon 1970, Morgan and Travis 1989, Marcus 1993. Even if negative evidence were available, there is no guarantee that children would

use it . Many researchershave pointed out the absenceof findings demonstratingthat children who are exposed to negative evidence (in experimental settings) use it to jettison incorrect grammaticalhypotheses. Studiesby Cazden(1972) and by Nelso~,

Carskaddon , and Bonvillian (1973) found that children who receivedexpanded parental input fared no better through the course of language development than children who did not. Explicit correction is even more rare than expansions (e.g., Brown

and Hanlon

1970 ), and when it does occur , there is little reason to believe that

children benefit from it . This is attested by familiar anecdotes that point to children 's resistance

to correction

.

22

Chapter 3

Child My teacherholded the rabbits and we patted them. Parent Did you say your teacherheld the baby rabbits? Child

Yes .

Parent

What did you say she did ?

Child Sheholded the baby rabbits and we patted them. Parent Did you say she held them tightly ? Child No, she holded them loosely. (Cazden 1972 )

Child Parent

Nobody don 't like me. No , say "nobody likes .me."

Child

Nobody don't like me. (Eight repetitions of this dialogue)

Parent No, now listen carefully; say "nobody likes me." Child Oh! Nobody don't likes me. (McNeill 1970)

Child

Want other one spoon, Daddy.

Parent

You

Child

Yes, I want other one spoon, please, Daddy.

mean , you

Parent

Can you

Chi ld

Other

want

the other

say " the other

spoon .

spoon " ?

. . . one . . . spoon .

Parent

Say " other ."

Child

Other .

Parent "Spoon." Child Spoon. Parent

" Other

Child

Other . . . spoon . Now give me other one spoon ?

spoon ."

(Braine 1971 )

As these examples illustrate, children are often unable to figure out what adults intend when they explicitly correct their speech. On the basis of current research findings, it seemssafeto infer that the universal achievementof languageacquisition is accomplishedwithout negative evidence.4 The debate surrounding the availability of negative evidencefocuseson children's nonadult linguistic behavior- for example, children's overgeneralization of the rule for past tense (resulting in forms like holded). We should ask, therefore, whether children violate constraints . The conclusion of the argument from the poverty of the

stimulus is that knowledge of constraintsis part of the initial state. If this conclusion is correct, then children should not be expectedto violate constraints at any stage of language development. If children do not violate constraints in the first place, then the question of the availability of negative evidenceis moot. 3.4 Investigating Knowledge of Constraints

Much of our own researchhas focusedon children's knowledge of putatively innate constraints. As far as possible, we have investigated children s knowledge of constraints that would be most susceptibleto violations, if these constraints were not

Poverty of the Stimulus

23

part of children's grammars . The two constraintsthat we introducedin chapter2 are good examples . First, considerthe constrainton wannacontraction . As we noted, therearemany constructions in which~ant andto mayor maynot contract,asin (1). (1) a. Who doesArnold wannamakebreakfastfor? b. Who doesArnold want to makebreakfastfor? Children, like adults, shouldbe expectedto contractthe verbalelementswantandto frequentlyin such constructions . This tendencytoward contractioncould follow from the generaldispositionof speakers to use"reduced " formsasmuchaspossible . This tendencyis pitted againstthe constrainton <;:ontraction in the exceptional construction . (2) a. *Who doesArnold wannamakebreakfast ? b. Who doesArnold want to makebreakfast ? Lackingthe constraint , childrenwho appliedthe commonstrategiesof induction, suchas analogyor stimulusgeneralization , would be expectedto producenonadult questionslike (2a) at the stagesof languagedevelopmentwhen they first attempt suchquestions . It is againstthis backgroundof what childrenwould be expectedto do, given certaintheoriesof learning, that children ' s behaviorwill be examinedin subsequent chapters . Thegoalwill be to seeif childrenadhereto linguisticconstraints or if they insteadadopt learningstrategiessuchas analogyor stimulusgeneraliza tion, consequently producingsentences and assigningmeaningsthat are prohibited by constraints . Chapter21 reviewsthe findingsof a study that was designedto investigatehow young childrenform questionswith the verbalelementswant and to, using the elicitedproductionmethodology . The mainfindingsarethese: childrenshoweda general willingnessto contractwhere the constraintis not applicable , but the same childrenproducedalmostno contractedformsin constructions like (2a), wherea constraintprohibitscontraction . This patternof resultsis just what the Innateness Hypothesispredicts.It alsoshowsthatnegativeevidenceis not relevantin theacquisition of the constraint ; children's adherence to constraintsobviatesthe needfor negative evidenceand arguesagainstthe acquisitionscenarioaccordingto which children learnconstraintsandmakeerrorsin the courseof learning. Considernext the constrainton meaning , PrincipleC. This constraintlimits the interpretationsthat canbe assignedto sentences like (3). PrincipleC renders(3) deviant whenthe pronounandthe namearecoreferential , asindicatedin (a). The only legitimateinterpretationof (3) is one in which the pronounand the namehavedisjoint reference . Children(and adults) who know PrincipleC shouldprohibit coreference in (3). (3) He dancedwhile the Ninja Turtle atepizza. a. *& dancedwhile the Ninja Turtle atepizza. b. & dancedwhile the Ninja Turtie atepizza. This restrictionon interpretationdoesnot hold in othercases , asin (4). (4) While he dancedthe Ninja Turtle atepizza. a. While ~ dancedthe Ninja Turtle atepizza. b. While ~ dancedthe Ninja Turtle atepizza.

24

3 Chapter

Lacking Principle C, children who apply the usualleaming-theoretic strategieswould be expected to permit the illicit interpretation of sentenceslike (3), on analogy with the sentencesin (4), assumingthat sentenceslike (4) are part of the primary linguistic data. If so, sentenceslike (3) would be ambiguousfor these ehildren, just as (4) is for adults. It is important, therefore, to observehow children respond to ambiguoussentences like (4) and to comparetheir pattern of performanceon such sentenceswith their responsesto sentencesthat are governed by the linguistic constraint, as in (3). The pattern of responsesto sentenceslike (4) could be used as a yardstick for sentenceslike (3). If children know the constraint, then sentencesthat are governed by the constraint, such as (3), should be unambiguous. In short, the researchdesign we have in mind pits ambiguous control sentencesagainst the (potentially) unambiguous test sentences . The question is whether or not children respond to sentenceslike (3) on par with ones like (4). To put it another way, the researchquestion is whether or not children find both types of sentencesambiguous. There are two possible outcomesfor any child subject. Children who lack the constraint should find both types ambiguousand should acceptcoreferencein response to both to a similar extent. Children who know the constraint, on the other hand, should not tolerate coreferencebetween the pronoun and the namefor sentenceslike (3), although this interpretation might even be their preferred interpretation of ambiguous sentenceslike (4). Chapter 26 reviews the findings of a study designedto investigate young children's interpretations of sentenceslike (3) and (4), using the truth value judgment task. The main findings are that children show a general willingness to assign coreferencein sentenceslike (4), but they consistently reject the coreferential reading of sentenceslike (3), where a constraint prohibits coreference. This pattern of results is as predicted by the InnatenessHypothesis. These findings also provide another indication that negative evidence may be irrelevant in the acquisition of constraints; as noted earlier, children's consistent adherenceto constraints obviates the need for negative evidence. This arguesagainst the acquisition scenario according to which children learn constraints, making errors along the way. Because young children consider sentencesgoverned by Principle C to be unambiguous, just as adults do, it follows that negative evidence is not needed for the acquisition of Principle C. Similar considerationssuggest the hypothesis that young children will also adhere to discourseconstraints, as illustrated in (5). (5) No mousecameto Simba's party. He was upset. a. No mousecameto Simba's party. JI .& was upset. b. No mousecameto !!!}Q~'s party. & was upset. Again, the alternative supposition is that children will judge discoursesequencessuch as (5) to be ambiguous. The findings are reported in chapter34. 3.5 An Alternative : Conservative Learning There is a different way to avoid the conclusion that children 's knowledge of constraints is innately specified. This is to abandon the claim that children apply the standard strategies of induction , such as analogy . It has been proposed , for example, that children restrict the sentence forms they produce and the meanings they assign

Poverty of the Stimulus to sentences cording have

to ones

to not

this

that

themselves

have

encountered

, learners

encountered

seen to avoid producing are conservative in this learning encoded

they

viewpoint

do

not

in the

in the primary

postulate

linguistic

sentence

environmental

sentence patterns that they sense , then constraints would

( 1995 ), within

Interestingly pronoun

, van

, as in (6 ) . A

pretation

Cognitive

constraint

of (6 ) in the adult

(6 )

h! ! ! ! .

~ apa Bear is covering

him .

Grammaticality

the language

or

to

if it

coreference

conflict

sanction it

NP the

is possible

new

that

proposal

and

a

(a) inter .

- established

schemas . As noted

. . . , speakers

to actually

expressions

as an example

is unacceptable

of

if it

- occurring

.... A

:

a particular

conventionally

exposure

is judged

, but

(which is advanced

a referential

B , prohibits

the following

judgments

with -

through

through

schemas

ceptable

with

, Principle

advances

constitute

in

established

or templates

those

(6 ), van Hoek

judgments with -

are

him .

b.

like

knowledge ) has been

the (b ) interpretation

)(oPapa Bear is covering

congruent mas

; only

a.

to sentences

sentences

coreference

grammar

they

, children

.

involves

on

Papa Bear is covering

Referring

Grammar

Hoek ' s example

that

have not yet heard . If children not be necessary for language

. This kind of account of the acquisition of linguistic in constraints in the framework of Universal Grammar

by van Hoek

data . Ac -

meanings

input ; similarly

25

expression

is

patterns

in

acquire

expressions

sentence

a pronominal

such

as [(6 )] is ac -

construction

is understood

to

sche -

, and use

involve

without coreference

.

(p . 337 ) On

this

view , constraints

learned basis

responses of their

They

too

argument learners .

by

to

denying

[P ] erhaps

prohibitions

proposal the

grammar

, perhaps

evidence is irrelevant

and Rosen

constraints

that

is advanced

conclusion

actually ample

consists

positive

can support

learning about

out , according

children by Rosen

of

are statements

to questions point

on sentence

patterns

circumvent

the mental

positive

innate

. A similar

that

sentences

mark

are not

" schemas ," linguistic

experience attempt

license

As Rosen

to

the

are

on

the

and Rosen

( 1995 ) .

- of - the - stimulus

grammars

of positive

. If so , then innateness

, but

poverty

in the

evidence

meanings internalize

CONSTRUCTS

is available the

perhaps

evidence

and Rosen

learning

that

, and

negative

. . . . (Rosen

to a conservative

of language

hall -

1995 , 73 )

account

children

should be inherently unresponsive to II construct " for them . That is , children

sentences until they have built the relevant should reject all sentence / meaning pairs that

are not

. The

their

experience

do not bear out this prediction

consistent

, however

( 1990 ), children (7) The Papa

were

presented

with

Is Papa Bear covering

question Bear

with

was

in ( 7 ) would shown

watching

Papa

Bear ; this

is a m .onkey

pictures

been himself

and were

presented , and

Bear . First , the experimenter " ); then

from

studies

study asked

of child

by Chien

questions

language

and Wexler like

(7).

him ?

have

covering

findings

. In a well - known

children

were

in the context

a monkey introduced asked

was the

the question

of a picture shown

standing

characters in ( 7 ).

(" This

in which nearby is Papa

,

26

Chapter 3

In

response

to

answered of

" Yes

the

time

referent the

( its

antecedent

rect

meaning children that

Let

[ T ] he

child

child

The

evidence

strategy

,

to

do

with

which

" extra

the

not

"

cor

,

, be -

. Several ,

-

himself

meaning

-~ eneration

but

at -

we

will

acquisition

of

.) , then

have

no

con

-

concerning they

should

not

corresponding

con

-

:

the

coreference

in

/ picture

the

' s judgments

, etc

they

remark

construct

a

way

pairing

interpretation

that . ( pp

of

that

be

. Because

describes

. 78 -

the

picture

, the

79 )

interpretations

learners

in

invoke

of

(8 )

,

,

Therefore

, it

a

response

to

conservative

same

who

learn

-

than

that discourse

formance

the

children

's

stantial

evidence

seman

,

for

tic

him

children overgeneration

against

the

1990

Bear

(8 )

construct

are

conspire

to

the

respond in

conservative

enough

"

expe

-

appropriate

against

which

to

. plausible

response

children

, to

who

judging

however

.

related

sen

incorrectly

( 8 ) to

be

as

questions

-

an -

an

inaccurate

to

(8 )

.

least

as

template to

only

this

).

corresponding

( 8 ) . In

same

at

or

. Being

factors like

the

,

to factors

the

is

in

appeal

had

nothing

them

errors

( 8 ) , correctly

. Papa

like the

template

performance

that to

( Thornton

factors

sequences

factors

context

"

be

respond

unknown

not

findings

few

)

developed

reject

recalcitrant make

found " No

that

performance

stipulates

been

sequences

unlikely

the

to

by

have

would

reason

to

knowledge

( 1995

impaired

have

fail

linguistic

Rosen

simply ( 7 ) to

, there

children

covered

discourse is

has

is

like

no

children their

and

children

template

of

that

( 7 ) answered

the

earlier

unknown

it

that

therefore

that to

performance

is

explanations

is

. Rosen

questions

this

noted

I know

Presumably

" their

One

extraneous

system

( 6 ) or

and

.

reasons

rejoinder like

example to

for

that

. Without

these

"

rejoinders "

performance

sentences

of

description

than

of

children

, schemas

for

proposal

possible

possible

" Yes

develop

account

overgeneration

the

, stating

should

. For

swered

on

the

template

Neither

tences

the NP

assign

, but

overgeneration

, if

sentence

sentence

" constructs

with

such

it

to the

' s semantic

sentences

or

compare

,

an

covering

interpret

to

6

half

, referential

else

over

and

possible

not

able

able

assign

3

only is

preceding

also

semantic

) acknowledge

( 6 ) , they

able

the

are

their

( p . 80 ) . Another

First

the

someone

semantic

( templates

like

be

children

explanation

schema

's

sentences

interpret

there

to

rience

are

learning

( 1995

( 7 ) undermines

course

of

children

of

a

of roughly

Bear

apparently

covered

one

because Papa

to

children

is

,"

ages

answer

.

according

kind

are

the

this

, and

children

had

( 7 ) . Because

constructs

not

reject

of

" No

, linked

these

error

is

monkey

children

Bear

conservative

for

will

like

having

on

cannot

should

questions

Of

the

sentences

child

to the

Rosen

based

. Concerning

,

that

explain

interpretations

the

ing

to and

are

then

Papa

"

the

gave

.

return

illicit

if

, to

speakers is

between

' s study

pronoun

noted

" Yes

made

Rosen

coreference

struct

be . So

here

now

. As

accept

should

been

adult

picture ,

children

Wexler

reflexive

speakers

them

us

a

pronouns

adult

have

for

the

young

and

adults

were

answered of

review

straints

it

) . It

these

in

) . Unlike

if

to

yond tempts

him

as

( 7 ) , many

Chien

answer

pronoun

(= him

in

correct

the

monkey

like

Children

. The

for

pronoun

not

questions

."

rare

corresponding

( 7 ) . This

leaves

a promissory to

make

absence

of

differently questions learning

note questions

any

good to

like

(7 ) scenarios

the , this like

reason

examples represents proposed

like

account

based

account ( 7 ) more , based like

(7)

strong in

(7 ).

would

merely difficult on

per

and circum the

-

(8 ),

liter

-

Poverty of the Stimulus

27

ature. We will further substantiatethis view by reporting several examplesof children's nonadult productions in chapter 5. The present example of children's semantic overgeneration illustrates a further point. There are two horns to the dilemma of languageleamability: one is how children recover when they have overshot the target language, the other is how children avoid undershooting the target language. One way to resolve the first horn of the dilemma would be for children to be conservative, thereby avoiding the problem associatedwith syntactic and semantic overgeneration. Such conservative learning models are then left facing the secondhorn of the dilemma, however. As we show in chapter 5, moreover, children are not conservativelearners. This leavesus to reconsiderthe solution to the overgenerationproblem advanced by the theory of Universal Grammar: overgeneration is held in checkby constraints. The secondhorn of the dilemma is taken care of by generalprinciples, namely, operating principles, such as moveand copy, which are governed by constraints. 3.6 Summaryof the Poverty-of-the-StimulusArgument This completesthe discussionof the argument from the poverty of the stimulus. The argument can be depicted using the notation introduced in chapter 2. The situation confronting the languagelearneris indicated in the following graphic:

Input (primarylinguisticdata) -t LAD -1- Finalstate (sentence, meaning)

(sentence, meaning) (sentence,{ meaningl,meaning2} ) (sentence,{meaningI' *meaningz} ) "'sentence

As the graphic indicates, certain aspectsof the linguistic knowledge mastered by learners are acquired in the absenceof input (== primary linguistic data). We have mentioned two linguistic phenomenathat bear this property: knowledge that certain linguistic forms are prohibited (as in the caseof wanna contraction), and knowledge that certain meaningsare prohibited (as in the caseof disjoint reference(3)). A similar argument can be given for the acquisition of knowledge about ambiguity (e.g., Hornstein and Lightfoot 1981). Eachof these aspectsof the final state (= adult grammar) is such that it must be masteredin the absenceof decisive evidence from the environment6This invites the inferencethat such facts are not learned, but are part of the human biological blueprint for language acquisition (i.e., the LAD ). In short, the logical problem of language acquisition is that the data available to learnersunderdetermine what they come to know. The solution offered by proponents of the theory of Universal Grammar is that the knowledge is innately given, as part of the LAD itself. Children are preprogrammed to adhere to these principles of linguistic analysisas part of the blueprint for their development. Just as a child cannot help but grow fingers and toes, and not wings or claws, so these linguistic principles, and not others, grow in the child. 3.7 Conclusion By adulthood, at least, language users have mastered a rich and complex linguistic system. A large amount of the linguistic knowledge encoded in this system is

28

Chapter 3

apparently acquired without corresponding evidencefrom the environment. Among the facts that do not seem to be learned on the basis of experienceare knowledge that certain sentenceforms and sentencemeaningsare prohibited. The principles that underlie this knowledge are constraints. In the absenceof negative evidence in the primary linguistic data, it is likely that constraints are innately specified. That is, it is likely that constraints are part of Universal Grammar. If the goal is to investigate which principles are part of Universal Grammar, and which are not, it is now clear why children are ideal subjects. As a consequenceof innate specification, the possibility exists that linguistic knowledge may emergeearly in the courseof development. If children evince knowledge of a principle, knowledge that cannot have been learned from experience, this evidence reinforces the view that

the principle is part of Universal Grammar. The theory of Universal Grammar is essentially a biological theory; it is concernedwith those aspectsof the human biological endowment for languagethat could emergeearly in development, despite the

lack of evidence for them in the environment

.

Chapter 4 Models of Language Development

Following

the

designed the

,

,

we

getting

into

the

clear

of

its

in

are

other

details

the

apparatus

that

of

characteristics This

of

model in

the

advocate

4 . 1

The

The

,

Modularity

language to

According are

to

sealed

off

syntactic

mechanisms

world

knowledge

plausibility

" funny Cats

"

express dogs said

or

a to

true

mean

something

,

take do

proposition

false

it

systems

.

of

of work

,

operating

forthcoming

.

adopted

describe

more the

model

Fodor

1983

for

for a

).

example them

simple

not If

do

.

This

they cats

did

,

to

why

they

people

would ,

they

read

the

the writer

of

general process about

cog

-

real

-

the

sentences hear

to

interpret take

or

a

to

sentence

make

the

sentence or

.

representa

judge

together

ac

construction by and

-

faculty

grammatical

can

as the

systems

considerations the

they

words

the

influenced

why

it ,

language

represent

influence

the

,

not

,

operates

cognitive the

consequence is

when

put

. Instead

a

explains

is

other

within

used

not

,

to

As

are

This ,

example

attempt

chase

.

First

speaking

faculty

by

operations

.

. Broadly

language

shared

sentences

that

modular

the

not ,

mechanisms

.

dogs

,

been

, we

in

models earlier

Wexler

the

involved

the

assumptions

is

that

and

modularity

cognitive

,

they

that

sense to

of

the

. To

our of

have

fundamental system

the

specific

construct

false dogs

two

representations

sentences

people

chase

in

are

other

( see of

makes

conception

-

and

three in

and

that

make .

.

- processing

modular

semantic

nitive

that

this

Crain

to

Model

that

from

and

tions

is

principles

examine

grammar be

-

Before

language

may

implicit

. First

Model

to ,

process

statement

models

.

need

outlines

is

in

other

inves

sentences

of

explicit

.

methodological

we

that

psycholinguistics

language

apparatus

cording

first

acquisition / pragmatics

perspective

chapter

appeared

Matching

,

cognition

models

the

two

Model

human

the

,

developmental

Matching

the

of

however

and

been

examine

experimental

interrelations

This

the

along of

of

learn

the

.

endorse with

Matching

that

to

of

others

we

Modularity

Modularity

sumes

of

used of

one and

contrasted

field the

,

on

divided results

,

, we

semantics

psychological

processing

Although

focus

understanding

broader is

we

have

follow

of

are

components

model

be

a

,

and

their

conceptions

Hyams

the

will

generally we

,

part

studies

that

acquisition

chapters

that

language

.

Wexler

most the

of

, many

chapters

methodology

system

in

the

methods

from

nonlinguistic and

language

and

the

experimental

different

and

the of

and

cognitive

learning

For

, of

argument

. In

methodology

of

several

linguistic

language

.

Grammar

the

- stimulus

aspects

productions

Universal

role

There

of

' s

- the

research

consider

descriptions

children

theory

- of

constraints

such

points

separate of

poverty

also

certain

with

tigations

the

of

but

clarify

lines

the

linguistic

features

syntax

To

of

investigate

design

of

logic

to

speaker

be like

sentence Cats

chase to

have

30

Chapter 4

Second, the model assumesthat the child's language-processing system is essentially the sameas that of an adult. For example, if adults do not allow their beliefs and opinions to influence the interpretations they assign to sentences, then, on this

model, the same is expected of children. Similarly, both adults and children should invoke the sameparsing strategiesin resolving ambiguities. Moreover, in the absence of evidence to the contrary , both adults and children are assumed to have similar processing capacity and memory limitations .

Maximizing the similarities between children's cognitive systems and those of adults is neededin part to explain language Ieamability- why all children success fully converge on an adult linguistic system despite the considerablelatitude in their linguistic experience. The model we will consider in section 4.3, the Competing FactorsModel, allows children to differ from adults in languageprocessing(possibly, in order to attribute grammaticalknowledge to children). This is not the Null Hypoth esis, however. It is also unwise. Permitting the characteristicsof the child and adult processingsystems to differ opens a Pandora's box of possible processing explanations (cf. Atkinson 1996). Moreover, any account of children's performancethat attributes different properties to the child and adult processing systems must face a

new question: how does the processingsystem of the child changeso as to converge on the adult system? To the extent that the cognitive mechanisms of children and

adults are similar, problems of learnability do not arise. For this reasonalone it is important to take the assumptionof equivalenceseriously. These suppositionsjointly form the basisfor the Modularity Matching Model. By "modularity matching," we mean that the principles of any component of the language apparatus are the same for children and adults; for example, the principles within the syntactic component are the same, and the principles within the semantic component are the same. Chapters 12, 13, and 14 provide a more completepicture of the languageapparatus, according to the Modularity Matching Model. In a nutshell, the language apparatusis a hierarchically organized system in which the results of (partial) analysesat lower levels of linguistic representationare rapidly transferred upward to higher levels for analysis. For example, once (partial) syntactic representations are built, the results are subsequently, and incrementally, shuntedto the semantic component, where meaningrepresentationsare built; the responsibility of relaying information between levels of representationwithin the system is undertakenby the verbal working memory system. Only the output from the language faculty makes contact with real-world knowledge. Extralinguistic knowledge, which is not itself contained in a "module," does not influence the construction of linguistic representations at any level, however. The graphic in figure 4.1 illustrates this point . According to the Modularity Matching Model, all of the linguistic abilities of a child are the same as an adult 's. Not only do we assume that children have access to Universal

Grammar , just as adults do , we also make the more controversial

assump -

tion that children are equivalent to adults in the mechanismsthey use to processlanguage; that is, they have accessto a universal parser. The assumption that the processingcapabilitiesof children and adults are equivalent is not generally accepted even within the generative approach to language acquisition. We take this assumption to be the Null Hypothesis, however.! The Modularity Matching Model borrows heavily from the theory of Universal Grammar. For one thing, the model follows the theory in maintaining that linguistic experienceplays only a limited role in grammar formation. According to the Modu-

Models of LanguageDevelopment

31

IReal -WKnowledge orldI Semantic / Pragmatic Principles Syntactic Principles The Language Faculty Figure4.1 TheModularityMatchingModel

larity Matching Model, innately specifiedprinciples of grammar circumscribethe linguistic hypothesesthat children formulate. The role of experienceis minimal, at least in the acquisition of syntactic and semanticprinciples; experiencesimply enablesthe learner to identify and set the parametersprovided by Universal Grammar. That is, experienceaids the learnerin deciding among grammaticaloptions madeavailableby Universal Grammar. This said, we want to emphasizethat it is consistent with both the Modularity Matching Model and the theory of Universal Grammar for children to provisionally select grammatical hypotheses that will later be abandonedon the basis of (positive) evidence from the linguistic community. There is a limit to the possiblemismatchesbetween child and adult grammars, however. The limit is stated in the Continuity Hypothesis, according to which children's developing grammars can differ from the target, the adult grammar, only in ways in which adult grammars can differ from eachother (seeCrain 1991; Pinker 1984). 4.2 TheInput Matching Model The observation that children sometimeshypothesize grammarsthat differ from the target is important in distinguishing among alternative models of linguistic knowledge and linguistic behavior. For one thing, evidenceof children's nonadult behavior indicatesthat children are not forming grammaticalhypothesesbasedentirely on the input. Therefore, children's honadult productions and their nonadult interpretations of sentencesprovide evidenceagainst models that maintain that children's grammatical hypothesesare securelytied to their primary linguistic data. We call this classof models the Input Matching Model. The Input Matching Model places little , if any1 emphasison innately specifiedlinguistic knowledge as a source of children's grammatical hypotheses. Instead, general-purpose learning algorithms are assumedto underlie languagelearning, as well as other cognitive processes .

32

Chapter 4

One example of the Input Matching Model is the CompetitionModel proposed by MacWhinney and Bates (1989) and their colleagues. According to the Competition Model, a learner relies on "cues" from the input to form connectionist networks; these networks are the leamer's "grammars." Examplesof cues include word order, morphological agreement between linguistic items, and semantic plausibility. The learner comesto place more or less weight on different cuesaccording to their availability in the linguistic environment. According to the Input Matching Model, these differencesaccount for crosslinguistic variation and for variation among speakersof the samelanguage. Another variant of the Input Matching Model is the Coalition Model of HirshPasekand Golinkoff (1996). As its name suggests, the Coalition Model views grammar formation as the leamer's response to a coalition of language-related input sources. Unlike the Competition Model, the Coalition Model admits certain innate knowledge that selectively guides the child to place various "weights" on different sourcesof input. On the Competition Model, the weights that the child assignsto different cuesare strictly determined by the input. By contrast, the Coalition Model of Hirsh-Pasekand Golinkoff maintains that "[t]hesecuesare availableto the child at all times; however, they are not equally accessible to the child at different points in development" (p. 189). On the Coalition Model, there are three different phasesof language acquisition. Thesephasescan be characterizedaccording to children's "biases" to attend to different external cues: prosody providescthepredominant cuesfor the child in phaseI, semantic cues dominate in phase II, and syntactic cues dominate in phase III . At each phase, the primary sourceof linguistic knowledge is the environment: However, we have suggested that the mechanismfor progress within each phase is guided distributional analysis. That is, first the child is primed to selectively attend to certain information within eachdomain that forms the coalition (e.g., prosody, syntax). Second, the child mines these selectedinputs across domains to construct an interpretation of the linguistic stream and to build mental models. (p. 189) . .. Children must learn to exploit the correlations that exist across(as well as within ) domains. . . . (p. 190) Knowledge of constraints develops in phase III, according to Hirsh-Pasek and Golinkoff. As far as we can tell, however, this knowledge too is basedon experience. As the following quotation illustrates, the Coalition Model views the learning of complex syntax as a IIdiscovery" process, in which the child is predisposedto notice co-occurring patterns in the input: Finally, in phaseIII, children learn to comprehendand produce languageto represent more complex events that they may not have even witnessed, as well as multipropositional cross-clause constructions. Here, . . ., children are biased to rely on syntacticcues(although other aspectsof the coalition still influencetheir interpretation of what they hear). With increasedcapacity to hold more than one event in mind, they are motivated to learn about specificlinguistic properties in their language. . . . (p. 187) The Coalition Model differs from the Competition Model in that the child's disposition to attend to particular environmental cues changesover time; this is the concessionto innatenessmade by the Coalition Model. It seemsclear, however, that

34

Chapter 4

in processing two -clause constructions than in processing one-clause constructions

(e.g., McDaniel and Maxfield 1992a, 668- 669). To take another example, it has been claimed that processing difficulty increaseswith the number of (animate) NPs that appear in a sentence (Goodluck

1991 , 175 - 176 ). It has been maintained

that children

also experiencegreater difficulty understanding sentenceswith three NPs than ones with two NPs (e.g ., Goodluck

and Tavakolian

1982 ).

Another factor that hasbeencited as contributing to children's processingdifficulty is IIcarryover" effects- for example, the similarity of materials acrossexperimental conditions. This, too, hasbeenheld responsiblefor children's errors in comprehension (Philip 1995, 109). As a final example, it has been suggestedthat young children attempt to interpret constructionsin an "order-of-mention" fashion. In comprehending constructionswith two clauses, for example, children appearto find it easierto interpret sentences in which the event mentioned in the first clause occurs first in the con-

text. This accountsfor the higher proportion of correct responsesto sentenceslike (2) than to sentences like (1) (see Amidon and Carey 1972; Clark 1971; but cf. Crain 1982 ; Gorrell , Crain , and Fodor 1989 ). (1)

Before you touch your nose , touch your ears .

(2)

After

you touch your nose , touch your ears .

According to the Competing Factors Model , factors such as these all compete with

children's linguistic knowledge in determining their linguistic behavior in an experimental task. Proponents of the Modularity Matching Model readily admit that certain factors may sometimesimpede the linguistic performanceof both children and adults, albeit artifactually. It is easy to make a subjectperform poorly (if the instructions to the task are nQt clear, if a subject is askedto perform two tasks at the sametime, etc.). Moreover, there are specific linguistic properties that make some sentencesmore difficult to cope with than others. It is well known, for example,.that adult language users cannot correctly understand deeply center-embeddedsentences , in many instances. However, on the Modularity Matching Model children should only be subjectto the same "processing difficulties " that curtail adult performance ; if adults can orocess L

sentenceswith three animateNPs or sentenceswith two clauses , or if they are able to respond differently acrosstrials despite being presentedwith similar materials, then children should be able to also, according to the Modularity Matching Model. It seemsevident that adults have little , if any, difficulty interpreting two -clause sen-

tences or ones with three animate NPs. If such sentences fall within the processing capacities of adults, then, by assumption , they should also fall within the processing

capacitiesof children. The observation that children perform less well than adults on such sentencesis inconsistent with the Modularity Matching Model, and the differencesbetween children and adults must be explained. Proponents of the Competing Factors Model cite another kind of factor as interfering with the demonstration of grammatical knowledge. This factor concerns the match between

a sentence and the conversational

context

in which

it is used . It is

maintained that children (and, to a lesser extent , adults) have a bias to interpret any

sentenceso as to make it true in the discoursecontext. For ambiguoussentences , this bias toward interpreting sentences so as to fit the discourse context is eminently rea-

sonable and is consistent with ctheModularity Matching Model; however, on the

Models of LanguageDevelopment

35

Competing FactorsModel, the bias to make sentencestrue can also weigh againstthe interpretation given by grammaticalprinciples for unambiguoussentences . If circumstancesare such that a sentenceis not true on the interpretation that is consistent with the grammaticalprinciple, a conflict arisesbetween grammaticality and the bias to assign a meaning-that makesthe sentencetrue in the conversationalcontext. According to proponents of the Competing FactorsModel, this conflict can inflate the proportion of errors by children; errors occur on trials where the bias to make a sentence true outweighs children's grammaticalknowledge. Similarly, it has been maintained that children (and adults) commit errors in judging whether an answer is or is not appropriate to a question, becausethey "unconsciouslychangethe word order of the question to match the answer" (McDaniel and McKee 1993, 288). On the Competing Factors Model, all of these factors are assumedto sometimes suppressthe proportion of children's correct responsesin a task. This follows only if linguistic behavior is viewed as a probabilistic weighted sum. In the sum total of different probabilistic factors, the influence of factors extraneousto grammar is sometimes enough to make an ungrammatical structure acceptablefor children (also see Lust. Eisele, and Mazuka 1992, 338; McDaniel and Maxfield 1992a). This would not happen on the Modularity Matching Model . According to this model, if children know a linguistic principle, then they should behave in conformity with it, just as adults do. In short, the fundamentaldifferencebetween the Competing FactorsModel and the Modularity Matching Model is that grammar has priority over other factors on the Modularity Matching Model; there is no notion of grammaticality in any abso1utesenseon the ComDetin ... -~ FactorsModel. Moreover, on the Modularity Matching Model, every child (and adult) is expected to perform perfectly, all things being equal. By contrast, on the Competing FactorsModel, a subject's performancedepends on the precisemix of contributing variables that influence linguistic behavior at different moments in time. Therefore, a subject's performancemay vary from trial to trial; different subjectsmay perform more or lesswell; younger children may perform lesswell than older children; and so on. 4.4 Conclusion The upshot of this chapteris that different models of languageacquisition lead one to expect different patterns of responsesby children in experimental tasks that are designed to assesstheir knowledge of linguistic principles. According to the Modularity Matching Model, children should abide by universal principles at all stagesof development. Both children and adults should therefore perform without error on innately specifiedlinguistic principles. If the Continuity Hypothesis is adopted, the differencesbetween children and adults are confined to differencesthat appear across adult languages. Other divergencein the linguistic behavior of children and adults is not expected. This contrastswith the predictions of the Input - Matching Model and the Competing FactorsModel . On the Input Matching Model, children's hypothesesare circumscribed by their experienceto a much greater degree. Therefore, there should be a much closer match in the linguistic behavior of children and adults on the Input Matching Model. On the other hand, the Competing FactorsModel, like the Modularity Matching Model, may assume that children have accessto a grammatical "module." The grammatical module does not have special status (i.e., preempting

36

Chapter 4

other determinantsof linguistic behavior} on the CompetingFactorsModel. This modelthereforetoleratesa muchgreatermismatchin the linguisticbehaviorof children and adultsthan doesthe Modularity MatchingModel (or the Input Matching Model). It shouldbe understoodthat the divisionswe havemadeamongmodelsof languagedevelopmentare somewhatartificial. For example , the CoalitionModel is an Input Matching Model, but it can also be considereda variant of the Competing FactorsModel. (Accordingto the CoalitionModel, the subsystems of cognitionare highly interactive , ratherthanhierarchicallyorganized .) It shouldalsobe understood that the divisionswe havemadetend to exaggeratedifferencesamongresearchers who work within the samelinguistic framework . As we have indicated , many researchers who adopt the generativeenterprise , includingits assumptions aboutinnateness andevenits assumptions aboutthe modularityof the languagefaculty, are countedas advocatingthe CompetingFactorsModel. The reasonis that these researchers adopt the methodologicalassumptionsof that model, rather than the methodological assumptions of theModularityMatchingModel. AssumingtheCompetingFactorsModel, manyresearchers employtasksandresearchstrategiesthat we would deemto be inappropriatefor the studyof childlanguagewithin the generative framework . We view our criticismsof the work of theseadvocates of the Competing FactorsModel asa disputewithin the family. For this reason , we ignorethe vastbulk of child languageresearch , choosinginsteadto examinethe studiesof researchers who shareour fundamentalassumptions about the role of generativegrammarin languagedevelopmentbut who neverthelessadopt the researchstrategiesof the CompetingFactorsModel. We will defenda differentset of methodologicalassumptions , which we believe to be moreconsistentwith the theory of UniversalGrammar . We hopeto convince thoseresearchers who adopt the basicassumptions of generativegrammarthat it is alsonecessary for themto ~dopt a corresponding setof methodological assumptions in assessingchildren's linguistic knowledgeand -to abandonthe methodological assumptions of the CompetingFactorsModel. The argumentrelieson an explicit statementof the interrelationsof grammaticalknowledgeand performancefactors (e.g., short-term memoryandextralinguisticknowledge ). On the CompetingFactors Model, both grammaticalknowledgeand theseand other performance factorscon.. tribute to linguistic behavior. On the Modularity Matching Model, grammatical knowledgedoesnot competewith theseperformance factorson most occasions . Before we examinethe differences betweenthesetwo modelsmoreclosely, we takea brief excursionto point out someempiricalproblemswith the Input MatchingModel and to indicatewherethe linguisticbehaviorof childrenand adultsmay differ, accordingto the ModularityMatchingModel.

Chapter 5 Continuity versus Input Matching

In conjunction with crosslinguistic research, the study of children's linguistic knowl-

edgeis a specialsourceof evidencefor decidingwhetheror not a principleof the theorv - of Universal Grammaris a viable candidateas a linguistic universal (i.e., likely

to be innatelyspecified ). As mentionedearlier, onehallmarkof innateness is mastery in the absence of experience. Because young children 's experience is more limited than that of older childr ,en and adults , they can offer relevant data bearing on the "nature/nurture " controversy . Clearly , too , different children encounter different

input. This observation leads us to expect different children to adopt different hypotheses, assuming that they are basing their hypotheses on experience. On the other hand, if all children in a linguistic community adopt the same linguistic hypotheses, then we can infer that languagedevelopmentis guided by innate principles, and not by learning-theoretic mechanisms . The most compelling evidenceof innatenessprobably comesfrom the observation that children sometimes make nonadult grammatical hypotheses . The Modularity

Matching Model readily acceptsthat children's grammarsmay differ from those of adults . There are heavy constraints on the different forms that children 's grammars

can take, however. The model adopts the Continuity Hypothesis- the claim that children's developing grammarscan differ from the adult grammar of the linguistic community only in ways in which adult grammars can differ from each other (e.g .,

Crain 1991; Pinker 1984). According to the Modularity Matching Model, there is one further source of differencesin performancebetween children and adults. Children's non adult behavior

may be derived

from the one module

that children

and adults do

not share, the LAD .l In chapter 14, we consider how certain nonadult interpretations by children can arise from a principle of the LAD - a principle that governs languagedevelopment. We postpone that discussionin order to develop the point in detail .

The present chapter offers additional evidenceagainst the Input Matching Model. The evidence consists of children's nonadult linguistic behavior. First we discuss children's nonadult utterances that were discovered using the elicited production task. Then we present evidence of children 's nonadult assignments of meanings to

sentencesthat were discoveredusing the truth value judgment task. 5.1 The Medial -Wh

One source of evidence against the Input Matching Model is based on children's nonadult (but Universal Grammar- compatible) questions. This is the so-calledmedialwh phenomenon. Using the elicited production task, it was found that someEnglishspeakingchildren around the age of 3 or 4 consistently insert an "extra" wh-word in

38

Chapter 5

their long-distance questions, as illustrated in (1) (Thornton 1990; Thornton and Crain 1994). (1) What do you think what pigs eat? The appearanceof the medial-wh in the languageof children learning English cannot be explained as children's responseto the input. Although these constructions are not grammatically well formed in English, structureslike (1) are attestedin dialectsof German and other languages. Interestingly, even after the medial wh-phrase is exLpunged from children's object extraction questions, it persists in their subject extraction questions. This developmental sequenceis consistent with the theory of Universal Grammar. As suggestedby Thornton (1990), for example, the extra whphrase in certain kinds of children's questions may be. an overt manifestation of a processthat appearsin French, for example, in the familiar que/qui alternation (Rizzi 1990). English-speaking children's nonadult productions therefore support the Continuity Hypothesis, since both the structures themselves and the course of their development fall within the boundaries established by the theory of Universal Grammar. The medial-wh phenomenon constitutes a serious challenge to the Input Matching Model, however. To conclude this discussionof the medial-wh phenomenon, we would add three commentsabout the nature of children's nonadult productions. First, not all children produce medial wh-phrases. Only about one-third of the children we have interviewed produce questions of this form. This shows that there are several possible paths leading from the initial state to the final state of grammar formation. It also eliminatescertain explanationsof the findings (e.g., accountsbasedon the maturation of linguistic knowledge). Second, the period during which these children produce medial wh-questions persists for several months, perhaps for years. Finally, the explanations that have been offered for grammaticalchange- that is, for the transition from medial wh-questions to adultlike questions- are based on the assumptionthat negative evidenceis not availableto children (seechapter4 and chapter22). 5.2 Repetitionof theAuxiliary The second example of children's nonadult productions is children's negative questions. It was discoveredby Thornton (1993) that children even as old as 5 often form nonadult negative wh-object and wh-adjunct extraction questions, and nonadult negative yes/no questions. Thus, they ask wh-questionslike (2b) instead of the adult (2a), and they ask yes/no questions like (3b) instead of the adult (3a) (also see Guasti, Thornton, and Wexler 1995). (2) a. What doesn't Big Bird like? b. What does Big Bird doesn't like? (3) a. Doesn't Grover like the ice cream? b. Does Grover doesn't like the ice cream? (4) a. What can't Elmo jump over? b. What can Elmo can't jump over? c. What does Elmo can't jump over? The (b) and (c) forms in (2)- (4) do not occur in adult English. Of course, there is no a priori reasonto expect that children's grammaticalhypotheseswill be closely tied to

Continuity versusInput Matching

their

linguistic

tures

experience

consistent

nonadult

are

,

Grammar

-

though

has

not

the

1995

stance

data

.

. 3

As

On

,

the

ones

without

dren

' s

-

study

7

wh

-

The

children

,

with

,

-

bare

wh

construction

,

children

'

s

children

'

s

-

phrases

appeared

wh

full

The

Spaceman

(

6

)

What

food

that

the

structural

5 .4

In

Split

adult

Even

seem be

a

" pied

the

full

wh , as

input know

- piped

.

,

but

-

phrases

bare

"

-

phrase

.

(

the

same

)

"

-

ones

( i

. e

. ,

chil

,

-

in

a

that

4

phrases

and

.

The

wh

-

auxiliary

-

also

structure

questions

doubling

"

appeared

appeared

in

as

complementizer

in

. ,

found

wh

structure

appeared

. e

example

contexts

an

that

i

phrases

of

1995

.

evoked

overt

-

For

full

negation

doubling

(

wh

properties

with

another

an

wh

"

Thornton

in

-

,

"

well

.

In

sometimes

children

'

s

questions

with

' t

like

?

were

' t

like

?

observed

'

s

arose

grammars

responsible

,

.

with

this

full

the

,

bare

wh

assigned

nature

of

same

case

and

representations

the

in

In

by

these

different

discourse

con

Thornton

-

-

suggested

phrases

moving

these

to

children

types

of

moves

as

.

wh

-

The

phrases

.

elephant

to

-

linguistic

Questions

position

Whose

though

English must

,

full

wh

contained

doesn

structural

23

-N

- initial

(7)

;

phrases

surrounding

English

sentence

phrases

-

the

position

auxiliary

children

is

the

chapter

Whose

-

and

in

primary

structural

evoked

all

The

that

controversy

in

,

another

their

and

The

task

Spaceman

by

in

discussed

.

)

structures

doesn

component

positions

theoretical

)

.

of

object

-

differences

imposed

syntactic

)

never

the

the

-

. 2

does

different

Al

con

Thornton

Model

"

nature

complementizer

however

be

.

other

doubling

,

Matching

food

what

were

wh

What

must

Input

sentence

wh

full

)

they

that

.

( 5

5

the

,

)

the

to

)

any

illustrate

tied

whose

children

in

with

6

not

a

Guasti

questions

are

1995

in

to

see

.

Universal

Wexler

related

(

nonadult

or

the

from

with

,

who

questions

extracted

(

and

' s

Grammar

are

questions

be

production

from

illustrated

in

s

or

different

(

Since

text

phrases

questions

as

as

elicited

the

questions

,

bare

such

an

although

as

,

negative

distinguish

boys

on

elicited

-

'

linguists

which

,

using

questions

Thornton

-

children

Universal

example

for

,

as

vroduced ...

phrases

wh

questions

depending

sometimes

bare

-

content

differ

of

children

children

of

questions

could

for

struc

some

following

negative

these

process

challenge

such

lexical

,

to

hypotheses

wh

,

s

grammatical

that

Phrases

on

content

sentences

,

another

Wh

Guasti

,

nonadult

argued

are

'

Italian

hand

pose

research

lexical

other

;

the

of

grammatical

of

recent

with

,

be

they

parallel

)

of

can

children

exact

dialect

they

Nature

an

it

that

that

English

their

such

The

In

.

which

existence

,

1993

of

Paduan

)

in

the

evidence

Thornton

child

in

is

argued

aware

( besides

struction

Wexler

strong

(

are

only

Grammar

been

compatible

we

language

is

it

Not

Universal

utterances

Accordingly

5

.

with

39

along

in

(7 ).

do

you

provides that

"

- phrase in

the with

a

think

the

- question

jumped

only noun

wh

the

questions inside wh

always

best

like a

- operator

whose who

whole

to

?

( 7 ), -N

a

not

phrase and

the

all

children

( e . g ., whose possessive

learning elephant marker

) 's .

40

Chapter 5

In

(

an

elicited

1996

)

long

-

production

found

3

distance

given

(

8

8

)

)

-

N

questions

do

you

a

.

Who

b

.

Who

:

kinds

of

,

wh

questions

,

not

-

)

a

,

wh

-

The

3

)

-

a

-

-

adult

it

lenge

,

. 5

insert

they

insert

to

use

the

to

One

are

-

wh

this

)

-

questions

their

kind

.

long

of

for

a

walk

?

for

a

walk

?

)

also

remain

illustrates

-

For

distance

sentence

form

is

intact

in

the

matrix

.

study

)

the

one

well

who

speaking

-

of

a

(

.

It

Sza

-

these

NP

.

N

be

,

The

al

-

non

-

chal

-

compatible

of

Model

see

possessive

therefore

matter

to

of

questions

are

is

Model

a

-

'

grammar

distance

children

phenomenon

' who

the

out

long

operator

case

in

move

to

-

whose

documented

that

can

that

possessive

wh

-

suggest

restricted

-

hypothesize

with

allow

is

8

apparently

questions

Matching

be

in

parametric

poses

varia

a

further

-

.

-

initial

their

into

their

,

the

wh

wh

-

an

must

be

,

the

tendency

Matching

like

questions

(

8

in

both

( 2

)

(

-

{ 4

I

)

is

)

,

.

chil

and

It

hard

is

note

,

only

to

,

the

-

rather

adults

pressed

who

' s

wh

-

why

and

into

moving

why

material

children

porridge

) ,

in

of

Model

whose

like

as

as

insertion

by

account

is

,

questions

phrases

alternative

explained

questions

involve

Input

-

provide

What

distance

negative

the

full

distance

-

forms

Moreover

to

.

long

discusses

position

,

and

-

operator

ex

-

por

-

.

the

Input

overlap

. 3

like

pressed

chapter

nonadult

21

.

hard

this

into

split

salvage

to

question

s

in

took

distance

Modularity

these

)

long

templates

'

intact

took

(

English

would

children

to

children

?

boy

that

Matching

chapter

forms

-

these

' s

-

this

auxiliary

forming

)

tried

boy

operator

Because

elements

of

sentence

?

Matching

extra

(

best

?

in

is

Input

several

some

"

forming

productions

?

the

Hungarian

-

reported

wh

deletion

.

wh

Model

an

way

s

?

the

that

languages

by

Input

extra

in

who

the

the

data

reduced

ridge

two

to

Matching

why

are

produced

for

its

plain

doubling

,

that

than

'

in

?

dog

long

extraction

.

best

remain

whose

forms

with

however

worthy

children

Pocahontas

in

Gavruseva

in

question

Grammar

production

dren

Gavruseva

noun

the

10

like

' s

and

,

think

operator

(

fronted

of

consistent

-

you

NP

-

dog

like

possessive

forms

Input

the

the

jumped

indicates

think

do

you

children

Implications

The

of

and

pipe

tried

apart

Example

there

split

think

you

#

wh

.

is

possibility

is

these

Pocahontas

which

you

the

that

a

speaking

question

,

of

Thornton

-

the

you

operator

think

Thornton

The

Universal

tion

5

.

porridge

symbol

dog

out

s

broken

do

do

alone

of

)

the

pied

. )

with

Who

'

-

you

Whose

out

though

,

to

jumped

do

( The

children

#

turns

1994

English

do

from

It

extracted

bolcsi

dog

Which

.

elephant

do

wh

know

Which

who

.

.

we

these

s

not

the

phrases

.

word

phrases

are

as

'

porridge

illustrates

#

wh

Whose

phrases

far

children

wh

with

)

.

b

the

9

of

10

subjects

examples

elephant

think

with

as

questions

(

-

.

b

Moreover

wh

(

,

9

Some

think

you

:

phrases

as

attested

(

child

failed

.

Whose

do

( Adult

Other

12

consistently

.

( Adult

example

with

who

whose

in

(

study

children

(

11

Matching

For

)

Model

example

(

=

=

(

5

,

)

)

could

would

consider

be

be

how

formed

a

.

Its

to

template

formation

allow

the

for

contents

an

would

"

of

auxiliary

combine

-

41

Continuity versusInput Matching the

template

one

of

process

left

in

children

duced

, is

children

What

( 12 )

a.

does

What

Unless their

does

some

other

" auxiliary

distance

questions

speaking that

they

adult

arise

5 .6

So

far

in

ductions tinue

this

chapter are

case

for

sentences child it

turns

t } 1an

out

easily

requires only

limited

To

the

decisions

about

tact

certain

with

This

makes

garding matic The There

present are

which

has

either

with

in

adults

( 15 ) . Notice

scope no

that

to

converged

; but not

an also pro

- questions

medial wh

-

.

wh

- questions

- phrasesin

on

-

English

. We

the

,

long by

Model

rely

-

suggest

final

state

, the

a

the

( 15 ) has

clause or

the

with

- world

same

an

in adverb

interpretation

, they

of

so - called a

second

have

do

of

.

making lost

to

con

-

children

.

adults

opposed

re -

to

or

donkey universal

type

quantification with

in

prag

-

.4

contains

( 14 ) ; the

have

adults

beliefs

those ( as

,

may

knowledge

available

principles

of

. First

than

than

,

clearly

children

may

remain

sentences

type

show

sentences

- world

and

revealing

interpretation

how

will

more

. Because

knowledge

of

shows

we

resources

that

it

)

. Real

gather

sentences

are certain

considerations

( ambiguous

pragmatic

. One

; . As

's

semantics

even

beliefs to

semantic

, as

language

following

and

more and

of

-

con

children

understand

principles

of

distinctions

sentences

relative

quantifier

child

the

time

representations

from

adults

pro

5 . 7 , we

evidence

distinctions

from

real

' s nonadult

section

' s acquisition

of

ambiguous

syntactic

donkey

using

linguistic

limited on

children . In

and

study

potentially

concerns of

,

children

take

more

of

the

of Model

knowledge

semantic

of

over

1987

wh

-

with

( 13 ) is

productions

yet

linguistic

these

have

' s judgments

overt

's

adult

interpretations

, and

and

deriving

the

arises

interpretation

types

non

children

basic draw

- world

come

example

two

in

affairs

real

syntactic

) in

from

alternative

contributions

principles

in

ungram

but

Fodor

split

instances

that

about

, they

the

question

the

- phrase

- distance

their

Model

fact

virtue

of

the by

children

the

a

experiences

that

of

. The

?

Matching

findings

sometimes

experience

extent

and

Matching

three Input

. The

state

manipulated

wh

- question

Input

not

Matching

. The

children

- world

the

informative

about

.real

presented

Input

reason

. This

"

children

and of

have

kind yield

for

beans

Sentences

on

reveals

judgments

be

have

the

be

that do

children

sentences

can

adults

adult

, we

differently

beans

examples the

template like

beans

explain ,

for

: Donkey

another

language

- N , these

unexpected

of

important

to

the

would

long

' t like

questions

't

?

' t like

found

the

doesn

.

against

comprehension

be

because

Semantics

that our

can

doesn

problematic

system

in

doesn

whose

simply

Continuity

like

wh

in

' t like

Spaceman

it

Crain

of

phrases

doesn

negative

appear

grammatical

-N

? and

yield

,

( see kind

Spaceman

"

with

children

whose

Spaceman

not

a " moved

know

, the

the

way

- doubling

we

like

Spaceman

does

, with

the

the

The

( 11 ) ) . Rather

noted

Spaceman

b .

( 13 )

as

Spaceman

,

question

far

split

the

process

( i . e ., of

as

the

what

, this

) . As

who

does

fact

kind

1984

does example

produce

attested

Peters

those

(I I )

( 12 ) . In

( 13 ) . This

not

and

by

What , for

actually in

behind

Wilson

like

counterparts

question

see

- question

Illustrated

that

matical

a wh

declarative

is

forms

N

for

its

is , such

without

the

a

sentences

.

quantifier

,

conditional as

always

adverb

, , as

.

42

Chapter 5 (14 )

Relative clause

Every farmer who owns a donkey feedsit . (15)

Conditional

If a farmer owns a donkey, he (always) feedsit . Both types of donkeysentencesadmit of two interpretations, called the strongand the weakreading. On the strong reading, (14) and (15) are true only if every farmer feeds every donkey that he owns. On the weak interpretation, each farmer must feed at least one of the donkeys he owns; he may feed them all, but this is not necessaryfor the sentence

to be true .

Beginning with Heim 1982 and Kamp 1981, and continuing through current versions of Discourse RepresentationTheory (e.g., Kamp and Reyle 1993), linguistic analysesof donkeysentenceshave had two main goals. One goal has been to provide a semanticsthat assignsthe sametruth conditions to both relative-clausedonkeysentenceslike (14) and conditional donkeysentenceslike (15). A secondgoal has been to ensurethat the truth conditions for sentencesof both kinds correspondto the strong reading, according to which every farmer feeds every donkey that he owns (also see Groenendijk and Stokhof 1991).

Chierchia (1995) has questioned both of these goals. On Chierchia's account, relative -clause donkey sentences are interpreted using the same mechanisms of dy-

namicbindingthat underlie discourseanaphora(e.g., the relation between an indefinite NP in one sentenceand a pronoun in a later sentence). According to Chierchia's theory, thesemechanismsestablishthe weak interpretation for relative-clausedonkey sentenceslike (14), which is true if all of the farmersfeed at least one of the donkeys they own. By contrast, the interpretation of conditional donkeysentenceslike (15) is not primarily determined by the mechanismsof dynamic binding but, to a larger degree, is influenced by pragmatic factors, the strong interpretation being more readily ava.ilable in many cases. If there are distinctions to be drawn between relative-clauseand conditional donkey sentences , as Chierchia suggests, then children may draw them more clearly than do adults; for the reasonscited above. On the Input Matching Model, by contrast, children would not be expectedto make fine distinctions about sentenceslike these, which are surely not abundantin the primary linguistic data. Children's understanding of both relative-clauseand conditional donkev - sentences was investigated in a series of experiments designed in collaboration with Laura Conway (Conway 1997; Conway and Crain 1995a,b; Crain , Conway and Thornton

1994). In the main experiment, children were presentedwith either a relative clause or a conditional donkeysentencein a context that was consistentwith the weak reading, but not the strong reading. The findings support Chierchia's theory of dynamic binding. First, children did not interpret relative-clausedonkeysentenceslike (14) and conditional donkeysentences like (15) in the sameway; as a group, children rejected the weak reading of conditional donkeysentencessignificantly more often than they rejected the weak reading of relative -clause donkey sentences. As individuals , children tell an even more inter -

esting story. Some children consistently rejected the weak reading of conditional donkey sentences, but no children behaved in this fashion in response to relative -

clausedonkeysentences .

Continuity versusInput Matching

43

If we interpret these findings as evidence that young children initially assign the weak reading to relative-clausedonkeysentences , then this is evidencethat children's earliest hypothesis correspondsto the dispreferredoption for adults (if we acceptthe claim in the literature that adults prefer the strong reading for both types of donkey sentences ). It is evidence, therefore, against the Input Matching Model. 5.7 InalienablePossession Further evidenceof children's nonadult linguistic analyseswas uncovered in Conway and Crain's study of donkeysentences . The donkeysentencesin that study were divided along an interesting semanticdimension. On half of the trials, including both relativeclauseand conditional donkeysentences , the nature of the relation between the possessorand the object of possessionwas one of inalienablepossession.An example of inalienable possessionis the relation between parents and their offspring, as in (16) (or the relation between a car and its engine, a table and its legs, etc.). In contrast to this relation is the kind of possessionexpressedin sentenceslike (17). Becausethe relation between men and snowplows is not one of inalienable possession , let us call relations of this kind alienablepossession. (A syntactic analysis of the distinction between alienableand inalienablepossessionis given in Hornstein, Rosen, and Uriagereka 1995.) (16)

Every frog who has a baby takesit to the pond.

(17)

Every man who has a snowplow usesit to push snow.

Roughly half of the children (8/ 15) gave different responsesto relative-clause donkeysentencesdepending on the relation between the possessorand the object of possession . For relative-clausedonkeysentences , if the sentenceinvolved inalienable .. possession , as in (16), these children rejected the weak reading. By contrast, if the sentenceinvolved alienablepossession , as in (17), these children acceptedthe weak reading. No child exhibited this distinction when the target sentenceswere conditional donkeysentences , however. With conditionals, children tended to either reject all of the test sentences , or acceptthem all. Adults we have interviewed on the same stimulus materials appear to be totally unaware of the alienable/inalienable distinction in interpreting donkeysentencesof either kind. If so, then it is unlikely that this distinction is manifestedin the input to children. Therefore, the Input Matching Model would be hard pressed to account for this aspect of children's linguistic behavior. This experiment reveals an interesting differencebetween children and adults. Be. L - world knowledge in making decisionsabout causeadults come to rely heavily on real the interpretation of ambiguoussentences , they apparently lose certain syntactic and semanticdistinctions that remain availableto children. Having accrueda sizablebody of knowledge about the world, adults appearto assign sentencesinterpretations that are consistent with their experience. Becausechildren have less real-world knowledge, their judgments appear to be more revealing with respect to the underlying contributions of syntactic and semanticprinciples in the derivation of linguistic representations. Being less contaminatedby experience, children may occasionallymanifest grammaticaldistinctions that are maskedin adult performanceby their wealth of extralinguistic knowledge.

44

Chapter 5

5.8 Conclusion

This chapter illustrated linguistic phenomenathat are difficult to reconcile with the Input Matching Model . This ends our caseagainst the model. We now take up a different alternative to the Modularity Matching Model, namely, the Competing Factors Model. Although we disagreewith severalfundamentalmethodological assumptions of the Competing Factors Model , it remains the currently received view . Our argu-

ments against it consumethe greater part of the next sevenchapters.

Chapter 6 The Competing Factors Model

According

to

perform

the

flawlessly

course

,

not

all

experimentally

.

subjects

may

may

lack

veal

what

tic

a

the

in

,

of

there

this

5 , the

we

of

particular

,

an

the

statistical

.

the

%

in ,

,

that we

the

( as

-

,

a

meth

suggest

researcher

1984

such

Occasionally

follow ,

the

of demands

performance

however

then

Fodor

task

task

) .

studies

thumb

research

influence

correct

-

extraneous

extraneous %

re

linguis

proper

knowledge

even

and

the

the

With 100

accurate

for

range degree taken competence

of

will

we

has

-

that

is

if

obliged

to

) .

it

. will

that the

that

the

and ,

no

linguistic

be

that

behavior

with

is

is is

only

observed the

. In

forces

knowledge

-

single

input

other

linguistic

principles

Thus

when

as one

only

principles

competence it

reflects

experimental is

. Such are ,

grammatical

, even

which chance

.

,

conformity

~ only

critical

children task

model

over

Model

linguistic

variable

variance

evidence in

reflec be

above

to

linguistic

in

mind

this

precedence

which

behave

between of

.

as .

As

:

assume

behavioral

of

of

directly always

relation

access

take Factors

,

to put

the

perspective

privileged not

knowledge

never

. It

significantly as

)

of

the

do

subjects

( 1987

. Thus

From

factors

of

Flynn

well

view .

Competing

of

will

as

petence

On

manifestation

[ P ] erformance

cal

90

grammar

aggregate

, . and

factors

be

, of

( Crain

apparatus of

tendency

Chien

a

least

Model

language

behavior

be ,

and

in rule

,

to

of

:

, they

test

these

eliminate .

from

alternative

Factors

principles

to to

,

to

task

numerical

of

Of

task

the

assessment

Model

linguistic

a

psychological

a

influence

.

investigated

by

the

equal

being

any

devise in

the

how

creeps As

in

should

being

confused

Matching hold

deviation

error

an

,ng

the

the

contribute

Therefore

introduced

Competi

component

sumed

at of

to

adults

things

Factors

chapter ,

not

" be

acceptable

requisite

.

and

all

is

may

competence

noise advise

is

is

explain

little

experimental

,

difficult

)

that

the

source

it

linguistic

be

children

" noise

they

Modularity

have

we

of ,

noise

to

' s

both

knowledge

sources

makes ( i .e .,

the

should

particular

Competing

This

book

subjects

performance

the

are instructions

"

,

knowledge linguistic

tasks

children

of

's

havior

Lust

.

the

.

to

minimum

that

identify

a

According

prescription

children

on

linguistic

check

there

" error

design

amount

In

so of

assessing

odological

,

when

the

and

purpose

in

suming

6 . 1

example

,

to

to

small

For

Model linguistic

equal -

understand

.

demands

held

are

proportion

is

is

-

not

knowledge

It

Matching their

accessing

things

interest

strategy

factors

Modularity in

because

data

constrained

by

significantly

consists

this the

experimental

aspect nature

other

a factors

their

performance

-

pattern

variance of

of

com of

constrained

accessing of

, but

grammatical

to will

grammati

data

,

46

Chapter 6

behavioral evidencefor significant grammaticalfactors will always involve significant contrasts of degree, not absolute successwith one condition and absolute failure with another. (p. 282) It is worth drawing out the predictions of the Competing Factors Model in more detail. As the quotation from Lust, Chien, and Flynn makes clear, this model anticipates "variance" in performancein any experimentaltask, rather than all-or-none linguistic behavior on at least certain tasks, as anticipated by the Modularity Matching Model. The quotation also makes another assumption of the Competing Factors Model clear: that linguistic competenceis revealed by above-chanceperformancein experimentaltasks. There is more to say about the "nature of performance data" according to the Competing Factors Model. Most importantly, the distribution of scores should be unimodal.1 By assumption, every subject is influenced by the same constellation of competing factors. Therefore, differencesamong subjects should be only a matter of degree. From a statistical point of view, the pattern of responsesshould approximate a normal distribution. In a normal distribution, the numerical scores '-' ~iven to individuals (i.e., the observations or data points) form a symmetrical "curve." This "bell-shaped" curve centers around some average (or mean) level of performance, suchthat observationstaper off gradually in both directions away from the mean. Any individual observation in a range of observations is taken to represent some "true" component of behavior, plus an "error" component, due to extraneousfactors. In an experiment designed to investigate children's linguistic knowledge, the true component is linguistic performance that conforms to the principle under investigation; other factors contributing to behavior make up the error component. The magnitu~e of the error establishesa "gravitational field" for the true component. This field keepsthe mean from skyrocketing, by pushing the true component to one side or the other, with equal likelihood. The result is a normal, unimodal distribution with a single peak or mode, as in figure 6.1. In the figure, the horizontal axis gives the percentage of correct responseson the hypothetical measure taken of a linguistic principle. The vertical axis gives the frequency of observations (e.g., by individual subjects) at varying points along the horizontal axis. The mean in figure 6.1 is roughly between 50% and 75% correct; the majority of observationsfall within these boundaries, and the frequency of responsesdiminishesaccording to how far they deviate from the mean. According to the Competing FactorsModel, if the effect of the true component (grammatical knowledge) is significantly above chance, then it can be inferred that the group as a whole knows the grammatical principle under investigation. In addition to grammaticalknowledge, which accountsfor the "center of gravity " in the distribution of observedscores, each observation is influenced to some degree by various ('unknown" factors. These extraneous factors cause the dispersion of scoresin both directions away from the mean. Responsesshould vary only probabilistically (randomly) from the central tendency, however, so there should be a single "group" behavior.2 The Competing FactorsModel would not expect two patterns of behavior within a group of children or a group of adults; it would not expect a bimodal distribution, either by children or by adults. A bimodal distribution is illustrated in figure 6.2. . There is another expectation of the Competing Factors Model that is not shared by the Modularity Matching Model. On the Competing Factors Model, children's

Competing FactorsModel 8

SUO ~leAJesqo JO#

7

25

S)

75

100

75

100

% Correct

Figure6.1 A unimodaldistribution

8 7

6 5 4

3 2

suo~~eJ\Jesqo lO #

1 0

Figure6.2 A bimodaldistribution

25

50

% Correct

47

Competing FactorsModel Although

this

tribution

view

has

permits

one

drawn

.

to

For to

knowledge

be

the

under

assumption

that

responses

by

titatively

.

that

are

If

of

adults

showing

the

case

tributions

6 .2

Remarks

Our

Statistical

trivial

,

of true

,

and

be

icant

difference

( using

and

items

of

the were

parts

of quan

example

,

different

, a

children ,

the

the

with

bimodal

with

each

pattern

normal

the

-

Competing

for

showing

a

affirmative particular randomly

us

of

re

-

distribution

form

of

.

normal

dis

-

part

to

meaning2

of

to

the

) .4

were

are

responses as

from

one

by

than

that satisfied

reading the -

usual for

the that

the

}

would

be would

experiment

,

statistically

the

other

,

If

} of

there

then

evoked

were the

the

con

greater

underlying both

-

meaning

proportion ) .

, that

we

signif to

part

-

not

experiment

, meaningz

a

am

assessment

The

assumptions example

an is

meaning

to

second

reading

of

subjects

compared the

the

of was

-

.3

the

part

parts there

( on

assessments

.

in

distribu

readings

,

second

Hirsh

display

normal

making

experiment the

not

)

two

proceed

both or

prefer

( provided test

we

and

statistics

such

would

; in

data

children

responses

selected

of

experiment

responses

statistical

that

affirmative

the

parametric of

, of

differences

visual a

Making

we

conduded

( using

that

.

use

example

7 are

another

which adults

whether of

or

the

for

chapter

gathered

using

false

,

approximate

data

part be

reject

measurements

( see

are

one

suppose

one

totally essential

example that

how

we are nondiscrete

should

is

Having

proportion first

one

. ascertaining

would .

use

to

Here

. In

Statistics

data

or

let .

determine

warranted

the

the

in

,

affirmative

be of

use

the

is ,

children

task

false

that

tasks

measures

useful by

to the

more

proportion

this

constructions

the

,

that

take

under

only

of

however

that

in

analyze

meaningz }

responses

would

be

two

position in

significantly clusion

in

from

to

concreteness

, and

a

data

affirmative

,

implying

attending

such

judgment

meaning

in

view

approximate

is

of

and

of

.

task

spend from

also

. For

context

as

this

preferred

conducted

would

the

may

value

the

to

linguistic

children

children

would

paradigm

In

appropriate

however

be

true

It

qualitatively

to

acquisition

assessed

children

is

truth

).

) . Data

tests

a

would

be

time

sentence

using

construed

looking 1996

therefore

biguous

lead

distribution

setting

values

responses

.

expectations

be

the

unimodal

acquisition

and

point

language is

preferential

of

is

be on

monitors . It

the

could

continue

that

from

parameter

two

linguistic

.

not

Golinkoff

amount

tion

mode

research

. The

television

statistical

' s

to

responses had

should children

knowledge

and

the

a

,

of

' s can

grammar a

the of

adults

parameter

adults

,

then case

and

the

From

should

linguistic

Pasek

the

of

differing

say

are

Statistics

in

behavior

.

each

position statistics

when

that

parameters

on

-

In

children

and

around

.

distribution

assuming

of

shift

unimodal

children

-

variation

also

sampled

by

question

children

assumes

anticipates adults

-

.

principle

Model

dis

distribution

distribution

discussion

with

normal of

normal

normal distribution

observations

the

competence

universal

,

empirical

by a

a

the

normal

which

that

a

of

a

investigations

present

adults

parametric

responses

were

sponses

the

to

,

setting

and

would

distribution

is

in

linguistic

Factors

use of

from

assuming

. The

Competing

shift

Model

pattern

disclaimer

statistical

recognize

used how

' s

the

assumption

population to

statistics

children

children

we

the

indicate

investigation

both

the

important

the

we

brief

the

subject

Factors

of

a

is

of 7 ,

about

add

knowledge

In

basis

it

,

because

about

,

chapter

should

discredited ,

inferences

inferences

We

largely

unabated

purposes

. In

erroneous

been

make

our

continues

has

continued

49

subjects

-

so

Chapter 6

Statistical tests are also used to decide if two ambiguous sentencesare analyzed in the sameway. For example, as pointed out in chapter 5, there are two types of donkeysentences : relative-clausedonkeysentences , such as (1), and conditional donkey sentences , such as (2 ). (1)

Relative clause

Every farmer who owns a donkey feeds it . (2)

Conditional

If a farmer owns a donkey, he (always) feedsit . As noted, donkeysentencesof both types admit of two interpretations: the strong reading and the weak reading. On the strong reading, every farmer feeds every donkey that he owns; on the weak reading, every farmer feedsat least one of his donkeys. Chapter 5 discussedtwo theories of donkeyanaphora. According to one, both types of donkeysentencehave the sametruth conditions, correspondingto the strong reading (e.g., Heim 1982; Kamp and Reyle 1995). According to the other, the weak reading is basic for relative-clause donkeysentences , but the strong interpretation is basic for conditional donkeysentences(Chierchia 1992, 1995). To evaluatethesecompeting accounts, Conway and Crain (1995a,b) presented children with donkeysentencesof both kinds in contexts that were consistent only with the weak reading. Children were asked to indicate whether the test sentence matched the context . Applying

a

statistical test, Conway and Crain found that children acceptedthe weak reading significantly more often for relative-clausedonkeysentencesthan for conditional donkey sentences , as predicted by the secondtheoretical account. According to the Modularity Matching Model, these uses of statistics are appropriate because (a) the observations obtained from subjects are nondiscrete , as in the

preferential looking paradigm, or (b) the observations concern ambiguous sentences . Statistics are inappropriate, according to this model, when the observations concern unambiguoussentences . For example, children's knowledge of constraintsis expected to be nearly perfect, if the Modularity Matching Model is correct- it is not expected to be simply at some above-chancelevel of performance. That is, children are not expected to contract the verbal elementswant and to in (3), and they are expected to consistently reject coreferencein (4). (3 )

"'Who does Arnold

(4)

".~

wanna make breakfast ?

danced while the Ninja Turtle ate pizza.

At this point, it would be prematureto attempt to explain the reasonsbehind these predictions of the Modularity Matching Model. However, we felt that we should flag these points about the use of statistics in anticipation of that discussion . 6.3 Comparing the Models

Admittedly , the Modularity Matching Model may be incorrect. PerhapsLust, Eisele, and Mazuka (1992) are correct in assuming that subjects must " consult pragmatic

context" (p. 353) and other factors in all behavioral tasks and in assumingthat they do not basetheir responsesprimarily on grammaticalknowledge, as the Modularity Matching Model claims. It may also turn out that there are differencesin the linguistic behavior of children and adults that derive from sourceswithin the language-

Competing FactorsModel

51

processing system. For example, verbal working memory may undergo maturation. If so, the Modularity Matching Model would have to be adjusted to reflect the observed differences between children and adults . However , if children and/ or adults

perform poorly becausethey succumbto competing factors- that isJif their performance is not dictated by grammatical knowledge- then this finding would be evidenceagainst the Modularity Matching Model.5 It is even possible that the patterns of observations for children and. adults will prove to be qualitatively different. Insteadof approximating a normal distribution, the observationsmadeby children could turn out to be better characterizedby a different class of distributions . If so, the observed differences in the patterns of responses by children and adults could be attributed to processing mechanisms, in accord with the

Competing Factors Model. As we noted, however, processing explanations of observed differences

between

children

and adults should be advanced

with

caution . If

the processing system of a child has radically different properties from that of an adult, then the question of learnability arises: namely, how does the child attain the adult processing system?

The Modularity Matching Model skirts the problem of learnability entirely, by assuming equivalencebetween the mental architecture of children and that of adults. We take this to be the Null Hypothesis. By contrast, most researchersin language acquisition assumethe Competing FactorsModel as the Null Hypothesis. For example, consider the following statementby Lust, Eisele, and Mazuka (1992): Behavioral researchmust assumethat data derived from experimentaltasks are modulated to some degree (at least chance) by performancefactors. Just as in adults, these factors must be expectedto produce variation in children's group data. . . . This fact motivates the use of large samplesand statistical analysis in behavioral

research . (p . 340 )

The Modularity Matching Model accepts the first proposition- that all tasks are susceptibleto "noise." By chance, a child will make the occasionalerror, regardlessof how tightly controlled an experiment is. However, these chanceerrors do not arise becauseperformance factors are invariably at odds with grammatical knowledge; they arise because the child subject gets distracted , or fails to pay close attention to some event , and so on. As indicated , we expect only about 5% to 10% of such errors

by children in responseto unambiguoussentences . If errors occur more often than 10% of the time, they should not be chalkedup simply to "noise"; in our view, there is something to explain.6 The Modularity Matching Model rejects the second proposition- that performancefactors necessarilyintroduce variancein all experimentaltasks, so much so that the experimenter's only recourseis to use statistical tests of group data. According to the Modularity Matching Model, grammatical principles are used preemptively in language processing; they are not "modulated" by performancefactors in most instances. Only rarely will task demands or grammatical complexity make excessive demandson the resourcesof the language-processingsystem; in suchcases , of course, production or comprehensionwill be curtailed. Ordinarily , however, critical thresholds in processingare not exceeded, so linr;uistic behavior is expectedto approach 100% ac~ racy. Aside from the minimal int~ sion of "noise," ling~istic perfo~ ancewill be. 100% accurateif appropriate experimental precautions have been followed. On the Modularity Matching Model, a child who knows the grammatical principles under

52

Chapter 6

investigation will rarely make an error , so children as a group will respond with over 90% accuracy. There is one exception . If subjects are confronted with ambiguous sentences, then it is " accurate" for them to respond according to any of its inter pretations . In such cases, statistical tests may be useful in evaluating subjects' behavior . According to the Modularity Matching Model , however , statistical tests of group data are usually unnecessary for unambiguous sentences.

6.4 Areasof Convergence According to the Competing Factors Model, knowledge of a linguistic principle is observed as a tendency to behave in conformity with the principle. On this model, extraneousfactors competewith linguistic knowledge and often mask it . To conclude that children know a principle, therefore, it is sufficient to observe above-chanceperformancein experimentaltests of children's performance. Moreover, quantitative differencesbetween children and adults are anticipatedbecausechildren's performanceis more likely to be impeded by these extraneousfactors, becausechildren are believed to have more severe limitations on working memory and a greater bias to analyze sentencesaccording to their extralinguistic knowledge. On the alternative model, Modularity Matching, children should generally perform as well as adults, because children are assumedto "match" adults in processingcapacity and in processingroutines. In chapter 7, we critically review somefindings from the literature that seemto be consistent with the Competing Factors Model and inconsistent with the Modularity Matching Model . There are circumstancesin which the Modularity Matching Model and the Competing Factors Model make similar predictions, however. In these circumstances , "performance" factors playa role in language processing, even according to the Modularity Matching Model. First, performancefactors are brought into play when children (and adults) are confronted with sentencesthat are structurally ambiguous and when these sentencesare evaluated in circumstancesthat are consistent with a range of interpretive options. Although such circumstancesare probably rare in real life, it is essentialto the design of many psycholinguistic studies to construct contexts that are appropriate for more than one meaning of a sentence. In such cases , the Competing Factors Model and the Modularity Matching Model make similar assumptionsabout the interpretation of children's responsesto ambiguoussentences . When principles of linguistic theory make multiple representationsavailable for a sentence, the human sentence-processing device (i.e., the parser) may attempt to resolve the ambiguity by engaging the servicesof real-world knowledge, for example. On the Modularity Matching Model, however, specific context supersedesextralinguistic knowledge. If the specific context surrounding a sentencesupports one reading, but not any others, then that reading will be selected regardlessof its a priori plausibility. We will return to this point in chapter 12. There is another factor that has been found to exert influence in resolving ambiguities in such special circumstances . We mentioned the idea that the sentenceprocessing system attempts to accessa linguistic analysis that makes the sentence true in the discoursecontext. On the Modularity Matching Model, this "tendency" is manifestedin resolving ambiguities. This results in a bias to say "Yes" to either reading of an ambiguous sentence in contexts that are compatible with both interpretations. The perceiver assumesthat the speakerintends to say something that is

Competing FactorsModel true

in

the

domain

ambiguous to

appreciate

that

,

exert

others of

. In

ferred If

or

there

is

,

it

the

" by

phrase ( as

the

Ronald

to

elements

want

use

linguistic

principle message

grammatical

may

to

short

intrude

on

Conclusion

To

sum

make

up

,

the

similar

accounts

,

particular

Model

Competing

bias

accept

interpretation

response

. The

different the

about

the

decisions

in

larity

the

particular

placed

context on

Model only

the

use

. This if

the

, of

,

most only it

conversational

of

another

to

the

the of

same an

be

then

%

pre

still

-

other by ;

.

were

selecting

, -

made A

"

similar

the

verbal by

priori

-

circum

prohibited

a

the there

Similarly

many

contract

any

plausibility

among

of

alternative

Model

,

and

on

and

the

inherent

are

key

ingredients

the

influence

sentence

also

sentences

.

sentences can

assigned

,

regardless ,

operative not

, be .

according only support

however

for a

the ambiguous

particular

a

,

people

's

, to

make

on

their factors

the a

.

decisions

these

on

' s

however

impinge of

a

person

Model ~ tences

Additional to

both for

in

None

assigned

On

preference

se

that

interpretation

Model .

Matching

unambiguous

ambiguous

does

Matching sentences

Modularity

unambiguous

is context

Modularity ambiguous

factors

knowledge

information

the

,

unambiguous

extralinguistic

source

true

and

the

one will

as

Model

of

If

then

50

the

) .

not

Factors

of

hears

of

of .

,

" Mistakes

is

in

and

interpretation

interpretation -

Model

the

as

available

in

English

Competing

over

Model

processing

Matching

one

interpretation the

the

sentence

the

Factors

preferred

often

administration

( including

interpretation

Factors about

about

weigh

a

Competing

Competing

as

structures

that

contraction

Model

the what

of

predictions

factors over

constructions

these

his

processing

Factors about

to

other

with

construction

of

where

on

many

declared

factors

both

an

important

.

predictions the

passive he

linguistic

"

made

in

during

performance

,

Matching

6 .5

On

,

" No

message

of

speakers ,

by

associated

options

optional

verbal

leads

is

sentence

say

particular

production

when

wanna

overruled

is

analyze

. It

ambiguous will

the

hostages

forms

form

representations

Modularity

then

"

to ,

a

the

clear

be an

response

among

the

to

proposition

.

is

for

" reduced

. In

)

made

weapons

and

"

expressing

from

can of

" No

choice

in

omitted

"

attempts

true

children

complementizer

Reagan of

the

of

a

" Yes

sentence

' s

perceiver

reading that

when

occasionally

be

exchange

tendency

the

can

say

found

means

the

only

"

been

the

express

one

ambiguous

one

,

appear

to of

someone

example

may

stances

than

it

bias

often an

influence

. For

fore

in

more

may

this

has

,

makes

selection

more of

accordingly

that

that

, it

even

;

way

the

studies

,

grammar

ing

, on

interpretation

factors

a

however

many

time

discourse

in

pressure

the

a

of

sentence

53

Modu

-

sentence

in

restrictions

are

Modularity

Match sentences

reading

and

.

Chapter 7 Competing Tasks: Reaction Time Studies

According to the Competing Factors Model, linguistic behavior is a composite of many factors, only one of which is linguistic knowledge. On some occasions, children's accessto and use of principles of grammar will remain submergedbeneaththe confluenceof other forces, suchas memory limitations and extralinguistic knowledge. On those occasions, children will make mistakes or they will produce the right re-

sponses for the wrong reasons - not because of their adherence to grammatical principles. On other occasions, the principles of grammar will surfaceand will determine children's responses . If the principles of grammar surfaceoften enough, then children are credited with knowing theseprinciples, where IIoften enough" meansabove some

preestablished level of correctperformancethat would be expectedby chance . The Modularity Matching Model rejects this argument. On this model, if children know a linguistic principle, then they should behave in conformity to it, just as adults do. This follows from the joint assumptions (a) that children have access to Universal

Grammar, (b) that children use the samelanguage-processingmechanismsas adults, and (c) that the grammars of both children and adults have priority over other factors

that contribute to linguistic behavior, such as memory limitations and extralinguistic knowledge. In contrast to the Modularity Matching Model, the Competing Factors Model makes the following methodological assumption. Supposethe researchquestion is whether

or not children

know two structures

that are similar but not identical . If chil -

dren distinguish between sentenceswith these underlying structures (in the adult grammar) often enough, then they can be credited with knowing the linguistic properties of the two structures even if their performancein responseto both types of sentencesis lessthan perfect. The Modularity Matching Model rejects this methodological assumption. On this model, knowing that children treat two types of sentencesdifferently does not prove that they assignthe samegrammaticalproperties to them as adults do; instead, it must be demonstrated that this is the case. The pattern

of children's linguistic behavior must conform to the Modularity Matching Model, according to which children are expectedto perform perfectly, just as adults do (up to the level of experimental noise). Like adults, children should consistently reject sentencesthat are ungrammaticaland accept ones that are grammatical. Only then

canchildrenbe creditedwith knowing the grammaticalpropertiesthat underliethe test sentences

.

7.1 Problemswith ReactionTimeMeasures The different methodological assumptionsof the two models lead to differencesin the selectionof tasksfor assessingchildren's linguistic competence. One way proponents

56

Chapter 7

of the Competing Factors Model have chosen to ascertainwhether or not children distinguish between two structuresis to use reaction time as the dependentvariable. Becausereaction time measuresdetermine significant differences (in the statistical sense) between structures, their use is entirely consistent with the methodological assumptionsof the Competing FactorsModel . From the vantage point of the Modularity Matching Model, however, reaction time measuresshould be used with caution unless they are accompaniedby more

direct measures of children's linguisticknowledge . To anticipatethe conclusions we reach: it is ill advised to use reaction time measures to investigate children 's knowl -

edge of linguistic constraints becausethesemeasuresdo not reveal the syntactic and semantic properties that children assign to the sentences and discourse sequences

under investigation. We take up this problem first, in discussingthe difference between discrete and continuous

variables .

A second problem with reaction time measuresof children's knowledge of linguistic constraints is that they cannot be used to tell whether or not sentences are

unambiguous for children. As we have shown, it is crucial in evaluating children's knowledge of constraints to establish that certain sentencesare unambiguous for them. Reactiontime measuresdo not provide this information. A third reason to avoid reaction time measuresin assessingchildren's knowledge of linguistic constraints is that the usual statistical tests that are applied to reaction time data treat subjects(and sentencetypes) as a group. These tests can yield significant findings even when the responsesof individual children (or responsesto individual sentences) do not conform to the pattern that is characteristic of the group .

Becausestatistical analysesof reaction time measurestolerate (a limited number of ) exceptions, yielding significancedespitethe outliers, suchmeasuresshould be avoided when the researchquestion is children's knowledge of linguistic constraints. Constraints are expectedto be universal and therefore do not tolerate exceptions. .

-

-

We want it to be clearthat theseshortcomingsin reactiontime measures pertain primarily to the investigation of children's linguistic competence; reaction time measureshave yielded substantialprofits in other areas, particularly in the study of adult sentenceprocessing. In that area, studies using reaction time as the dependent variable have been instructive in understanding such phenomenaas when various kinds of information are used in on-line sentenceprocessing, where the loci of difficulty reside in sentenceprocessing, and how ambiguities are resolved. Given the impressive achievementsusing reaction time as a dependentvariable, it might seemnatural, even highly desirable, to extend the use of reaction time measures to investigate children's linguistic knowledge. We urge caution in doing so, however. Appying reaction time measuresto adult sentence-processing routines is justified becauseit can be safely assumedthat adults have all and only grammatical analysesavailableto them. In studies of child languagedevelopment, by contrast, the researchquestion generally concernschildren's underlying linguistic competence. Reaction time measuresdo not set sufficiently high standardsfor attributing knowledge to children. From the perspectiveof the Modularity Matching Model, it is imprudent to relax the criteria used to assesschildren 's competence. This is why we favor mea-

sures that tap linguistic competencemore directly, such as the elicited production task and the truth value judgment task. Once children's competencewith the relevant linguistic structures is established, however, reaction time studieslike those used to study adult sentenceprocessingcan

ReactionTime Studies

57

be usedto studyrelatedquestionsaboutchildren's sentence -processingabilities: how childrenaccessand usedifferentkinds of information, wherethe loci of processing difficultiesresidein children's on-line comprehension of sentences , how childrenresolveambiguities , andso on (for discussion , seeMcKee1996). 7.2 Discrete and Continuous Variables

Essentially, the trouble with reaction time measuresis that they do not reveal what the specificanalysesare that subjects assignto test sentences . To clarify the point, it will be useful to divide measuresof language comprehensioninto two classes : discrete variablesand continuous variables. In experimentsusing discretevariables, eachresponseby a subjectis an instanceof iust one of a limited set of values. The truth value judgment task is one suchmeasure, # becausethe subjectsindicate their acceptanceor rejection of a sentence(i.e., "Yes" or "No") on each trial. The act-out task is another discrete measure, becausesubjects' responsescan be differentiated by the objects that they manipulate and by the order in which they act out events. Grammaticality judgments are another discretemeasure. So are children's productions, sincethesecan be categorizedin specificways. In experimentsusing continuous variables, by contrast, eachresponseby a subject falls somewherealong a continuum. By definition, the probability of the occurrence of any exact value for a continuous variable is zero; subjects' responsesassumethe smooth curves that are characteristicof normal distributions. By their nature, continuous variables mask the specific linguistic analysesthat subjects assign to sentences (for related discussion, see Steedmanand Altmann 1989, 116- 117; Gordon 1996, 212). Reaction time is a continuous variable. Reaction time studies measurethe time it takes subjects to respond to test sentences ; they do not categorize responsesinto discrete classes . Becausereaction time does not indicate the syntactic and semantic properties of the sentencesthat are presentedto subjects, suchmeasuresare not truly informative in assessingchildren's linguistic competence, or in comparing the pattern of responsesby children with the pattern of responsesby adults (whose linguistic knowledge is not in question). The fact that reaction time is a continuous variable becomesespeciallyproblematic when the research question is knowledge of one class of linguistic constraints, namely1constraints on the meaningsthat can be assignedto sentences . The problem is that reaction time measuresveil an essentialdistinction in the assessmentof children's knowledge of constraints, namely, the distinction between ambiguousand unambiguoussentences . As we have shown, children who lack a constraint should find sentencesthat are governed by it to be ambiguous- meaningsthat are ruled out by the constraint shouldbe availableto thesechildren. Children who know the constraint, however, should not find sentencesthat are governed by the constraint to be ambiguous. If a test sentenceis unambiguousfor children, they should consistently reject it in contexts that do not match the interpretation given by their grammars. Therefore, demonstratinQ - children's knowledge - of a constraint boils down to demonstratingthat sentencesgoverned by the constraint are unambiguous for them.! Becausereaction time studies do not indicate whether or not a sentenceis unambiguousfor children, they do not provide an appropriate test of children's knowledge of linguistic constraints. After discussingthe third limitation of reaction time measures , we illustrate

58

Chapter 7

this claim by reviewing a study of children's reaction time using a techniqueknown as the preferentiallookingparadigm. The study, by Hirsh-Paseket al. (1995), was designed to investigate children's knowledge of Principle B. The third limitation of reaction time measures also shows itself in the investigation

of children's knowledge of linguistic constraints. According to the theory of Universal Grammar, linguistic constraints should be universal. This is a consequenceof innate specification, as discussedin chapter 2. Becauseconstraints are putatively universal, all children (and languages) are expectedto adhereto them. The problem with reaction time stems from the custom of analyzing the responsesof subjects as a group , using (parametric ) statistical tests of the combined scores of all subjects. These

statistical tests determine whether or not the subjectsas a group performed significantly above somepreestablishedlevel of performance- for example, a level of performancethat might result by chance1 time in 20 (p < .05). A positive result obtains if a sufficient number of subjects demonstrate a consistent pattern of behavior , over

and above the preestablishedlevel that would be expected to occur by chance.2 As noted , however , it does not suffice to show that the distribution of responses by sub-

jects as a whole conforms to a linguistic constraint; it must be shown that all children abide by the constraint. In this connection, we review a second experimental study using reaction time measures . The study, by McKee, Nicol, and McDaniel (1993), used a reaction time measureknown as cross -modal priming to assesschildren's linguistic knowledge of Principle B. This measurewas used in combination with a discrete measureof children's comprehension; therefore, the study permits us to examinehow a continuous measure and a discrete measure stack up against each other .

7.3 The Preferential Looking Paradigm

Our first exampleof a reaction time study using child subjectsis by Hirsh-Paseket al. (1995). These researchersused a preferential looking paradigm. The dependentmeasure in the preferential looking paradigm is gaze duration, that is, the amount of time (in hundredths or thousandths of a second) that children spend looking at scenes depicted on video displays. The scenesare associatedwith auditory input that is presentedwhile the child subject is attending to the video displays. Typically, two displays are presentedat the sametime. Children tend to look longer at some displays than others; for instance, children have been found to spend more time looking at displays that are consistent with the meaningsthey assign to sentencesthan at ones that are not .

The study by Hirsh-Paseket aleinvestigated children's knowledge of Principle B, by recording how long children looked at scenescorresponding to the licit and illicit (in the adult grammar ) interpretation of a pronoun in sentences like (1). In the adult grammar , the pronoun him is not able to refer to Cookie Monster in (1). This assign-

ment of coreferenceis prohibited by Principle B. As noted in chapter 3, much previous researchusing other taskshas found that children often, but not always, assent to (1) in a situation in which Cookie Monster is turning , but is not turning someone else (e.g ., Chien and Wexler

(1)

1990 ).

Cookie Monster is turning him . a .

b.

"'Cookie Monster isturningh!!!l. ~ okieMonste !: isturninghim.

ReactionTime Studies

59

Hirsh-Paseket al. sought to pursue the issueusing the preferential looking paradigm with the hope that this techniquewould bear better fruit than previous tasks, because it is more " child -friendly ."

In this study, one video display depicted the meaning of the test sentencethat is

compatiblewith the adult grammar ; the other displaydepictedthe meaningthat is ruled out in the adult grammar. For example, one of the trials included the auditory input in (2). There were two video displays corresponding to the input. One display (the "match" display) showed Cookie Monster turning Big Bird; the other (the "nonmatch" display) showed both Cookie Monster turning and Big Bird turning. (2)

Look at Cookie Monster and Big Bird ! Cookie Monster is turning him .

Match display Cookie Monster is turning Big Bird

Nonmatchdisplay Cookie Monster is turning and Big Bird is turning

On the basis of previous researchfindings, Hirsh-Paseket alereasonedthat children who know Principle B should spendmore time looking at the match display than at the nonmatch display.3 Indeed, a group of older children in the studyI who ranged in age from 3;2 to 3;4, looked significantly longer at the match displays than at the

nonmatchdisplays(3.22 secondsand 2.23 seconds , respectively ). On this basis, Hirsh-Paseket aleclaim to have demonstratedthat children know Principle B by age 3, in contrast to other research that has shown that some children coreference This

between

conclusion

pronouns

as old as 5 or 6 allow

and names in sentences like (1).

is unwarranted

, however

. At

issue

is whether

children

know

a

linguistic principle (i.e., Principle B) that prohibits the assignmentof a certain interpretation to sentences. Without

the constraint , children should find sentences gov -

erned by it to be ambiguous. If children lack a linguistic constraint, they should be able to assign the kind of meaning (s) that the constraint rules out . In other words , for

children who lack Principle B, sentenceslike (1) will be ambiguous; thesechildren will

be ableto assignnot only the legitimatemeaning , but alsothe meaningprohibitedby Principle B. This situation can be depicted as follows: Child Adult

(sentence,{meaningl' meaningz} ) (sentence,{meaning}} )

Having laid the groundwork, we can now explain why the findings of the study by Hirsh-Pasek et al. do not demonstrate young children's knowledge of Principle B. Supposethat the 3-years-olds in this study lacked Principle B. If so, then both of the displays these children witnessedwould have depicted meaningsthat were consistent with their grammars. That is, both displays would have been "matches." Now, are children likely to look longer at one of the displays than the other, if both displays match a reading that is generatedby their grammars? It is likely . Just as adults do, children may prefer one meaning of an ambiguous sentenceover another. Children who prefer one interpretation should be expectedto look longer at the display that matches that interpretation. In the present context, if children's preferencefavored the display that depicts the meaning associatedwith the legitimate (adult) interpretation, then the findings could be explainedwithout appealingto Principle B. In short, there is a confound between (a) children's preferencefor one meaning of an ambiguous sentenceover another and (b) their knowledge of the linguistic

60

Chapter 7

principle governing the interpretation of pronouns. More generally/ without adequate controls, the preferential looking paradigm cannot distinguish in principlebetween children's knowledge of linguistic principles and their preferenceto resolve ambiguities in favor of one reading rather than another (see Lasnik and Crain 1985/ for a similar objection to the act-out task). As its name suggests, the preferential

looking paradigmcanbe usedto establishthe preferences of subjectsfor onedisplay over others. However, subjects' preferencesto perform in certain ways do not suffice to guaranteetheir adherenceto linguistic principles. To deal with the potential confound between preferencesand principles, studies using th~ preferential looking paradigm could borrow a researchstrategy discussed earlier, in the context of the truth value judgment task- namely, to comparechildren's responsesto ambiguous sentenceswith their responsesto sentencesthat are unambiguous, becauseof a linguistic constraint. (For example, we describedthis research

strategyin chapter3, asit wasappliedto evaluatechildren's knowledgeof Principle C (also see chapter 27).) The strategy involves examining children's responsesto both ambiguousand unambiguoussentences . The ambiguoussentencesserve as controls. For example, in investigating Principle C, we would first find out how often children accept coreference between the name, the Ninja Turtle, and the pronoun , he,

in sentences such as (3). (3 )

The Ninja Turtle

danced while he ate pizza .

Then we would compare the proportion

of coreference that children assign to am-

biguous control sentenceslike (3) with the proportion of coreferencethat children assignto sentenceslike (4), where coreferenceis ruled out by Principle C. (4)

He danced while the Ninja Turtle ate pizza.

Children who lack Principle C should find both sentences(3) and (4) to be ambiguous, so the coreferenceinterpretation should be assignedequally often to both sentences . On the other hand, children who know Principle C should assign only the noncoreference , or deictic, interpretation of (4)i this sentenceshould be unambiguousfor these children, just as it is unambiguousfor adults. Nothing prevents us from applying the sameresearchstrategy to study children' s knowledge of Principle B using reaction time as the dependent variable. To demonstrate children's knowledge of Principle B, for example, the first step would be to record the averagelength of time children look at two different displays- ones that correspondto the alternative interpretations of ambiguoussentences , such as (5). (5)

The Ninja Turtle tried his pizza.

a. I ~e Ninja Turt!~ tried ~ pizza b.

The Ninja Turtle tried his pizza.

Presumably, children would spend more time looking at one display than another. If fortune shines, children could prefer, and therefore look longer at, the display corresponding to the coreferenceinterpretation of the pronoun, (a), than at the display corresponding to the deictic interpretation, (b). Having establishedchildren's gaze durations in response to ambiguous sentences , the next step would be to present sentencesgoverned by Principle B, suchas (6). (6)

The Ninja Turtle covered him .

ReactionTime Studies a. b.

61

"'The Ninja Turtle covered him . The Ninj a Turtle covered him .

As in the truth value judgment task, the displays for (6) would correspond to both

the readingthat is permittedby the adult grammar , the deicticreading(b), and the reading that is ruled out by Principle B, the coreferencereading (a). If children lack Principle B, the amount of time they spend looking at displays corresponding to the coreferencereading of sentencessuch as (6) should be similar to the amount of time they spend looking at displays corresponding to the coreferencereading of the ambiguous control sentences , such as (5). If children know Principle B, then they should spend significantly more time looking at displays correspondingto the deictic interpretation for test sentencesthan they do looking at displays corresponding to the deictic interpretation of ambiguouscontrol sentences . Although the study just outlined would overcome the objection leveled agains.t the study by Hirsh-Paseket al., we would still have qualms about concluding that Principle B is innate based on such a study. As noted, one hallmark of innatenessis

that a linguisticprincipleshouldbe accessible to all children.Evenif thereis a significant differencein gaze duration by subjectsas group in responseto the displays corresponding to the coreferenceinterpretation of (5) and (6), we recommendcaution in interpreting the findings as evidence of innateness. Here is how Crain and Wexler (forthcoming) put the point: Suppose, for example, that subjectsfrowned more often when they heard one sentence than when they heard another sentence. There is no reason to think

that the difference in behavior representsa differencein grammaticalstatus of the two sentences . It may simply be that children failed to understand one of the sentences. Knowing that two sentences are not treated in the same way by

subjectsdoes not tell us about the grammaticalproperties they assign to each type of sentence.

To take another example, supposethat children showed a difference on the sametwo types of sentencesin an experiment using a different dependentmeasure , say reaction

time . All that we could conclude would

be that there is a dif -

ference for the children between the two types of sentences. We would not conclude that children know the syntactic and semantic properties of the sen-

tences. ... In ordet: to demonstrate children's knowledge of UG principles, it must be established that their pattern of responses conforms to the model ; that

is, to the relevant principles of UG. To concludethat the constraint .. . is innate . . . it must be shown that all (or at least the vast majority tested adhere to the constraint .

of ) children

who are

In the next section we follow up on the last point made in this quotation : that re-

action time studies in particular, and continuous variables in general, often conceal individual differencesin performance. 7.4 Cross-Modal Priming

Another example of the use of reaction time measuresto investigate child language is a study by McKee, Nicol, and McDaniel (1993). These researchersinvestigated children 's processing of coreference relations between ordinary pronouns (e.g., him,

62

Chapter 7

her, them) and their linguistic antecedents , and betweenreflexives (e.g., himself , herself , themselves ) and their antecedents . They adapteda techniquefrequently used to investigate sentence processing in adults : cross-modal priming . In a cross-modal priming

study, sentencesare presentedauditorily. As the sentenceunfolds, subjectsare asked to respond to a display that is presentedvisually. In many studies, the visual display is a string of letters , and the subjects' task is to indicate whether or not the string

of letters constitutes a word. Subjectsare instructed to respond IIYes" or IINo" by pressing the appropriate responsekey. In studies of lion-line" sentenceprocessing, subjectsare requestedto respond (as quickly aspossible) while they continue to listen to the remainder of the stimulus sentence. If subjects respond quickly enough, responses come as the sentence unfolds

in real time (i .e., before the sentence has been

heard in its entirety). This makesit possibleto examinethe influenceof specificproperties within a sentenceon subjects' responsesto the visual display. In the cross-modal priming technique used in the study by McKee, Nicol, and McDaniel, 17 children aged 4 to 6 (and a larger number of adult controls) listened to sentencesover headphonesand, for each sentence , responded to a picture that was displayed at a critical juncture in the sent~nce. The picture depicted an animateor inanimateobject, for example, a leopard. The subjects' task was to indicate whether the depicted object was "alive" or "not alive." Subjectsindicated their decision by press-

ing one of two response keys.

A "YesjNo" judgment task was also administered to the subjects. The judgment task was a picture verification procedure, in which subjects were asked to indicate whether or not a test sentence accurately described a picture . This was an " off -line "

task, becausesubjectslistened to sentencesin their entirety before responding to the correspondingpictures. In the cross-modal priming portion of the study, the critical manipulation was the alternation of an ordinary pronoun (e.g., him) with a reflexive pronoun (e.g., himself) or with a definite description (e.g., the girl ). One of these items appearedin each test sentence. To form a complete set for a given sentence , the same picture was

displayedimmediatelyfollowing eachof theseitems (as indicatedby the word [LEap ARD]). An exampleis (7). (7) The alligator knows that the leopard with green eyes is patting him/ himself/ the girl [LEOPARD]' on the head with a soft pillow . McKee , Nicol , and McDaniel

reasoned as follows . If children

and adults process sen-

tencesin the sameway, subjectsin both groups should respond fastestto the version of the test sentencesthat contain a reflexive pronoun. This prediction is based on previous research with adults, where it was found that pronominal elements re-

activate, or "prime," the meaning of words that subjectsassociatewith their linguistic antecedent. In the present study, then, the reflexive pronoun was expected to prime the preceding NP the leopardsincethis NP is the linguistic antecedentof the reflexive pronoun. What about ordinary pronouns? Assuming that subjectsassociatethe pronoun , him , in (7) with

an NP that is mentioned

in the sentence , the pronoun

should

shareits referencewith the NP the alligator. If so, there should be no advantagein responsetimes to the picture of a leopard when the critical word was an ordinary pronoun. In short, the prediction was that responsetimes indicating that the picture of the leopard depicted an object that was "alive" would be faster to the reflexive, himself, than to the pronoun, him (assumingthat children have the samelinguistic knowl-

63

ReactionTime Studies

edge

as

girl

adults

does

do

not

serves

only

as

There

is

tween

a

an

For

a

on

( 8

a

)

Papa

Returning

to

excluded

the

by

tioned

in

study

.

B

the

reflexive

for

both

,

,

some

.

always

refer

pattern

of

to

,

pronouns

icantly

,

A

What

for

should

we

statistically

as

sidered

as

individuals

if

,

the

it

remark

their

.

)

may

then

take

by

object

the

object

and

ordinary

a

leopard

-

do

to

definite

de

because

-

these

.

The

advance

responded

or

definite

Lacking

,

exact

because

.

adults

pronouns

the

.

they

the

-

the

pro

of

picture

in

men

of

as

,

the

-

are

ordinary

reflexives

in

con

is

leopard

picture

predicted

depicted

that

the

to

the

depicted

children

is

either

him

the

compared

be

It

that

the

pronoun

to

.

NP

,

NP

as

again

an

,

to

,

)

with

the

be

the

to

the

the

faster

the

to

definite

description

observation

that

adults

do

principles

?

?

between

statistical

did

of

the

re

-

descriptions

not

.

As

additional

we

.

differ

signif

to

the

the

-

and

Wexler

by

the

confirmation

from

same

more

and

there

other

NPs

of

the

responses

con

subtypes

)

remark

:

that

of

knowledge

direct

tests

-

of

mimicked

underlying

a

children

children

by

children

was

for

( forthcoming

differences

attribute

overall

children

Although

behavior

of

same

that

.

patterns

Crain

the

conclude

and

. to

time

- advised

show

necessarily

different

reaction

ill

Should

Not

applied

obscure

group

children

reflexives

test

could

be

without

to

of

grammatical

and

McDaniel

.

was

prophetic

of

children

analysis

patterns

( 8

.

NP

assigned

( 7

with

to

cannot

Both

as

pattern

Although

further

,

-

not

the

possibilities

responses

pronouns

and

overall

would

knowledge

study

the

This

the

children

on

,

within

adults

either

linguistic

group

.

then

consider

children

presumably

refer .

difference

whole

Even

,

behavior

same

a

a

,

do

with

meaning

coreferential

quickly

faster

to

from

significant

analyzed

be

adults

corefer

the

preceding

pronouns

conclude

the

.

.

linguistic

have

where

to

pronoun

these

as

be

pronoun

group

Look

of

to

the

either

Closer

pattern

adults

to

NPs

coreference

coreference

be

For

reflexive

not

,

allow

.

would

findings

than

times

,

sentences

.

immediately

would

need

as

McDaniel

the

leopard

ordinary

but

of

children

well

could

corresponding

essential

and

respond

result

to

may

Response

The

NP

,

the

responses

the

the

the

and

fastest

responses

are

flexive

The

pronouns

The

pronouns Here

or

NP

the

test

tolerate

these

as

the

types

him

For

,

would

pronoun

might

.

ordinary

Nicol

prime

children

himself

scriptions

7 . 5

alligator

) .

"

associate

the

sometimes

two

deictically

study

children

then

3

description

of

children

pronoun

interpreted

,

the

these

,

the

might

Principle

If

sentence

definite

.

McKee

in

The

version

other

some

the

chapter

is

) .

this

preceding

allow

himself

him

by

adults

NPs

him

covered

children

( see

;

the

,

immediately

covered

him

all

with

to

)

B

at

earlier

an

( 8

pronoun

some

the

preceding

noun

like

Principle

NP

noted

and

Bear

Bear

that

As

appear

" Papa

the

any

comparison

.

sentence

which

ceivable

for

however

children

meaning

adults

the

,

some

in

of

of

pronoun

,

the

knowledge

yardstick

ordinary

Bear

on

including

reactivation

catch

example

Papa

,

involve

it

of

the

and

responses

situation

adults

turned

out

.

The

that

arose

produced

similar

that

there

need

for

were

further

in

McKee

,

patterns

differences

analysis

Nicol

of

,

responses

between

of

the

as

the

reaction

' s

a

group

,

groups

time

in

data

64

Chapter 7

became evident , however , only when McKee , Nicol , and McDaniel considered the results of the picture verification task. On the basis of the picture verification task, it was determined that 7 of the 17

children did not make adultlike judgments for sentenceswith ordinary pronouns. The behavior of these 7 children was quite unlike that of children examined in previous

studies, however. These 7 children treated ordinary pronouns as if they were reflexives. That is, the exceptionalgroup of children in McKee, Nicol, and McDaniel's study rejectedthe direct reference(deictic) interpretation of pronouns. The finding that is consistent with earlier researchis that these children acceptedcoreferencebetween ordinary pronouns and an immediately preceding referential NP. A post hoc examination of the reaction time data of these 7 children was quite re-

vealing; McKee, Nicol, and McDaniel found that both reflexive pronouns and ordi-

nary pronounsshowedpriming effects , ascomparedto the definitedescriptions . The remaining 10 children showed a priming effect only for reflexives . Thus, the reaction

time data mirrored the judgment data, when the children were divided into subgroups on the basis of the results from the judgment task. The similar pattern of results in both tasks is important , because it means that reaction time studies can be

revealing about children's linguistic processing, if their grammatical knowledge is documentedby an independentassessment , basedon a discretemeasureof linguistic competence. Without the judgment task, McKee , Nicol , and McDaniel would have

had no causeto examinethe responsesof subgroupsof children. On the basis of the similar overall pattern of response times for both groups , the wrong conclusion could have been drawn : that children and adults have the same sentence-processing abil -

ities. To put the matter differently, if only reaction time data had been used, it might have been inferred that the children and adults in the study had similar grammars. But this inference is warranted only if it is reasonableto assumethat the pattern of responsetimes by both children and adults is a reflection of their grammaticalknowledge. If this assumption were correct, then the finding that children and adults showed a similar pattern of behavior would license the inference that they have sim-

ilar grammars. However, the study by McKee, Nicol, and McDaniel amply demonstrates that this assumption is incorrect .

Having criticized the use of reaction time measuresto study language development , we should say again why the use of such measures is justified in studies of

adult sentenceprocessing. The reasonis that linguistic competenceis not at issuein studies with adults . Because it is assumed that all adults have the correct grammatical

analysesavailableto them, tasksmay be used that only indirectly tap their grammatical competence , such as measuresof reaction time. Clarity is sacrificedin looking at the particular syntactic and semantic properties of sentences , for example, to determine the time course of the availability of alternative sourcesof information. As Gordon

(1996 ) remarks :

A major difference in the aims of language acquisition studies and those of adult studies is that the latter tend to test for performance variables where reaction times are used to make inferences

about the structure

of and access to

existing knowledge. (p. 212) In studies of children's language development, IIexisting knowledge'! is precisely what is at issue. As we have argued, the discovery that children and adults treat sentences in the same way does not entail that both groups analyze sentences in the

ReactionTime Studies

65

sameway. In order to demonstratechildren's knowledgeof particularprinciplesof Universal Grammar , it must be established that their pattern of responses conforms

to those principles. Once this has been established, however, the findings of the study by McKee , Nicol , and McDaniel become important , because they suggest that further questions about the availability of different sources of information in chil -

dren's processingof linguistic information may also be open to empiricalinvestigation. The findings of the study by McKee, Nicol, and McDaniel are quite instructive. They suggest that the processing routines of both children and adults are intimately

tied to their grammatical knowledge and that these routines may be equivalent for both children and adults, as anticipated by the Modularity Matching Model . The downside of using reaction time studies should also be evident . In the absence of

more direct tests of children's linguistic knowledge, measuresthat are essentially nondiscrete , such as reaction time , can sometimes obscure important

aspects of

children's linguistic knowledge, by obscuring individual differencesamong children. To conclude our discussion of the study by McKee , Nicol , and McDaniel , we wish

to tie up a few loose ends. First, we wish to underscorea point 'we will raisein chapter 9 regarding the imitation task. The relevant finding is that both children and adults show priming effects for reflexive pronouns. This finding invites the inference that people actively attempt to establishthe referents of NPs in the sentencesthey hear, even when sentences are present~d outside of any context .

Another point worth making also anticipateslater discussion. The point concerns how the referents

of NPs are determined

in the absence of a conversational

context

(i .e., in the so -called null context ). In the absence of context , as in both of the tasks

used by McKee, Nicol, and McDaniel, the direct referenceinterpretation of ordinary pronouns

seems to be unavailable . This was the case with the 7 children

coreference too liberally ; these children showed compara'ble priming

who allowed

effects with

ordinary pronouns and reflexive pronouns. Presumably, this is simply an artifact of the tasks; an anaphoriclink can be madeto a linguistically establishedantecedent, but no such link can be made to a referent

that has not been established

in the domain

of discourse. The priming effects with ordinary pronouns could be expected to be diluted if extrasententialreferents are available. In fact, intersentential anaphoramay not be preferred if there is a choice between inter- and extrasentential reference. If the goal is to understandnormal language-processingroutines, then the cross-modal priming technique is likely to present a distorted picture of children's sentenceprocessing routines . This is a consequence of using a task that requires children to construct sentence meanings in the absence of contextual support . 7.6 Conclusion

This chapter discussedthree limitations to using reaction time measuresin the study of children's linguistic knowledge. One limitation is that such measurescannot be used to establishwhether or not a certain type of sentenceis unambiguousfor children. We discussedthis problem in the context of the preferential looking paradigm. In this task, children are askedto respond to the sametest sentencein two different contexts , one that is consistent with the unique adult interpretation of the sentence

and one that requiresa nonadult interpretation. The finding of the study we discussed was that children generally respondedfaster in the context that supported the adult interpretation . The experimenters concluded on this basis that children have accessto

66

Chapter 7

the adult reading of the target sentence,but not to the other reading. This conclusion was unwarranted, however, becausethere was another interpretation of the findings: that children accessedboth'readings, but simply preferred the adult reading. Another limitation of reaction time measures is that they do not reveal the nature

of the linguistic representationschildren assignto sentences . This limitation was discussedin the context of a study using the cross-modal priming technique. In that study, children showed a differencein reaction time to two types of sentences , just as adults

did . It could

not be inferred

that children

knew

the syntactic

and semantic

properties of the test sentences , however, even though their pattern of reaction time differences

mirrored

that of adults .

The final limitation of reaction time measuresis that they lump all subjects together and ask about their performanceas a group. The performanceof individuals

and subgroupscanbe washedout if the majority of childrenbehavein a particular way . To underscore this point , we noted that when a more direct test of grammatical competence was undertaken , the initial reacti9n time data were shown to be mislead-

ing. In the light of the additional measurement , the study revealeddifferencesamong children in reaction time that were obscured by the initial analysis of the findings. The lesson to be learned

is that one can be confident

that children ' s responses

are

basedon linguistic knowledge only if they produce a targeted linguistic construction in appropriate circumstances , or consistently and correctly judge sentencesto be grammatical or ungrammatical , just as adults do .

Reaction time measuresare useful in finding out if subjects produce different patterns of responsesto different sentencetypes or if they produce different patterns of responses to the same sentence type in different contexts . In both cases, statistical tests are used to argue for or against the claim that subjects' responses are influenced

by a "true" component of behavior, such as linguistic knowledge. On the Competing Factors Model, significant differencesin responsetimes by children might be construed as evidence of linguistic knowledge. According to the Modularity Matching Model, however, reaction time data alone do not provide a reliable basisfor making inferencesabout children's grammatical competence. In the absenceof independent assessmentsof children's linguistic knowledge, based on discrete measuressuch as the elicited production task or the truth value judgment task, the results of reaction time studiesmust be interpreted with caution. The next two chapterstest the Competing FactorsModel in a different way, as it has been applied in interpreting the findings of studies that use other researchmethodologies. Chapter 8 examinesthe act-out task and chapter 9 examinesthe imitation

task .

Chapter8 CompetingTasks:The Act-Out Task

This chapter continues our discussionof the Competing Factors Model. The Com-

petingFactorsModel anticipatesthat childrenwill sometimes performlesswell than adults. Evidence of this would seem to be abundant. Children often fail to perform

aswell asadultsdo in manypsycholinguistic tasks, andthey sometimes producedifferent responsesthan adults when they confront the same linguistic materials. The Modularity Matching Model maintains that both the linguistic and performance componentsof the languageapparatusof children and adults are basically the same; this leavesthe Modularity Matching Model with little room to maneuverin explaining why children sometimesperform with less accuracythan adults on certain tasks and why they sometimesproduce systematicnonadult responses.In chapters2 and 4, we noted

one source of differences

between

children

and adults

that is consistent

with the Modularity Matching Model: differencesthat arise becausechildren adopt

distinctgrammaticalhypotheses . However, the problemof generallydecreased performanceby children, as comparedto adults, remainsto be addressed . This chapter and the next examine findings that seem, on the face of it, to be inconsistent with the Modularity Matching Model and supportive of the Competing Factors Model. The findings are less than perfect linguistic behavior by children on two tasks : act -out and imitation

. The recalcitrant

data do not force us to abandon the

Modularity Matching Model, however. The tasks that have evoked less than perfect performanceare highly unnatural and consequentlydo not accuratelyassesschildren's linguistic competence. In many (but not all) experimentaltasks, linguistic behavior is at the mercy of irrelevant task demands. The researchdesignsthat have evoked errors from children are typically onesthat are especiallysensitiveto irrelevant nonlinguistic

factors, as well asbeing sensitiveto grammaticalcompetence . This underscores the needto designexperimental paradigmsthat arefreefrom the effectsof confounding influences. One of the overarching goals of this book is to illustrate how to rid tasks of the influence of extraneous and confounding factors. First, however, let us look

more closelyat the problemsthat can arisewith tasksthat are frequentlyusedin assessingchildren's linguistic competence. This chapter examinesfindings from the act-out task. Chapter 9 examinesfindings from the imitation task. For illustrative purposes, we will cite two examplesof researchstudies using the act-out (or figure manipulation ) task. In this task, the experimenterpresentsa sentence to the child subject and instructs the child to act out (" Do what I say" ) the sentence

using toys and props that are present in the experimentalworkspace. We will argue that the results of this task canseriouslyunderestimatechildren's linguistic knowledge. The first study was mentioned already, when we pointed out the suggestion that young children attempt to interpret constructions in an " order -of-mention " fashion .

This study investigated children's understanding of temporal terms like beforeand

68

Chapter 8

after. In comprehendingtwo-clauseconstructionsthat are tied together by these temporal conjunctions, even children as old as 5 were found to be much more successful in responding to instructions like (1) than in responding to instructions like (2). (1)

After

you touch your nose , touch your ears .

(2 )

Before you touch your nose , touch your ears .

Having noted that the event mentioned first in (1) should be acted out first, but that there is a conflict between order of mention and the correct conceptualorder in (2), researchersconcluded that children lacked the grammaticalresourcesfor interpreting sentencescontaining subordinating temporal conjunctions. It was claimed that, lacking grammaticalknowledge, children relied on order of mention as a strategy for interpretation (e.g., Amidon and Carey 1972; Clark 1971). This strategy gives the right results for (I ), but the wrong

results for (2 ).1

The conclusion that children lack grammatical competencesits poorly with the Modularity Matching Model. Fortunately, however, another source of children's errors was identified by Crain (1982) and further pursued by Gorrell , Crain , and Fodor (1989). These authors argued that children found the instructions in (1) and (2) odd,

becausethesesentenceshave presuppositionalcontent that was flouted in the experiments that evoked high error rates. The presupposition is that the person being addressedhas already expressedan intention to perform the action mentioned in the subordinateclause. To satisfy the presupposition, all that is required is a lead-in sentence like th~ following : "In the next game I want you to touch some part of your face." As the gameunfolds, the child is askedto say what part of her face sheplans to touch . For a child who says, "J'm going to touch my nose," either (1) or (2) would be an appropriate request. When the presupposition was met in this way , children 's cor-

rect responsesclimbed to 82% accuracy in Crain's study. This figure excludes an 'Lother" response that accounted for almost half of the errors : that of acting out only

the main clause. Intuitively , this may also be a "correct" response; seeHamburger and Crain 1982 for discussion . In any event , it is apparent that children are extremely

sensitive to a presupposition that is subtle enough to have been entirely overlooked by numerousexperimentalpsycholinguists. A secondexampleof children succumbingto infelicity appearedin researchon the acquisition of restrictive relative clauses. In severalearly studies on this construction, children were found to commit errors in interpretation . The errors resulted from

experiments using an act-out methodology to assesschildren's comprehension of sentences with relative

clauses . In several such studies , children

committed

systematic

errors in comprehending sentenceslike (3)- (6). The differencesamong these construction types are indicated by the two-letter code preceding eachexample; the first letter in the code identifies the grammatical function of the NP that bears the restrictive relative clause, subject (5) or object (0 ) of the main clause; the second letter

indicatesthe grammaticalfunction of the "gap" within the relative clause. For example, an as

sentence is one in which

the relative

clause modifies

the main -clause direct

object (0 ), and the relative clause has a gap in subject position (5) .

(3) 55:

The dog that jumped over the pig pushedthe sheep.

(4) 00 : The.dog pushedthe sheepthat the pig jumped over. (5) 50 : The dog that the sheeppushedjumped over the pig. (6) as : The dog pushedthe sheepthat jumped over the pig.

Act-Out Task

69

Severalstudies found that 4- and 5-year-old children consistently acted out these sentences in a nonadult fashion (Sheldon 1974; Tavakolian 1981). One type of sen-

tence was particularly likely to evoke errors: the as sentence.When askedto act out the meaning of the as sentencein (6), many children had the dog push the sheepand then jump over the pig. For adults, the sheepis the agent of the action expressedin the relative

clause in (6 ). Children ' s nonadult

enactment

led Tavakolian

(1981 ) to

claim that they assigned sentenceslike (6) a structure appropriate for conjoined clauses, as if (7a) had the meaning of (7b). Accordingly, Tavakolian called her proposal the conjoined -clauseanalysisof relative clauses. (7) a. The dog pushedthe sheepthat jumped over the pig. b. The dog pushedthe sheepand jumped over the pig. According to Tavakolian, children initially lack the principles needed to interpret relative clauses in an adultlike fashion . Therefore , when they interpret sentences with

relative clauses , they are compelledto borrow syntactic and semanticprinciples from Universal Grammar that are used for other grammatical structures (i.e., for conjoined

clauses). These linguistic principles apply to some sentencesbetter than to others, however . The conjoined -clause analysis was able to account for 63% of children 's

responsesto as sentencesin Tavakolian's study, but it accountedfor only 19% of children's responsesto 00 sentences . In total, the conjoined-clause analysis explained 48% of all of children's responsesto the four types of relative clauses.

Later, using other experimentalprocedures , it was demonstratedthat Englishspeaking children and Italian-speaking children have mastery of sentenceswith a relative clause even before their third birthday (but see Labelle 1990, where French-speaking children's difficulty in producing standard adultlike relative clauses in some types of sentences is documented ; also see Guasti and Shlonsky 1995, Guasti

et at. 1995). The modified procedureswere motivated by the observation that sentences with relative clauses often bear two kinds of presuppositions . Consider the as sentence in (8).

(8) The dog pushedthe sheepthat jumped over the pig. The relative clausein this examplemodifies the NP the sheep , forming the constituent the sheepthat jumpedover the pig. Felicitous use of this phrasepresupposes , first, that at least two sheep are present in the conversational context. If there is only one sheep, there is no need to use the relative clauseat all; the speakercould just as well say, liThe dog pushed the sheep." In short, a restrictive relative clauseis felicitous when some restricting needs to be done. With only one sheep present in the experimental workspace, the restrictive relative clausethat jumpedover the pig is superfluous ; the sentence therefore

violates

one of the Gricean maxims

of manner : IIA void

Unnecessary Prolixity " (Grice 1975). The second presupposition of (8) is that the event mentioned in the relative clause

(the sheepjumped over the pig) took place prior to the assertion(the dog pushedthe sheep).2 Given that the past tense form of the verb is used, the event mentioned in the relative

clause should have occurred before the sentence was uttered .

In the researchthat evoked high error rates from children, neither of these presuppositions was satisfied. Only one sheep was present in the experimental workspacefor (8), and no event occurredbefore a test sentencewas presented; the child's task was simply to IIact out" the meaningof the test sentencesfrom scratch, using the available objects .

70

Chapter 8

To investigate the possibility that experimentalinfelicities, and not lack of linguistic competence, were responsible for children's comprehensionerrors, Hamburger and Crain (1982) made the apparently minor changeof adding more sheepto the experimental workspace for a sentencelike (8). This simple change resulted in a much higher percentageof correct responsesby children, including children much younger than the ones tested previously. However, the most compelling evidencethat young children command the structural knowledge underlying relative-clause formation, came in a second experiment, using an elicited production task. Pragmatic contexts were constructed in which the presuppositions of restrictive relatives were satisfied; then children were instructed to ask someone who was blindfolded to identify one toy from a set of identical toys . In such circumstances, even children as young as 2;8

consistently produce well-formed subject and object relative-clause constructions (Crain , McKee , and Emiliani

1990 ).

In our view , young children 's conjoined -clause interpretation of sentences with rel-

ative clausesis simply the result of their attempt to behave in a manner consistent with their knowledge of pragmatic principles (seeSperberand Wilson 1986). This is why children's level of correct performance approximates that of adults when the pragmatic felicity conditions are satisfied. Why do older children and adults not fall victim to pragmatic infelicities to the

sameextent as young children do? We believe it is simply becauseolder children and adults have learned to "see through" misleading circumstancesin which test sentencesare presented. To be successfulin previous studies, subjectswere required to accommodatethe presuppositional failures in the experiment. It seems that older children and adults are capable of the necessarymental gymnastics, at least for unambiguous sentences(we have more to say about this later); but many younger children are apparently unable to perform the necessaryaccommodationsas rapidly or as successfully. It should be understood that we are not condemning all studies using an act-out methodology. The task is not inherently defective; there are placeswhere it could be put to good use. It has a major drawback, however, becauseit introduces sentencesin the so-called null context. Later, we describethe inner workings of a different comprehensiontechnique, in which context is under the experimenter's control. This alternative technique for assessingchildren's comprehensionof sentencesis the truth value judgment task, to be discussedin part III.

Chapter 9 Competing Tasks: Imitation

There is another task in which the linguistic behavior of children and adults differs: the imitation task. In experimentsusing the imitation task, children have been found to perform in a characteristicallynonadult fashion. The idea behind the imitation task is to see whether children can correctly repeat sentencespresented to them, or whether they change the input sentencesin some way. It is anticipated that the changes - children makewill indicate how their underlying grammardiffers from that of an adult. In this chapter, we review some representativefindings from studies using the imitation task, to determine the extent to which the findings challengethe Mod ularity Matching Model or canbe explainedby it . 9.1 PreviousFindings In someimitation tasks, young children have been found to omit whole chunksof linguistic material. For example, in a study by Phinney (1981), 3- to 5-year-old children were askedto repeat sentenceswith overt complementizers , such as (1). (1) The bear said that the turtle tickled the horse. Many of the youngest children's "repetitions" consisted only of simple sentences , whereas older children successfullyrepeated both clauses; but many of the older children neverthelessfailed to reproducethe complementizerin Phinney's study. Assuming that adults do not experiencedifficulty with this task, the findings from Phinney's study seem once again to challenge the basic tenets of the Modularity Matching Model. We disagree. First, the omission of optional lexical material, such as the complementizer that in (I ), is not problematic for the Modularity Matching Model; this omissionby children could be the result of a parsing tendency that is also characteristicof adults, namely, the tendency toward using reducedforms. We therefore disagreewith Phinney, who interpreted this finding as indicative of a nonadult grammar. Our own view is confirmed by the results of elicitation experiments, where it is found that children even younger than the ones in Phinney's study can reliably produce complementizersin linguistic contexts in which they are highly preferred, for example, in questionsthat begin "Is it true that . . ." (Thornton 1990). What appearsproblematic for the Modularity Matching Model, however, is the main finding: that young children omit whole chunks of linguistic material in the imitation task. Of course, it would be easyto say that they failed to understandthe task, or that they didn't seeany point to it . But taking the finding at face value, there are at least two ways to explain children's nonadult behavior, short of abandoning the Modularity Matching Model. First, some of children's incomplete repetitions may be attributable to problems in lexical access . If children do not rapidly accessthe

72

Chapter 9

meaningsof the lexical items they are askedto repeat, the working memory system may overload. The result could be fragments of unanalyzedmaterial in the working memory buffer. It is therefore important to make sure that children have ready ~ access to all of the vocabulary items they are presentedwith in an imitation task. If children have not learned to retrieve all of these words with sufficient speed, their ability to perform the steps in syntactic and semantic processing needed to repeat an entire sentencemay be hindered. 9.2 Problemswith the Task Even if this step is taken, children may experiencedifficulties with the task beyond those experiencedby adults. In chapter 8, we reported children's misunderstandingof sentenceswith a restrictive relative clausein the absenceof context, becausethe ,80called null context does not satisfy the pragmatic "felicity conditions" associatedwith this construction. The sameis true of the imitation task. Becausesentencesare presentedin the so-called null context, the task is highly unnatural; it also fails to satisfy any presuppositions the test sentencesmight have. Consider example (1): The bear said that the turtle tickled the horse. The use of definite NPs- the bear, the turtle, the horse - presupposesthat some bear, turtle, and horse have been introduced into the conversational context.! If children (and adults) are required to (mentally) establish the referentsof thesedefinite NPs in order to produce the sentence, then this process could interfere with verbatim recall. Chapter 12 discusseshow the processof adding referentsto one's mental model of the context associatedwith a sentenceis costly for adults, too, in certain cases. For adults, the easeof setting up alternative referents is revealedin their preferencesfor interpreting ambiguous sentencesin one way versus another, when ambiguous sentences are presented in the absenceof linguistic or nonlinguistic context. Our suggestion is that children, even more than adults, find it highly unnatural to perform tasks that require them to analyze sentencesin the so-called null context. This has little , if anything, to do with children's linguistic knowledge, on the assumptionthat the accommodationof presuppositionalfailures is an ability that develops in children over time, as they gather more experiencewith the conventions of conversation. The imitation and act-out tasks may be unnatural for children for this reason. The conclusion reachedby Hamburger and Wexler (1973) is that the primary linguistic data for grammar formation must consist of (sentence, meaning) pairs; these authors proved that grammars cannot be learned without their associated~ eanings (i.e., on the basis of sentencesalone). It has often been suggestedthat sentencemeaningsare inferred by children from the discoursecontext, although no one deniesthat the processesby which such inferencestake place are poorly understood (see Wexler and Culicover 1980, Pinker 1984, and Gleitman 1990 for discussionof this assumption). For our purposes, the point is this. Becausechildren are in the processof language learning, they may rely on immediate context to support sentencemeaning to an extent that is not characteristicof adults. Therefore, children's relatively poor performance on the imitation task and the act-out task may not directly tap the normal language-processing routines of either children or adults; somehow, older children and adults have learned to compensatefor the absenceof context. Poor performance on these tasks by young children is, therefore, not inconsistent with the Modularity Matching Model .

Imitation

73

9.3 Comparing Tasks Now

that we have discussed the act -out and imitation

tasks in some detail , it will be

useful to look at how the results from experimental studies using these tasks have been interpreted by advocatesof the Competing Factors Model. Between-task comparisons have been made both in first language acquisition (Lust, Chien, and Flynn 1987) and in secondlanguagelearning (Flynn 1986a,b). Both

lines of research have reached

similar

conclusions . One conclusion

is that

production (i.e., the imitation task) provides a more direct assessmentof underlying linguistic competencethan does comprehension(i.e., the act-out task). First, production is seento be influencedto a lesserdegreethan comprehensionby extralinguistic knowledge. The availability of extralinguistic cues makes it possible for people to comprehendsentencesaccurately without actually constructing a full structural representationof the input sentence,according to Flynn (1986b): In comprehension, the lack of a fully developed structural competenceis not as critical as in production. Other extralinguistic knowledge and information available to the subject can be used to make a coherent interpretation of the stimulus

sentence . (p . 154 )

[W ]e can hypothesize that comprehension can be achieved with a minimal structural analysis. The role of pragmatic context and extralinguistic knowledge is more likely to be helpful here than in production. A subject could easily achieveaccurateperformanceon a comprehensiontask. (p. 138) There is a second difference, according to Flynn: production (i.e., the imitation task) directly taps structural knowledge without requiring a full semantic interpretation :

[I]mitation does not necessarilyrequire establishingthe referents or generating a complete representationof sentencemeaning. (p. 138) Given the requirementsof an elicited imitation task, we might expect the results of such a test to demonstrate that we are tapping . . . structural knowledge. (p . 140)

[W]e can concludethat while production (elicited imitation) and comprehension (act-out) both elicit data that can be evaluated for evidence of linguistic competence, the degree to which each accessesthis knowledge differs significantly. SpecificallyI . . . elicited imitation more directly evaluatesa leamer's structural knowledge. (p. 154) In sum, on Flynn's view there are fewer factors competing with structural knowledge in the imitation task; the factors that compete with structural knowledge in the actout task often result in appropriate linguistic behavior, but right answersarisefor the wrong

reasons in that task .

With thesepoints in mind, it will pay to look more closely at the findings that are said to support them . The imitation and act-out experiments in question were de-

signed to test for the relative easeof processingright -branching (RB) sentenceslike (2) and (4), as comparedto left-branching (LB) sentenceslike (3) and (5). Imitation (2 )

task

The man answered the boss when he installed

the television . (RB )

(3) When he delivered the message , the actor questionedthe lawyer. (LB)

74

Chapter 9 Act -out task

(4) The yellow squaretouched the red triangle when it turned around. (RB) (5) When it turned over, the blue triangle touched the red circle. (LB) In Flynn's study, the "directionality" factor was significant in the analysis of the results of the imitation task, but not for the act-out task; right -branching constructions evoked fewer errors in the imitation

task . However , the act -out task resulted in

fewer errors overall than the imitation task, although the difference was not statistically significant. The fact that subjects performed better on the act-out task was interpreted by Flynn as evidence that "even without a fully developed syntactic competence, subjectscan productively rely on and use pragmatic context and knowledge in comprehensionto an extent not helpful in production" (p. 149). In each task , certain responses were counted

as correct and others were counted

as

incorrect . In the act-out task, a response was considered correct if either the subject or the object of the main clause was selected as the referent of the pronoun . Because the pronoun is " free" in sentences like (3) and (5), it can corefer with either NP . Among the responses that were considered incorrect in the act-out task were (a) acting out the two clauses of a sentence in the wrong order , and (b) acting out the content of only one of its clauses. There was another manipulation within the act-out task: the presence or absence of a "pragmatic lead." The pragmatic lead drew attention to the direct object NP in

the test sentencethat followed. For example, (6) illustrates the way sentence(5) was introduced in the pragmatic lead condition . (6)

I am going to tell you a sentence about the red circle .

When it turned overI the blue triangle touched the red circle. With a pragmatic lead, there was a significant reduction in responsesin which the pronoun and the subject of the main clause had the same referent . This makes sense, because the pragmatic lead mentioned the referent of the object NP in the main clause of the sentence

that

followed

.

In the imitation task, a sentencehad to be repeated verbatim by subjects to be counted as correct. Among the incorrect responsesWere"anaphoraerrors." Anaphora errors were found in previous studies of first language acquisition (e.g., Lust 1981). Anaphora errors are responsesin which "a subject might reverse the order of the direction of anaphora" (p. 145), as when sentence(7) is converted to sentence(8). (7)

When he was at school, John rode the bike .

(8)

When John was at school , he rode the bike .

In Lust's study, anaphora errors occurred in 44% of the sentencesproduced by the younger children (2;6- 3;5) and in 28% of the sentencesproduced by the older children

(3 ;6 - 5 ;7).

We are now in position to evaluate the conclusions drawn by Flynn . Let us look

first at the claim that the imitation task directly taps structural knowledge because "imitation does not necessarily require establishing the referents or generating a complete representation of sentence meaning " (Flynn 1986b, 138). As noted by Las-

nik and Crain (1985), the fact that children make anaphoraerrors in the imitation task entails that they do establish the referents of the NPs in the test sentences. An ana-

phora error like (8) is a paraphraseof (7); the two sentencesexpressthe samemean-

Imitation

75

ing using different word order. In light of the fact that anaphoraerrors rearrangeNPs in a way that preserves the gist of the original sentence, we would conclude that such errors occur only if the referents of the NPs in the original sentencehad been established .

Anaphora errors indicate that children understood the meaning of sentenceslike (7) as involving backward anaphora. On the basis of the anaphora errors, some researchershave concludedthat children experiencedifficulty with backward anaphora (e.g., Solan 1983). The backward anaphorainterpretation cannot be too difficult for children to construct, however, becausesentenceslike (7) need not involve anaphora at all. An alternative would be to interpret the pronoun as referring IIdirectly," to someonewho is not mentioned in the sentence. The high proportion of anaphora errors shows that the backward anaphorainterpretation is frequently favored over the IIdeictic " (or IIdirect reference" ) interpretation ; if children had consistently

adopted the deictic interpretation, they would have avoided anaphora errors and would have repeatedthe test sentencescorrectly. The availability of the backward anaphorareading in the imitation task is interesting , because the deictic interpretation does appear to be favored by children in the act -out task . For examole . two -thirds

of the children

studied

bv Tavakolian

(1978 )

acted out sentenceslike (7), in which a pronoun precededa lexical NP, by interpreting the pronoun as referring to a figure present in the experimental workspace, but not mentioned

in the sentence .

Why , then, is the deictic interpretation of the pronoun suppressedin the imitation task? In our view , this is the consequence of presenting sentences outside of context ,

that is, in the so-called null context. In the act-out task, (toy ) figures are presentin the experimentalworkspace. Thesefigures establishpotential referentsfor the NPs of the test sentences .2 In the imitation task, by contrast, there are no potential referents other than those denoted by the lexical NPs contained in t,he test sentences . Therefore , for a pronoun

to refer to someone other than the referent of an NP mentioned

in

the sentence, its referent would have to be conjured up by the hearer. In chapter 12, we provide evidence that even adults avoid interpretations of ambiguous sentences that refer to entities

that have not been introduced

into the conversational

context .

The finding that children and adults adopt similar strategiesfor interpreting ambiguous sentencesis expectedon the Modularity Matching Model . With respectto the construction of a complete representationof sentencemeaning, we would

draw

the same conclusion

as Potter

and Lombardi

( 1990 ), who

inves -

tigated verbatim recall of sentencesby both children and adults: [I]mmediaterecall involves regeneration of the sentencefrom a conceptualrepresentation[,] . .. a representationthat is based on the deepest, messagelevel. [Immediate] Memory is (nearly) verbatim not becauseof a special short-term representation of the surface sequence, but because the regeneration process of recall makes use of recently activated . . . entries in the lexicon to express

the ideasin the sentence,using the normal mechanismsof sentenceproduction. (p . 650 )

Further evidencefor this conclusionis provided by Martin (1993), who reports the findings of a battery of linguistic tests administeredto a patient, E.A ., who has an auditory memory spanof two, as assessedby serial recall of auditorily presentedwords. Despite a severe limitation in immediate memory , E.A . was able to repeat the gist of

76

Chapter 9

two-clausesentencesin an imitation task. Here are two examples(from Martin 1993, 180 ):

(9)

After eating dinner, the man walked the dog. E.A . IIAfter supper, the man took his dog for a walk."

(10)

Before calling her mother, the girl had a cup of tea. E.A . "The girl drank somehot tea before she went to talk to her mother ."

Notice that in repeating (10), E.A . converted a left-branching construction to a right branching one. This is interesting, becauseit is the kind of IIerror" based on directionality that children were found to make in the studiesby Lust (1981). The fact that E.A . correctly paraphrased(10), however, shows that she processedthe syntactic and semantic structure of (10). This counts against Flynn 's (1986b) claim that the imi -

tation task directly taps syntactic processing, but does not require the construction of sentence meaning . Martin notes :

Most of E.A .'s attempts at repetition were paraphrasesthat either preservedthe meaning of the original sentence(50% of the sentences ) or were only slightly different in meaning (10% of the sentences). (p. 180)

Along with Martin (1993) and Potter and Lombardi (1990), we would conclude, therefore , that the imitation task requires the formation of both a syntactic and a

semantic representation and includes establishing the referents of the NPs in the target sentences , Flynn's claim notwithstanding. We also question Flynn's second conclusion: that extralinguistic knowledge artificially inflates the level of correct performancein the act-out task. It seemsunlikely that people have formed opinions about the plausibility of the events mentioned in the sentencesused in Flynn's study, where sentenceswere about circles touching squares , triangles turning over, and so forth. Therefore in the interpretation of such sentences , extralinguistic knowledge is unlikely to increasethe proportion of correct responses

.

This leavesus to consider the possible contribution of the pragmatic lead to subjects' responses. Performance did not improve in the pragmatic lead condition ; in fact,

it plummeted. There were 65% correct responsesin the absenceof a pragmatic lead, but only 49% correct responseswhen a pragmatic lead was supplied. This shows that subj ects were not aided by the pragmatic leads; they were confused by them for

some reason. The overall decreasein comprehensionwith pragmatic leads is not inconsistent

with

the observed

decrease in coreference

between

the pronoun

and the

subject NP in the pragmatic lead condition; presumably, this reduction occurred on trials where subjects were not confused.

Nor is the drop in correct responsesin the pragmatic lead condition inconsistent with the observation

that there were more errors overall in the imitation

task than in

the act-out task. Our accountof the (nonsignificant) differencein performanceon the two tasksis quite simple: more kinds of responseswere counted as errors in the imitation task than in the act-out task. Superior performancein the act-out task is interpreted by Flynn as evidence that 'Jsubjectscan productively rely on and use pragmatic context and knowledge in comprehensionto an extent not helpful in production " (p . 151). As noted , the pragmatic lead sentences were actually detrimental to

77

9.4 Conclusion

Where do things stand? As we see it, the imitation task and the act-out task both induce artificially high error rates, in large part becausethey present sentencesoutside of context. If this is correct, then it lets the Modularity Matching Model off the hook. The question remains whether tasks can be devised that produce results in keeping with the Modularity Matching Model, or whether, as the Competing Factors Model maintains, all tasks are influenced to a greater or lesser degree by a host of factors in addition to grammatical competence. Adopting the Competing Factors Model, Flynn (1986b) infers that the factors that impede the imitation task and the act-out task will also impede performanceon other tasks: [O]ther commonly used tasks, suchas grammaticality judgment tests, .. . yes/no answersto comprehensionquestions, are all mediatedby a set of extralinguistic factors that could significantly attenuate accessto a leamer's structural knowledge of a language or produce results seriously confounded by the task requirements themselves . (p. 155)

Not necessarily, according to the Modularity Matching Model . If there are tasksthat do not impose extraneousdemandson languageprocessing, performanceshould be nearly perfect. Later, we will describetwo suchtasks: the elicited production task and the truth value judgment task.

Chapter 10 Judgment Tasks andCompeting Factors

This chapter compares the predictions of the Competing Factors Model and the Modularity Matching Model as they pertain to experimental studies of child language using IIdiscrete" measures , such as yes/no judgments. The researchdesignswe discussin this chapter investigate children's knowledge of linguistic principles that prohibit the assignmentof certain meanings to sentences - that is, with (sentence, *meaning ) pairs. Therefore , these research designs attempt to assess the range of meanings that subjects can assign to the test sentences.

In chapter 7, we noted a problem in interpreting the findings from nondiscrete (== continuous) measuressuchas reaction time, when the issueis whether or not sentences are ambiguous for children. In responding to ambiguous sentences , children (like adults) might exhibit a preferencefor one interpretation over its competitors. It could turn out, however, that children favor the interpretation that correspondsto the only interpretation that adults assignto a sentence.That is, the sentencecould be unambiguousfor adults and ambiguousfor children, and yet children could favor the adult interoretation and therefore respond to it in the same way as adults do . This scenario limits the utility of quantitative measuresin assessingchildren's linguistic knowledge. Weare therefore led to seekmeasuresthat ascertainwhen a meaning is authorized or proscribed by children's grammars, not preferred or dispreferred. Becauselinguistic constraintsproscribe meanings, reaction time measuresare of limited application in the study of children's knowledge of constraints (unless they are augmented by measuresthat directly reflect the linguistic analysesthat children assignto sentences ) .

At first glance, it might seem that judgment tasks do not suffer from the same limitations. In this chapter, we will show that there is more to the matter than meets the eye. The samekinds of problems that confront nondiscretemeasuresof behavior arise with judgment tasks, depending on the level of successfulperformancethat is taken as evidence of linguistic knowledge. According to the Modularity Matching Model, if children can parsea sentenceand relate it appropriately to the context, then they will know (" cognize " in the sense of Chomsky 1981) whether or not the repre-

sentation they assign to the sentenceis grammaticalor violates a principle of grammar. If children understand the experimental task and are cooperative , they will

perform perfectly, or nearly so. Becausethe Competing Factors Model acceptsless than perfect performanceas evidence of linguistic knowledge, it is difficult on this model to ascertainwhen children find sentencesambiguousand when they find them unambiguous. But, as we have shown, this is crucial in deciding whether or not constraints are operative in children 's grammars .

Putting aside the issue of how data should be interpreted, we note that two basic researchdesigns have been implemented in the use of judgment tasks. In one

80

Chapter 10

procedure

,

one

the

same

of

pretation

.

context

,

In

test

whether

their

.

.

Although

Matching

in

defer

of

of

1

Manipulating

This

section

against

hibited

that

a

is

The

permitted

nouns

.

and

her

,

is

(

1

)

a

b

Each

trial

which

.

of

In

the

self

Meaningz

( deictic

)

not

for

child

was

cally

linked

.

If

)

a

coreference

Big

)

I

video

sentence

pronoun

saw

3

)

Big

a

by

(

ordinary

Principle

the

-

.

We

on

the

Competing

sentences

that

the

is

sentence

1990

)

.

pro

-

mean

This

-

study

B

(

see

,

chapters

in

(

1

)

was

ordinary

pronouns

illustrated

'

a

s

31

-

pro

such

33

as

)

.

-

him

One

re

were

Bird

then

:

!

a

commentary

.

Big

:

!

!

!

k

patted

him

incorrectly

this

video

.

,

!

.

short

video

clip

by

were

is

him

given

two

in

( 2

in

the

ex

trials

)

also

presented

doing

.

Big

Bird

on

the

was

hitting

these

trials

pronoun

to

interpretation

as

him

,

be

then

-

the

anaphori

meaning

!

-

,

or

the

.

something

( 3

)

with

showed

Big

correct

referring

deictic

to

interpretation

Ernie

Ernie

Bird

description

.

will

the

call

pronoun

Big

Bird

patting

of

We

of

.

patted

Ernie

the

video

this

interpretation

.

-

for

.

but

allowing

illicit

a

There

trial

hit

!

controlled

commentary

,

to

of

one

!

.

puppet

Ernie

' s

hi

patted

Bird

near

puppet

refer

:

~

the

for

Ernie

the

! !

hand

in

standing

will

to

the

-

.

consisted

;

happened

' s

:

experiment

scene

was

We

)

)

with

grammatically

as

gives

focus

-

mod

.

(

corresponding

was

we

containing

of

is

Rosen

sentence

.

Rosen

sentences

interpretation

accepted

the

like

( 3

findings

meaning

to

interpretation

Bird

Bird

interpretation

Sentences

of

the

,matching

sentence

and

interpretation

puppet

child

Big

,

of

corresponding

Grimshaw

pronouns

standing

,

other

the

of

had

The

interpreting

to

by

the

thought

was

( 2

Ernie

,

out

it

Bird

here

Modu

,

from

strategy

understanding

,

and

.

Big

the

.

coreference

acting

sentence

;

same

about

.

(

what

video

,

him

Meaningl

said

( 2

s

of

were

target

is

'

9

Grimshaw

perimenter

the

*

the

constrained

puppets

each

on

chapter

patted

.

the

of

stemming

research

grammar

interpretation

Ernie

in

Model

III

to

underlying

advocates

part

same

informative

interpretation

to

to

and

adult

children

in

the

the

strategies

the

,

syntactically

on

by

be

Factors

features

corresponding

focus

test

noted

as

design

utilize

one

the

we

As

striction

meaning

by

to

both

can

Competing

well

research

that

constraint

experiment

used

their

respond

,

the

.

consider

:

linguistic

designed

in

ones

contexts

by

also

The

but

different

as

subjects

same

respond

in

constant

the

with

-

the

Context

to

,

design

findings

continues

Model

are

of

differences

the

actors

ing

the

research

that

inter

subjects

differences

held

favors

in

associated

that

observation

context

advocates

of

be

context

alternative

presented

observation

about

designs

by

features

discussion

.

and

certain

interpretation

the

the

similar

Model

differ

with

can

the

one

an

are

sentences

,

informative

,

,

;

favors

sentences

the

be

contexts

context

different

not

Similarly

different

other

constant

can

sentences

grammars

els

or

differently

the

two

held

knowledge

larity

,

context

different

in

and

procedure

the

sentences

to

presented

sentence

other

With

grammatical

way

10

is

the

determine

.

the

the

to

meaning

F

sentence

interpretation

him

.

if

the

Therefore

.

,

child

the

test

interpreted

meaning2

;

this

JudgmentTasksand Competing Factors

81

The finding was that children said IIYes" to (3) in the second situation a higher proportion

of the time than they said IIYes" to (2) in the first situation . To put it

another way, children acceptedthe deictic interpretation (meaning2) when it was a correct description of a situation more often than they acceptedthe coreferenceinterpretation (meaning! ) when it was a correct description of a situation. On the basis of these findings, Grimshaw and Rosen conclude that children know the linguistic constraint governing the assignment of reference to pronouns . They contend that the IIdifference " in children ' s correct responses to similar sentences in different

situations

is sufficient to guarantee that children know the constraint .

Here is their argument. Supposethat children lack the grammatical principle instructing them that the test sentence is ungrammatical on meaning! . If so, then meaning} and meaning2are both possibleinterpretations for children. On the basisof the proportion of children 's IIYes" responses to the sentence in a situation that makes it true on meaning2' a "best guess" estimate can be derived about the proportion of

"Yes" responsesthat are expected in a situation that makes the sentencetrue on meaning}. If children go out of their way to reject the sentencesignificantly more often on meaning} than is expected according to the "best guess" estimate, then something children know must be responsiblefor the different pattern of behavior in the two cases.The most likely candidateis the principle governing the interpretation of pronouns, which prohibits meaning}. In Grimshaw and Rosen's study, children acceptedthe test sentences83% of the time on meaning2' the meaning that is consistentwith the adult grammar. According to Grimshaw and Rosen, children who lacked the grammaticalprinciple that prohibits meaning! should have accepted meaning} the same proportion of the time (i.e., roughly 83% of the time ). This did not happen, however . Children gave "Yes" responses to meaning } in the relevant situation only 42% of the time . The difference between

83 % and 42 %, Grimshaw

and Rosen conclude , is due to children ' s knowl -

edge of the grammaticalprinciple prohibiting meaning}. We disagreewith this conclusion. Its validity hinges on two assumptions: that the Competing Factors Model is basically correct and that children can recover either meaning of an ambiguous sentencewith equal ease. According to Grimshaw and Rosen, the assignmentof either meaning! or meaning2simply dependson which interpretation makesthe sentencetrue in the experimental context. This is what made their study seem to have more promise than the preferential looking paradigmchildren were actually rejecting the illicit interpretation some prop .ortion of the time .

Let us walk though the situation a bit more slowly. Weare attempting to evaluate the hypothesisthat children lack the constraint, that is, Principle B. If this hypothesisis correct, then the secondsentencein (2)- namely, (4)- is ambiguousfor children. (4)

Big Bird hit him .

Suppose, now, that this hypothesis is correct, namely that (4) is ambiguousfor children. But supposefurther that children do not have equal, accessto the alternative interpretations of an ambiguous sentence, as Grimshaw and Rosen suppose. That is, children might have a preference for one interpretation over the other. If so, then at least one other factor

contributes

to children 's responses , in addition

to the bias to

accessthe interpretation that makesan ambiguoussentencetrue in the context. This additional factor is the preferenceto resolve the ambiguity in one way rather than in other

ways .

82

Chapter 10

With

two

one

of

( or

the

parsing

to

coreference

the

parsing

absence

suffice

,

of

we

are

reading

is

also

a

are

wrench

,

sponses

to

instances

.2

in

children

' s

children

ence

the

ably

tial

,

the

the

,

of

up

pretations

by

of

the

an

ambiguous

conversational

interpreting

ambiguous .

this

does

inter

-

reading

the

fact

.

that

If

,

this

availability

the

of

sentences

and

in

adults

,

~n

vantage

' s

re

who

of

confuses

knowledge

11

-

many

point

analysis

their

chapter

its

of

others

the

there

Modularity

as

Rosen

if

over

the

the

From

:

pattern

to

over

detail

clear

same

preferences

) .

and

meaning

the

and

more

is

one

according

12

- Pasek

interpretation

problem

same

chapter

Hirsh

correct

of

sentence

in

,

the

of

.

In

;

in

we

that

.

The

is

therefore

its

to

-

a

,

.

the

act

.

to

has

to

have

adopt

the

be

not

-

similar

Modularity

-

-

The

than

mentally

con

avoid

been

- out

. l

other

adults

-

poten

someone

-

inter

-

introduced

strategies

.,

To

Presum

establishes

that

. g

pre

context

workspace

refer

on

( e

of

task

experimental

that

tasks

pronouns

task

this

observation

null

different

- out

children

expected

the

-

on

consequence

of

act

corefer

out

certain

- called

referent

entities

of

that

is

in

evidence

that

.

evidence

the

jury

in

so

to

,

the

the

the

the

provide

refer

finding

,

account

simply

the

pronoun

sentence

,

far

from

in

a

the

12

sentences

sentences

for

So

interpretation

because

as

favor

preferred

in

findings

deictic

sentences

chapter

context

to

differ

not

mentioned

.

is

preferences

taken

demonstrated

' s

be

,

parsing

be

nevertheless

Rosen

context

the

test

)

.

may

of

' s

can

independently

( 4

preference

favor

does

be

like

and

this

tasks

of

can

pronouns

point

to

NPs

NP

,

would

two

it

children

study

sentences

outside

contrast

child

if

' s

sentences

Grimshaw

view

we

on

Rosen

ambiguous

for

our

hinges

only

to

in

the

the

the

The

,

Grimshaw

and

of

In

,

by

,

deal

B

appear

for

referent

great

children

from

task

jured

into

) .

children

referents

imitation

a

help

to

results

on

with

produce

Moreover

ambiguous

pronouns

conjecture

where

then

this

interpretation

task

this

,

it

deictic

the

the

favor

to

in

,

Principle

little

sentences

support

task

is

imitation

senting

discussed

incorrectly

coreference

the

.

in

ambiguous

Grimshaw

of

respond

There

as

argument

exactly

an

explain

from

of

.

that

will

data

interpretation

question

a

despite

Principles

knowledge

who

,

that

the

then

makes

not

noted

boost

for

expected

exhibit

interpreting

We

that

the

' s

be

of

are

apparent

,

be

predicament

.

Model

.

be

-

with

than

true

preference

Rosen

not

should

versus

example

the

context

assign

,

the

outweighs

that

should

should

that

consistency

often

preferred

ambiguities

reading

preferences

should

it

"

)

,

interpretation

sentence

" Yes

same

interpretations

preferences

Preferences

For

and

children

principles

say

( 4

greater

children

the

parsing

resolving

Matching

' s

to

a

should

one

( these

children

It

Grimshaw

favor

Modularity

linguistic

Model

,

to

,

that

make

the

of

alternative

Model

known

into

children

these

Matching

back

for

then

Finally

that

example

conversational

more

is

bias

possibility

preferences

competitors

the

the

into

inherent

are

;

an

possibility

for

in

the

access

interpretation

the

in

correctly

that

,

pronoun

true

establish

contexts

the

to

.

to

the

possible

.

drawn

study

bias

)

consider

is

show

grammars

with

It

responses

respond

deictic

higher

been

' s

the

( 4

' s

the

their

must

.

makes

scenario

consistent

have

Golinkoff

in

in

,

we

of

children

will

extent

even

throws

B

larger

If

with

this

,

interpretation

children

out

however

competitors

that

than

supposing

reading

We

deictic

Principle

a

,

its

.

the

rule

to

play

interpretation

then

to

pretation

this

the

preference

true

the

as

the

at

outweigh

interpretation

sentence

not

factors

for

accept

the

in

)

could

preference

bias

10

more

factors

-

Matching

for

JudgmentTasksand Competing Factors

83

The final difficulty with Grimshaw and Rosen's study is the low success rates by children who , they conclude , have mastered the linguistic knowledge being in vestigated . Such low levels of accuracy are accept~d by researchers who adopt the Competing Factors Model , because they view children 's linguistic behavior as an aggregate of linguistic and nonlinguistic factors , their expression of linguistic knowl edge sometimes being suppressed by extraneous factors . Because the Modularity Matching Model assumes that children and adults share the same cognitive mechanisms, the observation that adults correctly comprehend sentences with ordinary pronouns , but children do not , is inconsistent with the model . Children 's poor performance with sentences governed by Principle B (i.e., roughly 50% acceptance of Principle B violations ) does not sit well with the Modularity Matching Model . Even if children 's failures do not stem from a lack of grammatical competence, but are due to limitations in processing capacity or the lack of certain principles of pragmatics , as several researchers have claimed (Grodzins ~y and Reinhart 1993; Chien and Wexler 1990), the differences are not anticipated by the Modu larity Matching Model (but see Thornton and Wexler , forthcoming , for an account of children 's errors that is consistent with the model ). Therefore , the model would have to be amended in appropriate respects, depending on how the controversy is resolved . However , amending the model is preferable to abandoning it altogether , in our view . As we said, maximizing the similarities between children s cognitive systems and those of adults is responsive to the general problem of language leamability - why all children successfully converge on an adult linguistic system despite the considerable latitude in their linguistic experience . To the extent that the cognitive mechanisms of children and adults are similar , leamability problems are cir cumvented . This is why we assume equivalence as the Null Hypothesis . By contrast , the Competing Factors Model abandons the Null Hypothesis from the start . On this model , children differ from adults in language processing . This adds unwanted degrees of freedom , by tolerating a wide range of processing explanations of differences between children and adults . Moreover , it leaves a new question 'to be addressed for each such account : how does the processing system of children change, so as to converge on the adult system?

10.3 Implicationsfor Research Design In experimentswith children, the contexts that researchersconstruct are made to be consistent with more than one interpretation of the test sentences . One interpretation that is suited to the context is the adult interpretation. As the graphic below indicates, this will be referred to as meaning2 ' Presumably, meaning2is also availableto children. If children lack the linguistic principle being investigated, however, they will alsobe able to assignthe interpretation that is prohibited by the principle: meaningl' That is, children will have accessto an interpretation of test sentences that is not available to adults. Experimental contexts are devised to see whether or not children accessthe nonadult interpretation, meaning! . In designing this hypothetical experiment, let us supposethat the experimenter decided to make the test sentencesfalse in the context if meaning2 was assigned to them; and let us suppose that the contexts were also constructed such that the sentenceswere true if meaning! was assignedto them. In chapter27, we explain when this particular design is appropriate.

84

Chapter 10

Child (sentence,meaning}) (sentence,meaning2)

Adult (sentence,* meaning}) (sentence,meaning2)

Ruled out

If children consistently rejected the test sentencesin these (ambiguous) contexts, their behavior would be compelling evidence that the only interpretation available to them is the adult interpretation, meaningz' the interpretation of the sentencethat makesit false in the context. The conclusion would be that children, like adults, find the test sentenceunambiguous, by virtue of knowing the linguistic principle under investigation. Suppose, however, that the outcome is less conclusive, with children responding "No" to the test sentencesonly 60% of the time. Input (sentence, context) meaning}, true meaningz, false

Response fIYes" 40% fiN0" 60%

The Competing FactorsModel maintainsthat sucha level of performanceis sufficient to infer knowledge of the linguistic principle. All that is required to demonstrate children's linguistic competence is above-chance performance in response to sentencestesting the linguistic principle; the 40% "errors" may be attributed to "performancefactors." We take this to be the viewpoint advancedby Grimshaw and Rosen (1990), for example. There is another possibility to consider, however. On this scenario, children lack the linguistic principle. In this case, the test sentenceswill be ambiguous for them: both meaning! and meaning2will be accessible . But now supposethat children have a strong preferencefor the interpretation that is permissiblein the adult grammar (i.e., meaning2). This preferencecould be quite strong, say, by a factor of 3:1. If there were no other influences on a particular child's decision, then, the child would access meaning2 75% of the time, and meaning! 25% of the time. Becausethe sentenceis false on meaning2in the experimental context, the result would be 75% "No" responsesby children. However, children's preferencefor one interpretation of the sentence over another is not the only factor that influencestheir decisions. In addition, they exhibit a tendency to select a reading that makesthe speaker's sentencetrue in the context. In the present experimentalsetup, this bias favors meaning!, the reading that is true in the context (but ungrammaticalfor adults). It seemsreasonableto conjecture that this bias could boost the availability of meaning! by 15%, such that meaning! is adopted 40% of the time, rather than 25% of the time- the result that would have been obtained in its absence . There will be a correspondingdrop in "No" responses , from 75% to 60%.2 In short, on both of the scenarioswe have considered, children give "No" responses60% of the time. The reasonsunderlying this low rate of rejection are different on the two scenarios, however. In one case, children know the principle under investigation but do not display this knowledge perfectly; their low level of correct performanceis attributed to competing factors. In the other case, children lack the principle, and their low level of performanceis attributed to parsing preferencesfor one interpretation of an ambiguous sentence. Since both accountsexplain the findings, it would clearly be unwarrantedto infer from children's lessthan total rejections that they commandthe linguistic principle under investigation. This discussionpoints out the main problem in adopting the methodological stance of the Competing Factors Model. Becausethe alternative interpretations of ambig-

JudgmentTasksand Competing Factors

85

uous sentencesare not necessarilyequally easyto access , children's occasionalor even

substantial rejectionof sentences on aninterpretationthat is prohibitedby a constraint is only circumstantialevidencethat the constraint is part of their grammars. The pos-

sibility exists' that the sentences are ambiguousfor children, but that they prefer the interpretation associated with the IN 0" response (i .e., the adult interpretation ).3

Therefore, if we adopt the methodological stanceof the Competing Factors Model, we may confuse behavior that reveals preferencesamong alternative interpretations with behavior that revealsknowledge of linguistic principles.4 To avoid this confound of parsing preferencesand grammatical knowledge, the Modularity Matching Model advocatesa different researchstrategy: an experimenter can infer that children know a linguistic principle only if they perform with near total accuracy, as adults do. Fortunately, it follows from the basic architectureof the Mod -

ularity MatchingModel that it is reasonable to expectnearperfectperformance by children, as well as by adults. Both children and adults have grammatical modules, processing modules, and perhaps other modules. When a child or an adult is confronted with sentencesthat are ambiguous, the processingand "other" modules can createpreferencesfor one reading or another. But neither children nor adults will assign an incorrect analysisto an unambiguoussentence; as long as the correct analysis is felicitous in the context , the pressures to interpret the sentence by means other

than grammatical knowledge will be resisted. The reason is that the grammatical components preempt these other factors ; they do not compete with them .

One might object that, even on the Modularity Matching Model, there is also no way in principleto distinguish principles from preferences.Pursuing the sameline of reasoningthat we brought to bear against the Competing FactorsModel, one might counter that cnildren could find a test sentence ambiguous , but favor the inter -

pretation associatedwith the "No" responseto such an extent that they display the same pattern of behavior as adults do (i ~e., near total rejection of the test sentences),

even though they lack the linguistic knowledge that underlies adult behavior. Similarly, the interpretation of the findings of production studies might be called into question on the grounds that children's consistentproduction of sentencesconforming to one structural analysis does not entail that other structural analyses are not compatible with their grammars- the alternative analysesmay simply be less preferred. Although such a state of affairs may seemunlikely (in light of the inherent bias to analyze ambiguous sentencesin a way that makesthe speaker's statementtrue), we take this objection seriously, becauseit points to a potential limitation on the effectiveness of both of the methodologies we will introduce. We will therefore mention experimental maneuversthat can be instigated to overcome this potential confound. For the moment, let it suffice to underscorethe importance of distinguishing princi-

plesfrom preferences in the designof experimentalstudiesof child language . In our view, the distinction between principles and preferenceslooms larger than any other in the construction of appropriate experimentalmethodology. 10.4 Conclusion

In this chapter, we have argued against the researchstrategy of the Competing Factors Model . According to this model, all that is required to demonstrateknowledge of a linguistic principle is conformity to the principle at a level significantly above

86

Chapter 10

chance. From a theoretical standpoint, the problem with the Competing Factors Model is that it introduces additional degreesof freedom beyond those of the Mod ularity Matching Model. The Competing Factors Model views language as a large set of "tendencies," so that knowle~ge of language amounts to certain statistical tendenciesin behavior. Therefore, showing that the statistical behavior of children is different for different constructions shows that thev of - have learned adult caoacities .. grammar. The Competing Factors Model claims that grammar is just one linguistic ability that competes with other linguistic abilities to yield a statistical portrait of behavior. In the next chapter, we review three more studies, the findings of which appearto support the Competing FactorsModel, in order to draw out the differences between this model and the Modularity Matching Model . Reasonsfor the alternative claims of the Modularity 'Matching Model are presentedin chapters 12 and 13.

Chapter 11 Context and Competin 19 Factors

In the use of judgmenttasks, two basicresearchdesignshave beenimplemented within the CompetingFactorsframework . One procedurewas discussed in the previous chapter. The procedureinvolved presenting the samesentencein different contexts: one context favoring one interpretation of the sentenceand the other context favoring an alternative interpretation. In this chapter, we discussanother researchde-

sign. In studiesadoptingthis design, differentsentences are presentedin the same context , to determine

whether

or not the sentences can be associated with

the same

meaning. With context held constant, the observation t~at children respond differently to different sentences serves as a basis for inferences about children 's under -

lying grammaticalknowledge (similarly, if they respond in the sameway to different sentences , with context held constant). Resultsfrom studiesusing such a design have led some researchers to reach conclusions that are inconsistent with the Modularity

Matching Model . Therefore, it is necessaryto evaluatethese studies, if we are to retain the model. As always, we will examine actual studies, beginning with an investigation of Principle C of the binding theory, by Lust, Eisele, and Mazuka (1992). 11.1 PrincipleC Principle C prohibits anaphoricrelations between pronouns and names. Consider examples(1)---(3),' adaptedfrom the study by Lust, Eisele, and Mazuka (1992). As before, the intended interpretation of the sentenceis indicated by underlining; if two NPs are underlined, then they are taken to have the samereference.

(1) JI .& atethe applewhen!!!~_~!E~ touchedthe pillow. (2)

When~ atethe apple~!~-~~~ touchedthe pillow.

(3)

!!!~~!E~ touchedthe pillow when~ atethe apple.

Principle C prohibits coreferencebetween the pronoun, he, and the name, Big Bird, in (1), but does not prohibit coreferencebetween these sameelementsin (2) or in (3). Principle C is a linguistic constraint; it prohibits coreferencein certain sentences . To assess children's knowledge , Lust, Eisele , andMazukapresented - of this constraint pictures to children and had them judge whether the sentenceswere, or were not, ac-

curatedescriptionsof the pictures. In one of the picturescorrespondingto (1)- (3), there were two characters , Big- Bird and Oscar the Grouch. Big Bird was holding a pillow and eating an apple. Oscar the Grouch was in a garbage can, just observing. Lust, Eisele, and Mazuka call this the coreferencepicture. Children who know that co-

referenceis blocked in (1), but not in (2) or (3), would be expected to accept (2) and (3) as accurate descriptions of the coreferencepicture. By contrast, children who

88

Chapter 11

know Principle C should indicate that (1) is not an accuratedescription of the picture: Principle C dictates that the pronoun, he, cannot refer to Big Bird; the only other potential referent is Oscar, but Oscar is not eating the apple. Sentences(2) and (3) performed another service in this study. Notice that in these sentences, the linear order of the pronoun and the name differs . Sentence (2) is an in -

stanceof backward anaphora, becausethe pronoun precedesthe name, whereas(3) is an instance of forward anaphora. Lust, Eisele, and Mazuka call this difference directionalify. By comparing the rates at which children accepted(2) and (3) in responseto the coreferencepictures, Lust, Eisele, and MazUkaattempted to measurethe effect of directionality. One of the findings of their study was a statistically significant effect of directionality: children were found to accept forward anaphorasentenceslike (3) more often than they accepted backward anaphora sentenceslike (2), in response to the coreferencepictures (83% and 57% acceptances , respectively). Of course, it is not "incorrect " for children to reject either (2) or (3), because the grammar permits a direct reference (deictic ) interpretation of the pronoun in both sentences. And on

the direct referenceinterpretation, these sentencesare not true descriptions of the co reference picture .

BecauseLust, Eisele, and Mazuka advocatethe Competing FactorsModel, they assumethat the directionality effect will influence children's responsesto unambiguous sentenceslike (1), as well as to ambiguoussentenceslike (2) and (3). Moreover, they assumethat directionality will influence children to a similar extent in interpreting sentenceslike (I ), as comparedto ones like (2). But, in contrast to sentenceslike (2), sentenceslike (1) are also subject to Principle C, the grammaticalconstraint that prohibits coreferencebetween the pronoun and the name. In short, the directionality factor and Principle C conspire to promote the direct reference interpretation of the pronoun in sentenceslike (1), whereas only the directionality factor is operative in sentences like (2 . ). .

Therefore , to test children's knowledgeof the constraint , Lust, Eisele , andMazuka proposed that children must reject sentenceslike (1) significantly more often than they reject sentenceslike (2) in response to coreferencepictures. This was indeed what happened. Children rejected sentenceslike (1) 68% of the time; they rejected sentenceslike (2) 43% of the time. The significant differencein the rates of rejection was attributed to knowledge of the grammaticalconstraint.l Children's less than perfect performance(i.e., the 32% erroneousacceptancerate for sentenceslike (1)) was blamed on "performance factors ." -

In an earlier study, Lust, Loveland, and Komet (1980) sought to determinewhether

or not children'Iconsult" pragmaticcontext in decidingupon the semanticinterpretation of sentenceslike (1) and (2). To answer this question, they comparedchildren's responses to such sentences with and without a pragmatic lead. The pragmatic

lead was presented before each test sentence , for example, J'I am going to tell you a story about Big Bird. He ate the apple when Big Bird touched the pillow ." As this example illustrates, the pragmatic lead sometimesinvites children to initially assign Big Bird as the referent of the pronoun in sentence (1); this turns out to be incorrect ,

however, once .the name is encountered. Therefore, during on-line sentenceparsing the pragmatic lead sometimesconflicted with the linguistic constraint under investigation . Lust, Loveland , and Komet found that IIchildren did allow the presence of a

pragmatic lead to modulate their coreferencejudgments to someextent . . ., increasing the percentageof coreferencejudgments" (Lust, Eisele, and Mazuka 1992, 338). The pragmatic lead resulted in a 35% reduction in correct responses for sentences like (1)

Context and Competing Factors

Ambiguous

89

Unambiguous Noise

Noise

.

Pragmatic Context

Pragmatic

C on t ex t

Directional ity

P .

. I C

rlnclp e

Directionality

Figure 11.1 Predictions of the Competing Factors Model for ambiguou~; and unambiguoussentences

and a 59% reduction in coreferencejudgments for sentenceslike (2). Again, the difference was significant. Interpreting the findings of Lust, Loveland, and Komet's study, Lust, Eisele, and Mazuka remark that ". . . the pragmatic context appearsto interact with the knowledge that distinguishes [these] sentenceson independent, presumably (grammatical) linguistic and structural grounds" (p. 338).2 The studies we have just reviewed are typical of one researchstrategy used by proponents of the Competing Factors Model. Therefore, it is important to compare the methodological assumptionsof thesestudieswith those of the Modularity Matching Model. According to the model assumedby Lust, Eisele, and Mazuka, children's responsesto ambiguoussentenceslike (2) and (3) dependon two factors: directionality and the pragmatic context. In responseto unambiguous sentenceslike (1), an additional factor interacts with directionality and the pragmatic context, namely Principle C. Added to these interacting factors is some level of "noise," contributed by uncontrolled factors. We illustrate these assumptionsin figure 11.1. As the figure indicates, the Competing FactorsModel simply adds linguistic knowledge as another factor to the equation for unambiguoussentences , as comparedto ambiguoussentences . The Modularity Matching Model is not " additive" in this way. On this model, factors like directionality, pragmatic context, and other factors we have discussed , sometimesplaya role in children's understandingof ambiguoussentences ; however, these factors do not playa role in the interpretation of unambiguoussentences , such as (1). Figure 11.2 illustrates the assumptionsof the Modularity Matching Model. Comparing the two figures, we seethat the differencebetween the two models can be pinned on the analysis of unambiguous sentenc,es. According to the Modularity Matching Model, unambiguous sentencesare n<:>t subject to "performancefactors," except for a small level of "noise" that is present in any experimental context. The reason for this is that the (partial) syntactic and semantic representationsthat are assignedto unambiguoussentencesare completed beforethese other factors enter the picture. Therefore, grammaticalprinciples alone dictate children's responses . The differences between the models have consequencesfor research design. In contrast to proponents of the Competing Factors Model, proponents of the Modularity Matching Model are not content with less than nearly perfect performanceby children on sentenceslike (1), .~ ate the apple when ~~ - ~!!~ touchedthe pillow.

90

Chapter 11 Ambiguous Noise Preference

Unambiguous Noise

for

One Meaning Bias

toward

Meaningthat is True

Principle C

in the

Context

Pragmatic Conte >
Directional ity

Figure11.2 Predictions of theModularityMatchingModelfor ambiguous andunambiguous sentences

Ambiguous sentenceslike (2), When~ ate theappleBig Bird touchedthepillow, are useful as controls becausethey establisha baselineleveJofpefjformancefor the particular interpretation that is not governed by the linguistic constraint in question. Suppose that children manifest a strong preferencefor an interpretation that is not governed by the constraint, when they are askedto respond to ambiguoussentences . If so, then the fact that they reject this interpretation for unambiguoussentenceslike (1) is compelling evidence of linguistic knowledge- in the present case, knowledge of Principle C. To put the point another wayI children's responsesto the ambiguouscontrols are used primarily to rule out an alternative explanation of their responsesto the unambiguous test sentences . As noted earlier, if a linguistic constraint is absent from children's grammars, sentencesgoverned by the constraint would be ambiguousfor them. Nevertheless, they might have a preference for one interpretation over the others. Moreover, their preferencemight be consistentwith the constraint under investigation; that is, they might prefer for the adult interpretation. If factors conspire in this way, children could produce responsesthat would be counted as correct, even though they lacked the constraint under investigation. In short, children would give the right answersfor the wrong reasons. Ambiguous control sentencesenablean experimenter to rule out this explanation of children's (seemingly) correct behavior. In the best of all possible worlds, children will prefer to interpret ambiguouscontrol sentenceslike (2) by assigning coreferencebetween the pronoun and the name, but they will not tolerate this interpretation of test sentenceslike (I ). If these are the findings, then it may be unnecessaryto perform statistical tests of the differencesin children's performancein the two conditions; "eyeball" statisticsmay sufficein drawing conclusionsabout children's knowledge of the constraint. Suppose, however, that children assign coreferencein the control condition less than half of the time. In this case, it may be necessaryto establishthat there is a statistically significant difference between children's responsesto the ambiguous control sentencesand the test sentences. According to the Competing Factors Model, a positive finding resolves the issue; the hypothesis that children lack the constraint can be rejected. On the Modu larity Matching Model, by contrast, there is an additional consideration, namely, the proportion of children's correct responsesto the test sentences . If children truly find test sentenceslike (1) to be unambiguous, then they should respond correctly

91

Context and Competing Factors to

them

this

nearly

Modularity

Eisele

, and

ciple

C , they

larity

Matching

C , make in Our

the

the

draw

Factors

Model

conclusions

does

about

brings

. The

cussed

in

control

us

not

children

impose

's

linguistic

questions

(5 )

to

in

indicate

by Who He

to

like the

critical

know

on

the

know

( 1 ) . We truth

of

to

Lust

,

Prin

-

Modu

-

Principle take

value

up

this

judgment

conducted

within

differences

the

between

that

.

the

and

by 29

puppet a

second he

has he

researchers

did

"

the

best

of

an

like was smile

that the

same

strategy

' s responses and

different

question puppet

research

McDaniel

. These

. On

the children

" appropriateness

either

said did

chapter

adopts

comparing

conducted

another

the

that

, while

was

produced

Q :

experiment

constant

by

A :

incumbent

research

the

results

assumed

assumed

of

of

expose

Model

is

are

features

examples to

. It

sentences

design

provide and

second

detail

provided (4 )

a

experiment

to

this , who

the

are

Questions

to

asked

than

children

the

accept

children

unambiguous

Matching

/ context

greater

better why

discuss

is

Modularity

group

question

we

cannot

. Because

perform

framework

in

therefore value

interpreting

here

C

meaning

tences

to

in

purpose

face

explain

III , where

Principle

answer

Competing to

Model

- to

errors

part

and

the

. The

order

at

expected Model

Factors

model

This

time in

' s study

are

Competing

11 . 2

the

Matching

Mazuka

32 %

matter .

of

.

The

task

%

requirement

knowledge

a

100

additional

McKee asked

answer

of

to ( 1993

given

by puppet

in

one

( 4 ) or both

cases

like

and

and

, the

in

)

children

occasions

holding

different

one

that

is an

sen

~

dis

-

adult

puppet who in

to asked

( 5 ) . The

.

?

.

Q: Who did he sayhasthe bestsmile? A: He did andhe did.

These questions and their corresponding answers followed a short story in which two charactersperformed a similar action. In the stories associatedwith (4) and (5), there were two characterswho each said that he, himself, had the best smile. The answer, He did and he did, is an appropriate responseto the first question in this context, but not to the second. Principle C dictates this, as we will explain in chapter29. To see if subjects know Principle C, McDaniel and McKee compared their judgments about whether the sameanswer is appropriate for the two different questions in (4) and (5). Both children and adults indicated that this answer was appropriate to the question in (4) significantly more often (according to a statistical test) than to the question in (5). McDaniel and McKee inferred that the findings demonstrated children's knowledge of ~he grammaticalprinciple under investigation. It turned out, however, that the absolutelevel of performanceby both children and adults was not very good. Both groups correctly rejectedthe answeras inappropriate to the question in (5) only about half the time, although, admittedly, neither group rejected the answer as inappropriate to (4) very often. On the basis of the greater acceptancerate to question/answer pairs like (4) than to ones like (5), McDaniel and McKee nevertheless concluded that both groups know the grammatical principle under investigation. Here is their argument: Overall, both the children and the adults clearly differentiated between the . . . cases.The extent to which the adult group madethis distinction shows how the

92

Chapter 11 grammatical principle .. . is expressedin this particular performancetask. The fact that the children made the distinction to the same extent as the adults ["whose grammarswe presumablyknow about" (p. 278)] shows, therefore, that their grammarsalso contain the principle. .. . (p. 287)

Although this argument and its conclusion are consistent with the Competing Factors Model, they are not consistent with the Modularity Matching Model. According to the Modularity Matching Model, both children and adults should systematically reject question/answer pairs like (5). On this model, it does not suffice for there to simply be a statistically significant differencein the rates of acceptancebetween different sentencetypes, even if lithe children's distinction between types . . . [is] .. . significant and as great as the distinction madeby adults" (p. 289). The findings of this study permit us to make another point. In discussingthe results of the cross-modal priming study by McKee, Nicol, and McDaniel (1993) in chapter 10, we showed that a group mean can obscureindividual differenceswithin groups. The samepoint can be made using the study by McDaniel and McKee. A closeinspection of the data from this study revealsthat the distributions of responses by children and adults to question/answer pairs like (5) do not conform to the expectationsof the Competing FactorsModel. This model predicts a unimodal distribution of responsesto constructions that are generatedby children's grammars. In the present study, however, the distribution of responsesby both children and adults was not unimodal, but more closely approximated the kind of bimodal pattern described earlier. One group acceptedthe answer "He did and he did" as appropriate to questions like (5), so these subjectsdid not distinguish questionslike those from ones like (4). A secondgroup of subj~cts were successfulat the task; their responsesto questions like (5) fell under a different mode than their responsesto oneslike (4). The differencein the proportions of correct responsesto (4) and (5) was due to the group of children and adults who were successfulat the task. Becausemany children and adults responded incorrectly to the test sentences , however. the findinQs '-' of McDaniel and McKee's study are inconsistentwith the Modularity Matching Model. The point here is that the pattern of responseswas also inconsistent with the Competing FactorsModel. Therefore, we must look elsewherefor an explanation of the pattern of responsesby both children and adults. We discussthe range of possibilities open to the Modularity Matching Model in chapter 15, and we discussour particular view of the results from McDaniel and McKee's study in chapter 29. 11.3 Universal Quantification

We now discussa third example of an experiment, within the Competing Factors framework, that adopts the researchstrategy of holding the meaning/context constant, in order to comparechildren's responses to different sentences . The topic . of this L experiment was children's interpretation of sentenceswith universal quantification. For at least 30 years, it hasbeen widely believed that even children as old as 4 or 5 often misunderstandsimple sentenceswith a universal quantifier, suchas (6) and (7). (6) Is every farmer feeding a donkey? (7) Is a farmer feeding every donkey? The basic finding has been replicated with many children and across several languages. If shown a picture like that in figure 11.3, which we will call the extra object

94

Chapter 11 Transitive

Adult

43 %

Response

Intransitive

43 %

Adult

Response

gym metrical Response gym metrical Response

Figure 11.4

Proportion of correct and incorrect responsesfor transitive and intransitive sentences

where every cat is waving , and every cat is in a box , but where there is an IIextra " box with no cat in it . A symmetrical response would be to reject (8) because there is a box without

(8)

a cat in it .

Is every cat waving ?

Questions like (8) do not fall within the purview of the symmetrical interpretation ,

because(8) has only one noun for the universal quantifier to bind. Therefore, the Symmetrical Account anticipates that children will give significantly fewer symmetrical responsesfor sentenceslike (8). This is exactly what happens. Subjectswho meet the criteria of being "pure cases " of a symmetry child give significantly fewer symmetrical responses to questions like (8), than to questions with transitive verbs, such as (6) and (7).

Only advocatesof the Competing FactorsModel would concludethat theseresults support the SymmetricalAccount. From the perspectiveof the Modularity Matching Model , the results are troublesome . It turned out that there was only a 19% reduction

in the proportion of symmetricalresponsesto sentenceswith intransitive verb phrases, such as (8), as comparedto responsesto sentenceswith transitive verb phrases, such as (6). Symmetrical responses were given by children on 38% of the trials for sen-

tenceslike '(8), even though they lack the ingredients (two NPs) essentialto forming the linguistic representationthat underliesthe SymmetricalAccount. Theseerroneous responsesby children were chalked up to "responsestrategies" (Philip 1995, 107), "strong carry-over effects [that] were observed to confound the performanceof all subjects" (Philip 1995, 124),. and "the similarity of the picture-types across all experimental conditions" (Philip 1995, 109). However, the proportion of the data attributed to "carryover" effects is roughly twice that of the data supporting the experimental hypothesis, according to which the "pure symmetry child" adopts a nonadult linguistic analysisof quantificationalsentenceswith transitive verb phrases. Figure 11.4 summarizesthe findings. High errors rateslike this are tolerated by researcherswho view linguistic behavior as an aggregate of linguistic and nonlinguistic factors, the expression of children's

Factors

95

linguistic knowledge often being overcome by unknown factors. Assuming that adults correctly comprehend sentenceslike (8), the acceptanceof such high error rates by children would be tantamount to surrendering a fundamentalassumptionof the Modularity Matching Model : that children and adults share the samecognitive mechanisms . A high proportion of errors in responseto sentencesthat are structurally quite simple does not comport well with the Modularity Matching Model and must be explained. It will be, in chapter 35. We postpone discussionuntil then, however, becausethe source of children's errors is quite subtle and can be properly understood only after an extended discussionof methodological issues.

Chapter 12 LanguageProcessing

This chapter and the next discusstwo factors that researchershave proposed to be in competition with grammaticalknowledge in tasks that assesschildren's linguistic competence: memory and extralinguistic knowledge. On the Competing Factors Model, such factors are expected to exert an influence on performance, to some degree, in any task. According to this model, extralinguistic knowledge can be either harmful or helpful, depending on its affinity with the representationgeneratedby the grammar: if grammaticalprinciples and extralinguistic knowledge agree, performance is improved; if they conflict, performanceis depressed . By its perceived nature, on this model, memory is detrimental to performance. On the Modularity Matching Model, by contrast, memory demandsand extralinguistic knowledge playa much more limited role in language processing. In this chapter, we will argue that the influence of memory demandsis felt only in certain tasks and on certain linguistic constructions. In the next chapter, we make similar claims about extralinguistic knowledge. On the Modularity Matching Model, extralinguistic knowledge is used only as a last resort. It is engaged only when children (and adults) are confronted with sentences that are structurally ambiguous and when the specificcontext in which the sentence is introduced is consistentwith more than one of its readings. Becauseof the minimal contributions of factors such as memory and extralinguistic knowledge on the Modularity Matching Model, performanceis expected to accord, first and foremost, with a person's grammaticalknowledge. This leads to the expectation that children will perform (nearly) as well as adults in tasks that tap their linguistic knowledge, such as the elicited production task and the truth value judgment task. 12.1 Competence versusPerformance According to the Modularity Matching Model, the linguistic performanceof children and adults should be (close to) perfect in properly designed experimental studies, provided that children know the relevant points of the adult grammar. According to some advocatesof the Competing FactorsModel, by contrast, perfect performance is an "idealization." These advocatesof the model assumethat this follows from the basic distinction between competenceand performance, as Lust, Chien, and Flynn. (1987) make clear: [P]erformance will never directly reflect only grammatical competence, but other factors as well. It will always be variable, even when it reflectsgrammatical competence.0. . Thus, becauseof the nature of performancedata, behavioral

98

Chapter 12

evidencefor significantgrammaticalfactorswill alwaysinvolve significantcontrastsof degree , not absolutesuccess with one conditionand absolutefailure with another.(p. 282) Of course[UniversalGrammar ] will neverbedirectlyevidenced in anyparticular behavior.. . . (p. 279) Psycholinguisticresearchis specificallyempiricalin the sensethat it pursues the assessment of languageknowledgethroughthe measurement andanalyses of variousmodesof languagebehavior , principallyspeakingandunderstanding . Thelanguagefunctionswhichthesebehaviorsreflect, viz., languageproduction and comprehension , constitutelanguageperformance . Eachperformancefunction involveslanguagecompetence , but it alsoinvolvesvariousotheraspectsof languageknowledgerelatedto the specificbehaviorsof eachperformance function. Evidenceof languagecompetence . . . is thus alwaysmediatedby the processingfactorsinvolvedin eachbehavioralinstanceof eachlanguagefunction, aswell asby the basicvarianceassumed in any samplingof performance data. (p. 273) This view of the interrelationof competenceand performanceis not the only one, however. One's view of the relationbetweenthem dependson a specificset of assumptions about the role of grammarwithin the performancesystem. Few researchers in child languagemakecleartheir basicassumptions aboutthe theory of performance or the positionof grammarwithin it. Admitting that usefultheoriesof performance aregenerallyonesthatincorporatetheoriesof grammar , mostresearchers are contentto draw the immediateinferencethat grammaris just one factoramong many, with the consequence that all performance is lessthanoptimal. In this chapter , we indicatewhy our own conceptionof verbalworking memorydoesnot adhereto this expectationof the CompetingFactorsModel. 12 . 2

The Role of Working Memory in SentenceProcessing

As Chomsky(1965) concludes , to gain insight into the performance system , it is instructiveto look at the specificgrammaticaldevicesthat exceedthe resources of the system , aswell asat thosethat do not. Accordingto the modelof the performance systemwe endorse , linguisticbehaviorordinarilybeginsto degradeonly when specific boundariesof the systemareexceededor when specificprocessingroutinesare utilized. When certaingrammaticaldevicesare exploitedby the performancesystem , the resultis unacceptable . Onedevicethat is knownto resultin unacceptability is repeated self-embedding . This characteristic of the humansentence -processingmechanism explainsthe difficulty peopleexperience with multiply center-embeddedsentences like (1) (ChomskyandMiller 1963; Chomsky1965). (1) Thecheesethat the rat that the catchasedateis on the table. Ambiguity gives rise to unacceptabilityin certaininstances . An exampleis the classicgardenpath sentence , Thehorseracedpastthe barnfell. We will discussthe reasonfor the unacceptability of this kind of sentence in chapter13. Thehumansentence -processing mechanism alsohasa boundedmemory. However, the consequence of a finite memoryfor acceptabilityis not aspronouncedasit is in

LanguageProcessing

99

the caseof sentencesthat contain repeatedself-embeddingor ones that evoke garden path effects. This is attested by the acceptability of multiply right -branching structures , as in (2 ).

(2) This is the cat that caught the rat that ate the cheesethat is on the table. Further evidence that memory limitations are not always detrimental to linguistic performance comes from the literature on language breakdown. Studies of braindamagedpatients with severememory deficits reveal that these individuals are often able to processsentencesof considerablelength, and ones that involve grammatical

devicesthat are structurallycomplex, accordingto a pretheoreticalmetric of complexity. For example , Martin (1993) reportedthat her patient, E.A., who hasa severe immediate -memorylimitation, couldsuccessfully comprehend simpleactiveandpassive sentences , as well as several kinds of center -embedded

relative -clause structures .

In addition, E.A . was 90% accuratein detecting gender mismatches , as in (3), where

the antecedent andthe reflexiveareseparated by over 10 words. (3) The girl fell down the stairs at least once a week and often hurt himself. 12.3 Acceptabilityand Processing Difficulty If the goal is to discover the operating characteristicsof the performancesystem, then a good researchstrategy is to examinethe relative acceptability of different grammatical devices, suchas self-embedding, branching direction, and ambiguity (seeChomsky

1965). It alsomakessenseto utilize experimental measures that areextremelysensitive to processing difficulty . In this way , the researcher will not overlook factors that

help or hinder linguistic processing, but do not cause unacceptability. Earlier we mentioned one such technique: the cross-modal priming paradigm. Another useful tool is eye-movement recording (eye-tracking). Recording the length of eye fixations during reading has been used effectively to assessprocessing difficulty in studies of adults. These studies amply demonstratethat acceptablesentences may impose measurable processing difficulties at certain junctures . It has been

found that the eyes pauselonger in specificregions of certain types of sentencesand that they regress from certain regions of the sentence to earlier regions . Such studies

have added to researchers ' understandingof the nature of the performancesystemby indicating the relative amount of computation that occurs at various points during sentencecomprehension. We will describe two experiments using eye-movement recording in chapter 13, in an empirical demonstrationthat extralinguistic knowledge

doesnot competewith linguisticprinciplesin the constructionof structuralrepresentations

for

sentences

.

The observation that there are nonlinguistic constraintson performanceshould not lead one to infer that a generative grammar does not also underlie linguistic performance. As Fodor and Pylyshyn (1988) remark, the fact that performance improves as memory constraints and the like are lifted is evidencethat linguistic knowledge is also accessedin languageprocessing: [T]here are a number of considerations, which suggest that, despite de facto constraints on performance, one's knowledge of one's language supports an unbounded productive capacity in much the sameway that one's knowledge of addition supports an unbounded number of sums. Among these considerations

100

Chapter 12

are,forexample, thefactthataspeaker/hearers performance canbeimproved

by relaxing timeconstraints, memory constraints, increasing motivation, or supplying pencil andpaper. Itseems verynatural totreatsuch manipulations as affecting the transientstateof thespeakersmemoryandattentionratherthat whatheknows aboutor howherepresentshis language. Butthistreatment isavailable onlyontheassumption thatthecharacter ofthesubjects performance isdetermined byinteractions between theavailable knowledge baseand

theavailable computational resources. . 802inGoldman (p 1993)

Aswehaveshown, theinteraction between theknowledge baseandthecomputationalresources isnotsogreatasto impede performance inmostsentences. More-

over,in certainexperimental tasks,performance constraints canbe relaxed,suchthat

childrens (andadults) grammatical knowledge isrevealed onnearly everytrial.This

is important in studiesofchildlanguage, wherethegoalis to determine thenatureof

thegrammar embedded in theperformance system. Forsuchstudies, then,it makes senseto seektasksthatarenotinfluenced to anappreciable degreebyfactors that contribute eithertounacceptability ortosubtle processing difficulties.

Granting thebasic distinction between grammaticality andacceptability, thegoal ofempirical research inchildlanguage boilsdownto determining which sentences

children findacceptable. Asa simple example, consider (4). (4)

Him left.

Suppose somechildren consistently produce suchsentences (incontexts whereadults

sayLieleft). It wouldbereasonable to conclude thatsuchproductions directly reveal

the grammatical competence of thesechildren. It wouldnot be reasonable to con-

clude thatsuch productions areperformance errors, unless themodel ofperformance thatis beingassumed is quiteunlike thatbeingassumed foradults. Foradults, as noted, memory effects beginto impede processing onlywithcertain grammatical

structures andonlyaftera certain threshold hasbeensurpassed. Thisdoesnotmean, ofcourse, thatmemory isnotusedinprocessing sentences like(4),justthatthelimits

onthememory system donotinterfere withsuccessful production orcomprehension Asananalogy, consider another resource-limited device, theairplane. Inflight, airplanes consume fuel.Fuelconsumption depends onseveral factors, including headwinds, cargo weight, thenumber ofstops, thecourse, andsoon.Ifenough ofthese to any appreciabledegreein suchsentences.

factors conspire against it,theairplane mayrunoutoffuelbeforeit reaches itsdesti-

nation. fortunately, airplanes carryenough fuelmostofthetime;so,despite anin-

herentlimitation in resources, theysuccessfully reachtheirdestinations mostof the

time.Ofcourse, luggage isanother matter. Theintentoftheanalogy should beclear;

humanlanguage processing is similarly constrained by resource limitations such

asmemory andattention. Nevertheless, thisdoesnotentailthatperformance is

negatively affected mostofthetime.Onthoseoccasions where factors conspire against it,thelanguage-processing system doesbreak down, asinprocessing multi-

plyself-embedded constructions, gardenpathsentences, andthelike.Mostof the

time, however, people succeed inproducing sentences withthemeanings theyintend toconvey, andtheysuccessfully understand sentences thatareproduced byothers.

Tosummarize, thenature ofthegrammar thatisincorporated intheperformance system canonlybeexamined bylooking atlinguistic behavior, thatis,performance.

LanguageProcessing

101

In assessinggrammaticalknowledge, however, it is necessaryto present an explicit model of the interrelation of grammar and other componentsof performance, such as memory. - As the case of E.A . makes clear, linguistic constructions of considerable complexity (from a pretheoreticalpoint of view) can be produced and understood by people with extremely limited memory capacity, as measuredby memory spantasks. It follows that measuresof memory span do not tell us all that we need to know about the human verbal memory system. The systemmust be more than a short-term storage buffer, as suggestedby the name "short-term" memory. Let us therefore use another name to refer to the memory system used to processlanguage: verbalwork. Ing memory. 12.4 VerbalWorking Memory On our conception of verbal working memory, the memory systemhas two working parts. The first component of memory is a storage component that holds linguistic input briefly, between a second or two, in phonological form. It is this component of verbal working- memory that is used in tests of memory span. Given its limited capacity, only short stretchesof unstructured linguistic input can be maintained in this component of verbal working memory. To overcome the limitations of the phonological storage buffer, linguistic input must immediately be recoded into a more durable form and then transferredupward through the system. This is the function of the secondcomponent of working memory: the control mechanism. The control mechanismtransfers structural descriptions of the linguistic input between the different levels of linguistic representationwithin the language-processing system: the phonology, the lexicon, the morphology, the syntax, and the semantics. The control mechanismoperatesat severallevels simultaneously. However, its computational resourcesare also limited. Therefore, once the linguistic input has been assigned a partial structural analysisat any level of linguistic representation, the result is immediately transmitted to the next higher level. The lower-level information can then be discarded, so that additional material can be assimilated. The control mechanism ensuresthat the results of computations at lower levels of representation are transferred upward through the system quickly enough to promote the on-line extraction of meaning from the linguistic input. As long as the control mechanismof the working memory system remains intact, even people with a severely limited phonological storage buffer, such as Martin 's patient E.A ., should be expectedto understand sentencesof considerablelength and complexity. However, if the partial structural analysisat any level within the system cannot be immediately integrated into the structural representationthat has already been coffiouted at that level, the assimilation of further input will be blocked. This .. can curtail linguistic processingin patients with limited short-term memory capacity, but only for certain linguistic devices- for example, constructions that require the results of linguistic analysis to be maintained at one level for subsequentuse at the next higher level. (SeeCrain et al. 1990 for a more detailed description of this model of the verbal working memory system and for empirical findings consistentwith the model.)1 For the vast majority of sentences , however, this conception of verbal working memory leadsus to expect linguistic behavior to be almost flawless. As long as neither the sentencesnor the task exceedsthe limitations of the system, both children

102

Chapter 12

and adults should be expected to exhibit linguistic behavior that is consistent with their underlying grammar. This expectation does not hold on the Competing Factors Model, however. This model assumesthat performance factors always weigh in against grammatical knowledge. No sentencesor tasks are exempt from processing demandssuchas memory limitations. We have now sketched the role of memory in language processing, according to the Modularity Matching Model . The topic of the next chapter is extralinguistic knowledge. According to the Competing Factors Model, extralinguistic knowledge contributes to all linguistic behavior, just as memory does. The influence of extralinguistic knowledge is more complex, however. Like memory, extralinguistic knowledge sometimesinterferes with children's responses , reducing the number of correct responses(i.e., onesconsistentwith their grammars). At other times, however, extralinguistic knowledge inflates the number of correct responses ; if extralinguistic knowledge and grammatical knowledge both give rise to the correct response, then this responsewill increasebeyond the level that would be attained by grammaticalknowledge alone. Becausethe Modularity Matching Model disagreeswith the Competing FactorsModel's assessmentof the role of extralinguistic knowledge in languageprocessing, it is worth examining this role more closely.

Chapter13 ~ic Extralinguist

Since

parsing

the

considerations

Coffioeting

Ao the

human

the

role

on

sentence

or

,

process

reading

that

bility

.

knowledge

13 . 1

the

what

to

this

lel

the

. As

Matching as

, the

soon

as

one

only planes

, and

Shankweiler

, Crain

ambiguous

are

shunted

that

the

parser

, but

than

only

(a

one

of

, if

its

Model

how

children Model

specific

sentence

the

read

and

-

the

not

in

any

regardless

of

permissive

,

and

which

others its

a

allowing

sentences

adults

, extralinguistic

context

, but

selected

.

reading be

. One

discourse Crain

a sen , then

priori

-

the

plausi

-

extralinguistic

.

, Ni

for

, et

adapted ;

are

processed , the

in ( partial the

ambiguity

strategy

and

) . According

, where

resolving

from

Altmann

al . 1996

encountered

. This ( 1985

is

1985

component

strategy

succeeds

model

. ( p . 331

in

of

the

referring domain

called

it

is

-

) syn

-

current

is

is

paral

to the

pick Prin

-

).

Success

unquestionably

, when

an

sentence . This

to of

entities

already

discourse

established

, then

favored

in over

)

Ordinarily

dangerous

is

/ pragmatic

Steedman

parser

sentences

a sentence

Referential

the

; Crain

representation and

the

Steedman

Success

not

of

of and

ambiguous

of

checked

that

' s mental

of

of

semantic

by

architecture

1996

the

the

reading

does

resolution

can

fits

Referential a

perceiver

Principle

one

about

( Crain

fragment to

Success

is

analyses

is

best

of

there

the

first

ambiguous

Matching

unambiguous

, the Theory

an

that

Model

alternative

Referential

If

more

is

focus

the

engaged

Matching

more

of

Referential

representation

reading

Principle

guity

the

; Ni

analyses

of

processing

chapter

. We

structurally

be

with

. If

is is

this

of

.

the

context Model

of

. When

Modularity

context of

topic

is

and

Ambiguity

model

discourse

that

Modularity

reading the

to

1988

tactic

by Factors

known

Steedman

the

Model

characteristics

processing

) may

the

the

ambiguities

predictions

specific

Competing

) . The

sentence

predictions

. On

on

sentence

consistent ,

similar

one

influence

is

Matching

longer

parser in

plausibility

different

supersede supports

Modularity is

priori

make

supported

a

that

make

Modularity

little

structural with

circumstances

Model

not

Resolving

resolving

sentences

is

The

of

models

produced

plays

circumstance set

the

may is

The

a

unambiguous

knowledge

ciple

in

Factors

However

tence

in

the a

( i .e ., the

confronted ( i . e ., a

limited

Competing

the

is

both

dwell

device

used 's)

appears this

in

will

knowledge

knowledge

. In

On

- processing

adult

extralinguistic sentence

we

extralinguistic

an

large

Model

characteristics

's

ings

loom

Factors

that

the

child

Knowledge

sentence

is

underwrites

ambiguous

felicitous may

most

sentence . Consider

describe

is the

the

linguistic

hazards

cases

presented

of

of in

chestnut flying

an

ambi

context Flying airplane

,

104

Chapter 13

or the hazards of standing too near a runway, for example. Despite the ambiguity outside of context, only one reading will successfullymatch the conversationalcontext in ordinary circumstances ; the speaker/hearerwill either be flying an airplane or standing too near a runway, but not both. Concerning the influence of plausibility and general world knowledge, Crain and Steedmanpropose another parsing principle: Principleof A Priori Plausibility If a reading is more plausible in terms either of general knowledge about the world, or of specific knowled~e - about the universe of discourse,. then.-, other - ----things being equal, it will be favored over one that is not. . . . In caseof a conflict between general and specificknowledge, then the latter must clearly take precedence. (p. 330) The final sentencepertains to the issueof modularity. If one reading of an ambiguity is more plausible according to general knowledge of the world, but another reading is consistentwith the conversationalcontext, then the reading that matchesthe conversational context will be selected. This insulates the language-processing system from the influenceof extralinguistic sourcesof information, which cannot serveas the basis for the parser's decisionsin resolving ambiguities, as long as there is a specific context. Adopting a similar stance, Fodor (1984) compares a modular parser with what he. calls a "New Look" parser, in which extralinguistic factors infiltrate language processIng: In the extreme case, aNew Look parser can bring to the processof assigning structural descriptionsanythingthat theorganismknows(or believes, or hopes, or expects. . . etc.). . .. [B]y definition, a New Look parsertends to hear just what it ex-pects to hear. . . . [T]his . . . suggestsone of the reasonswhy encapsulatedperceptual modules might be quite a good thing for an organism to have: background beliefs, and the expectations that they engender, from time to time prove not to be true. (p. 132 in Goldman 1993) Fodor's argument is familiar, but it bears repeating. Extralinguistic knowledge had better not infiltrate the initial operations of the parser; if it did, the parserwould tend to disregard analysesthat lead to the unexpected, according to one's previous experience. This would wreak havoc in exceptional circumstances , where "deviant" sentencesturn out to be true (although they would be falsein more usual circumstances ). It does not follow , however, that a priori plausibility (i.e., general knowledge of the world) is never used in parsing, just that its role is quite limited: it is accessedonly after the alternative structural analysesof an ambiguous sentencehave been computed and then only if specific contextual information has not already resolved the ambiguity . 13.2 ResolvingAmbiguity in the So-CalledNull Context Crain and Steedman(1985) suggest another parsing principle for sentencesthat are presented outside of context. When sentencesare presented in the absenceof context, as in most studies on adult sentenceprocessing, the perceiver actively attempts to construct a mental model of a context that would makethe sentencefelicitous. The relative complexity of constructing the alternative mental models dictates which

Extralinguistic Knowledge

107

Theseelementsare already in the mental model of the child, having been introduced by the preceding sentences . Therefore, they can be accessedin interpreting a new sentence. In short, presenting sentencesin the null context may impede children's normal processingroutines, as comparedto presenting sentencesin a specificcontext. There is another negative implication of presenting sentencesin the null context: loss of experimentalcontrol. As Crain and Steedman(1985) put it : [T1heso-called null context is in fact simply an unknowncontext. .. . [T1heset of assumptionsunder which the subject constructsa context are not under control. Any processing preferencesthat are observed may therefore be due to facts about the construction of that context, rather than about the structural properties of the different readings. It is therefore necessaryto control context in a much fuller sense, offering the target sentencesin a matrix of preceding sentences[or situations] that unambiguously establisha known set of assumptionsunder which the subjectwill approachthe target sentence.(p. 338) Both tasksdescribedlater in the book take this advice to heart. The design featuresof these tasks are concernedwith manufacturing specificcontexts that are felicitous for the linguistic constructionsunder investigation. 13.3 Memory and Modularity According to the Modularity Matching Model, the human sentence-processing device is a parallel processing system. When structural ambiguity is encountered, the control mechanismof v.erbal working memory relays partial structural descriptionsof each of the alternative analysesfor evaluation and resolution by the semanticprocessor. The analysisthat best fits (the perceiver's mental model of ) the discoursecontext is the one that is selected. If none of the alternatives succeedsin referring to entities in the perceiver's mental model, then a reading is selected based on the number and kinds of accommodationsthat must be made to the mental model. If the ambiguity still cannot be resolved, information outside the languagefaculty (i.e., extralinguistic knowledge) can be consulted in order to resolve it . In short, the Modularity Matching Model conceivesof the language-processingsystem as hierarchically organized and encapsulatedfrom the influence of nonlinguistic sourcesof information. Following Fodor (1983), we can refer to this last architectural feature of the model as the Modularity Hypothesis . Recent researchwith Weijia Ni and Donald Shankweiler has produced empirical evidence that is consistent with the Modularity Hypothesis. The data suggest that syntactic and semanticstructuresare used for ambiguity resolution earlier than extralinguistic knowledge (a priori plausibility). These conclusionsrest on experimentsin which subjectswere divided into groups according to differencesin verbal working memory. It will be instructive to sketch the experiments as a way of clarifying the role of memory and extralinguistic knowledge in languageprocessing. We will report on the findings of two experiments. Both experiments exploit the techniqueof eye-movement recording (eye-tracking). By monitoring subjects' eye movements during reading, we are able to obtain two measuresof processing difficulty : first-pass reading times (i.e., the total duration of all fixations in each region of a sentence) and incidence of regressiveeye movements. It has been our working

108

Chapter 13

hypothesis, and that of others, that first-passreading times reflect on-line processes ; theseresponsesreflect the influenceof information that is immediately assimilatedby the reader. Regressiveeye movements are another indicator of processingdifficulties that readersencounterduring their first passthrough the sentence.In contrast to firstpassreading times, regressiveeye movementstestify to the reader's failure to use information on-line for interpretation. Subjects in both experiments were also assessedfor verbal working memory capacity, using a memory span test adaptedfrom Danemanand Carpenter 1980. For purposes of analysis, subjects were put in two groups of equal size: subjects with higher scoreson the memory test were put in one group, called the high-spangroup, and subjectswith lower scoreswere put in another group, called the low spangroup. Given our conception of the role of verbal working memory, we did not expect differencesin memory capacity among the subjectsto influence sentenceprocessing as long as the information neededfor recovering from a misanalysiswas structural. Structural principles are accessedwithin the language-processing system and should be costly of memory resourcesonly in unusualcircumstances , as discussedin chapter 12. If this is correct, then the eye-movement profiles of subjectsin the high- and lowspan groups should not differ in resolving ambiguities as long as the disambiguating information was structural. We did expect, however, that individual differencesin memory capacity would be relevant when extralinguistic knowledge was needed in order to resolve an ambiguity . It is assumedthat querying extralinguistic knowledge taxes working memory, especiallywhen the alternative analysesof an ambiguousphrasemust be maintained until the ambiguity is resolved. The combined effort of these computations is likely to create processing difficulties that can place high demandson memory resources. Therefore, we expected the eye-movement profiles of subjectsin the high- and lowspan groups to differ when the task involved resolving ambiguities using extralinguistic knowledge. Each eye-tracking experiment presented a different set of sentences , but the sets had certain features in common. In both sets, a semantic property of the test sentenceswas manipulated, by alternating the definite determiner the and the focus operator only. The first set of sentencesinvolved the main-clausejreduced -rel.ative..;clause ambiguity characteristicof the classicgarden path sentenceThe horseracedpast the barn. . . . The secondset involved an ambiguity in the site of attachmentof a prepositional phrase; the prepositional phrasecould attach either to the precedingnoun or to the preceding verb, as in . . . "pottedthe man with binoculars . On one reading, the man was spotted by someoneusing binoculars; on .the other reading, the man with binoculars was spotted. An important differencebetween the sentencetypes was the type of information that disambiguated the sentences . In the first set of sentences , a misanalysiscould be detected on the basis of structural information; test sentencesvaried in the presence or absenceof the conjunction but, as a signal of the misanalysis. In the second set, misanalyseswere signaled by the subject's general knowledge of the world : for example, the fact that it is implausible to paint using large cracks, or for doors to have new brushes. According to the Modularity Matching Model, the alternation between the definite determiner the and the focus operator only should result in a different analysisof the ambiguity in each experiment. The predictions of the model, as they pertain to

Extralinguistic Knowledge

109

acceptability, are as follows: Experiment 1: Structural disambiguation Acceptab Ie a. The horsesracedpast the barn but were unable to clear the jump. b. Only horsesracedpast the barn were unable to clear the jump. Unacceptab Ie c. The horsesracedpast the barn were unable to clear the jump. d. Only horsesracedpast the barn but were unable to clear the jump. Experiment 2: Disambiguation by extralinguistic knowledge Acceptab Ie a. The man painted the doors with new brushesbefore the festival. b. The man painted only doors with large cracksbefore the festival. Unacceptab Ie c. The man painted the doors with large cracksbefore the festival. d. The man painted only doors with new brushesbefore the festival. To see why the sentenceswith only and the should be analyzed differently, consider the on-line derivations of the meaning representations(i.e., the mental models) in the first set of sentences . First, the initial NPs, the horsesand only horses , compel the parser to establish a set of horses in the mental model. We have already explained why the main-clauseanalysisof the (a) and (c) versions, with the definite determiner, is preferred over the reduced-relative-clause analysis. The consequenceof this is a garden path effect for the (c) version, but not for the (a) version. It remainsto explain why the reduced-relative-clauseanalysisis preferred over the main-clauseanalysisfor the (b) and (d) versions, leading to the opposite pattern of acceptability. The focus operator only requires the parser to checJ < the mental model for a contrast set corresponding to the head noun it is in construction with - in the present example, horses . However, no contrast set has previously been established, because these sentencesare presentedin the absenceof context. The only entities in the mental model are the horses, which have just been incorporated into it . According to the Principle of ReferentialSuccess , the parserfavors a reading that succeedsin referring to entities already establishedin the perceiver's mental model of the domain of discourse over one that does not. Therefore, the parser should attempt to resolve the ambiguity, if possible, by adopting a reading that refers to the horses in the mental model. Consequently, the parsershould chooseto resolve the ambiguity by using the ambiguousphrase, racedpast the barn, as a way of modifying the preceding NP, only horses . The resulting semanticinterpretation establishesa contrast set, as required by the focus operator. The contrast set is formed by partitioning the horses into ones that were racedpast the barn and ones that were not.l Having analyzed the ambiguous phrase as a reduced relative clause, the parser should not experiencea garden path effect when the verb phrase continuation (were unable. . .) is encountered, as in (b). On the other hand, a garden path effect should arisein (d), becausethe reduced-relative-clauseanalysisis inconsistentwith the mainclausecontinuation (but wereunable. . .). Similar differencesare expected to turn on the alternation of the and only in the ambiguities involving the attachmentof a prepositional phrase, as in the sentencesfrom experiment 2. The presenceof the focus operator only should causethe prepositional phrase to attach to the preceding NP

110

Chapter 13

in the (b) and (d) versions. The NP that is createdin the (b) version is doorswith large cracks ; it is plausiblefor someoneto paint such doors. A pragmatic anomaly results in the (d) version, however, becausethe phrasethat is createdis doorswith new brushes ,a nonsensicalreferenceset. We are ready to consider the experimental findings. They are just as expectedby the Modularity Matching Model. The eye-movement patterns of both high-span and low -span subjectswere virtually identical in experiment I , where misanalysisof an ambiguous sentencewas signaled by structural information. That is, there were no between-group differencesin subjects' responsesto the test sentences . There was no evidence of garden path effects for the (a) and (b) versions either in the first-pass reading times or in the incidence of regressions, for either high-span or low-span subjects. Both groups of subjectsexperiencedgarden path effects for the (c) and (d) versions, however. This is attested by elevated first-pass reading times for both groups; the incidenceof regressionswas not significant for either group. The patterns of eye movements by high-span and low-span subjectsdiverged in experiment 2, however, where the misanalysisof the ambiguity was signaled by extralinguistic knowledge. High- and low-span subjects exhibited a similar pattern of responsesfor the (a) and (b) versions, which were expectedto be acceptable , on the Modularity Matching Model. The patterns of eye movements were also similar for both groups for the (c) versions, which contained the definite determiner the. Both high-span and low -span subjectshad higher first-pass reading times at the point of disambiguationfor these sentencesthan at the correspondingregions in the (a) or (b) versions. Where the subjectgroups differed was in responseto the (d) versions. For these sentences , high-span subjectsexhibited different kinds of eye-movement patterns than low -span subjects. High-span subjectswere able to recover from misanalyseson-line, although recovery led to inflated first-passreading times. High-span subjectsmade few regressionsto earlier regions, however. The pattern of responses for low-span subjectswas exactly the opposite. Low-span subjectswere largely unable to recover from misanalyseson the first passthrough the (d) versions. This resulted in the finding that low-span subjectshad faster first-passreading times in the disambiguating region for these pragmatically anomalous sentences , although the difference between groups was not significant. The impact of the anomalies took a different form for these subjects; they made frequent regressions from the disambiguating region to earlier regions. In processingthe (d) sentences , it seemsthat the memory resources of low-span subjects had been exhausted by the time the anomaly was detected, so they were forced to deal with it by returning to earlier portions of the sentence. The eye-movement patterns of low-span subjects suggest that, although they were aware of the anomaly, their limitation in verbal working memory did not allow them to recover on-line from a misanalysis; they had to look back in order to reanalyzethe ambiguity.2 In short, low-span subjectswere unable to use extralinguistic knowledge as effectively as high-span subjects, who were able to recover on-line from misanalysessignaledby semanticplausibility.3 Taken together, the findings of theseexperimentsare evidencethat the processing of extralinguistic knowledge lags behind syntactic and semanticprocessing. This explains the absenceof individual differenceswhen structural information signaled that an ambiguity had been misanalyzed, and the appearanceof group differenceswhen extralinguistic knowledge was neededto infer that the wrong analysis had been assigned. The findings suggest that the operations of the syntactic and semanticcom-

Extralinguistic Knowledge

111

ponents occur within the language-processingsystem and do not exceedthe capacity of verbal working memory. Differences in memory capacity begin to matter, however, when the disambiguating information comes later, from outside the languageprocessing system . If all of this is correct , then the evidence from these experiments

supports both the Modularity Hypothesis and the Modularity Matching Model . 13.4 Children's Parsing Strategies

So far, we have been discussingthe parsing strategiesof adults. Of course, according to the Modularity Matching Model, these same parsing strategies should also be manifestedby children, as long as children's parsing decisions are tested in similar ways. As noted earlier, however, the experimental contexts in which children are askedto respond to ambiguoussentencesare not the sameas those in which adults are askedfor their judgments. In experimental investigations with children, contexts are often designedto be appropriateto more than one reading of a sentence.The contexts are specifically constructed so that the test sentenceis true in the discourse context if one analysis is assigned, and false in the context if a different analysis is assigned. Crucially, then, both of the relevant interpretations of the sentencemust be felicitous in the context . Althoug :h such contexts are undeniably rare outside the labo-

ratory, they are necessaryelementsfor evaluating children's grammaticalknowledge. Little is known, however, about the strategiesthat guide the parsing decisionsof children or adults in circumstancesin which both readings'of an ambiguity are felicitous.

.I

-

In our experience with children , we have been able to identify several factors that

conspire to determine which reading of an ambiguous sentenceis selected.4 First, children pick the reading that makesan ambiguous sentencetrue in the context, assuming that both readings are consistentwith the context. On the basis of this finding, many researchersassumethat children should always accessthe reading that makes the sentence true (see Grimshaw

and Rosen 1990 ). However , there are other

influences on children' s parsing decisions in "ambiguous" contexts. Another is the structure of the context itself- for example, the order in which events take place. If one reading correspondsto the final event in the context, that reading will be favored over readingsthat refer to events that occurredpreviously. FinallyI there seemto be inherent biasesfor one reading or another of an ambiguous sentence. We have already indicated some preferencesby adults to interpret an ambiguity in one way rather than another. The sameholds true for children. For example, children have been found to prefer the direct reference(deictic) reading of the pronoun in questions like (2)- that is, the reading on which the pronoun refers to some salient male individual (2 )

in the conversational

context .

I know who thinks he has a hat .

The alternative , bound variable interpretation

of the pronoun allows he to refer to

more than a single individual (although it need not). It is noteworthy that many children continue to assign the direct reference interpretation of the pronoun , rather

than the bound variable interpretation, even if the direct referencereading makesthe sentence false in the context .

The preferencefor the direct referenceinterpretation of the pronoun in sentences like (2) is not fully general, however. For example, this preferencedoes not extend to the VP-ellipsis construction in (3).

112

Chapter 13

(3) Snoopy thinks he has the best hat, and Mickey Mouse does too. In interpreting (3), children generally assign the reading on which the pronoun he refers to both Snoopy and Mickey Mouse, and not to a third character(say, Donald Duck). The accessibility of the bound variable interpretation (the so-called sloppy reading) in sentenceslike (3), but not in sentenceslike (2), must apparently be chalked up to a construction-specificparsing preference(seeThornton 1990). 13.5 UnambiguousSentences According to the Modularity Matching Model, but not the Competing Factors Model, the principles used in deciding among competing discourse representations do not hold sway over the parser's interpretation of unambiguoussentences . In the comprehensionof unambiguous sentences , according to the Modularity Matching .

-

-

Model, factorssuchasplausibilityandgeneralknowledgeof the world play no role whatsoever. If the principles of grammar made only one reading available, then that reading is parsed, even if it would be highly dispreferred had it been pitted against another reading. To help reinforce this point, it will be useful to return to the classicgarden path sentence , The horseracedpast the barn fell. Notice that the difficulty inherent in the reduced-relative -clause analysis of this sentence is not experienced by the perceiver

in processingits unambiguouscounterpart in (4).5 (4)

The horse ridden past the barn fell .

In understanding(4), there is no consciousdifficulty associatedwith constructing the discourse representation corresponding to the reduced-relative-clause analysis, becausethere is no "simpler" alternative analysis. Therefore, the sentenceis assigned the correct , reduced-relative -clause analysis without conscious effort . Given that chil -

drenhave essentiallythe sameparsing apparatusas adults, which is the basictenet of the Modularity Matching Model, children too should have no difficulty analyzing unambiguoussentences , even if they are compelled to assign an analysisthat would be highly dispreferredif the sentencehad another interpretation. A further example may be helpful. As mentioned earlier, both children and adults attempt to accessthe interpretation of an ambiguoussentencethat makesit true in the discoursecontext, all other things being equal. According to the Competing Factors Model , this "parsing strategy " also holds sway over children (and adults, to a lesser

extent) in processing unambiguous sentences . On the Competing Factors Model, then, even unambiguoussentencesare subject to the vagariesof performancefactors. As a consequence , children who know a linguistic principle can neverthelessperform lessthan perfectly in responseto sentencesgoverned by it . The model maintains that lessthan perfect performanceis sufficient to indicate the effect of a linguistic principle. Proponentsof the Modularity Matching Model take issuewith this way of assessing knowledge of linguistic principles. On this model, if the sentenceis false in the discourse context on the interpretation assigned by the grammar, then children (and adults, to a similar extent ) will reject the sentence; they will not attempt to reanalyze

it in a way that makesit true. Therefore, the Modularity Matching Model maintains that only near perfect performanceis sufficient to indicate the effect of a linguistic principle, either for children or for adults.

Extralinguistic Knowledge

113

13.6 Conclusion The Competing FactorsModel and the Modularity Matching Model diverge in their expectations about the influence of memory and extralinguistic knowledge. On the Competing Factors Model, these factors playa role in processing sentencesof all kinds. On the Modularity Matching Model, they playa much more limited role. In the resolution of ambiguity, extralinguistic knowledge is relevant only if the mental model of the domain of discourse cannot be engaged in deciding among the alternative readings. The Competing FactorsModel and the Modularity Matching Model also diverge in their predictions about children's responsesto sentencesfor which there is only a single grammaticalanalysis(i.e., unambiguoussentences ). According to the Competing FactorsModel, the kinds of factors that influence children's interpretation of ambiguous sentencesmay also influence their interpretation of unambiguoussentences . By contrast, according to the Modularity Matching Model, if linguistic principles eliminate all but one semantic representation, leaving a sentenceunambiguous for children, then they will have no difficulty interpreting the sentencecorrectly. To repeat the main point, the Modularity Matching Model predicts that both children and adults will invoke the sameparsing strategiesto resolve ambiguity. This follows from the basic architecture of the model. Children and adults differ in one important respect, however; until two or three years after they begin to speak, children lack full linguistic competence. In the next chapter, we show that even this brief delay in acquisition has consequencesfor the Modularity Matching Model and implications for the Modularity Hypothesis.

Chapter14 When Principles and Preferences Collide

This

chapter

presents

evidence

comes

vestigate

children

possible

,

acquired

could are

may

be

children

the

not

. 1

the

The

This

is

whether

The

big

a .

The

only

thing

b .

The

only

elephant

giving

is

refers

to

the

use

possible

,

operator

of

the

contrast

The

.

There

of

meanings

.

This

finding

that the

adults

data are

learning

under is

seen

the

LAD

) .

to

different

( i . e .,

set

is

the

set

only

playing

,

be

-

chapter , however

12

in

.

. Most

guitar

the

( 1 ) ,

of 3 -

to

taken sets

.

of

but

to

dictated

responses 5 - year

for

by - old

(a )

On

the

a

nominal

element

the

proform

only set

one

one for

struc

the

interpretation (b )

the This

-

focus ,

every

interpretation

,

the

.

adults

,

( say

the

Principle

children

children

.

has

animal by

of

for

revealed

another

interaction substitution

of

discourse

true

the

contrast

the .

adults be

the

.

one

sentence

On set

( 1 ) .

.

referent

as

for

. Consider

elephant

contents

the

domain

is

pattern of

is

and

( 1 ) . For

preference

The

the

big

as

one

contrast

children

a

This

the

the

though

what

in

to than

contrast

elephants

of

.

on

the

involves

back

even

elephant

known

more

,

is

playing

well

If

possible

both

is

ambiguous

least

.

big

ambiguity

refer .

turns

interpretation

as

to

results

of

elephant

one in

of

guitar

are

prefer

guitar

the

phenomenon

one

two

the is

the

that

adults

playing guitar

. The

sentence

discourse

studies (b )

adults

ambiguity are

one the

linguistic

a

sentences that

playing

proform

of

interpret

only

for

the

domain

the

adults

and of

principles

initial

prefer

,

meanings

architecture of

children

principles

interpretation

the

ambiguity

Experimental

of

adults

mental

are others

various

the

is are

over

by

that

interpretation

of to

playing

the

in

. only

in

is

and

semantic

analysis

cussed

that

modular

modules access

the

ambiguous

mentioned

the

readings

proper

reading

-

It

sentence

governed

out

in

.

meanings

of

is

turn

to adults

ambiguous

acquisition LAD

could

of

ever

them

elephant

only

previously

prefer

the

. The

children

( 1 )

operator

thing

the

it

kind

the

of

for

one

. The

designed

alternative for

order that

not

have

initially

sentence

be

that adults

by

focus

tural

the

,

the

an

which

preferences

principles

Model

in

,

ambiguous

of

order

Hypothesis adults

Substitution

question

adults

the

,

for

observation not

,

adult

are

are

meanings

so

Modularity and

that

. Assuming

support

but

One

LAD parsing

Matching

children

If

the

children

sentences

Unlike

sentences

empirical

on

cause

.

of

with

principles

the

support

alternative

.

by

Modularity

hinge

the

children

semantic

to

offer

in

experiments of

that

by

and

assign

would

is

on

determined ,

of

instructive

based

leamability

lies

, by

be

evidence

set

understanding

likely

piecemeal

which

a

' s

even

acquired

14

further

from

,

following

the

big

an

not

interviewed

:

Adults

elephant

octopus of

was we

the

)

must might

be

Parsimony the

same consistently

, as

dis that

-

116

Chapter 14

rejected sentence(1) if any characterother than the big elephantwas playing a guitar. That is, children adopted the (a) interpretation of (1). On the face of it, children's nonadult responsesto (1) invite the inferencethat children have internalized a parser with different operating characteristicsthan the adult parser. It looks as though children are not using the Principle of Parsimonyto resolve the ambiguity, as adults are. There is another interpretation of the findings, however. On this interpretation, certain semanticrepresentationsthat are initially adopted by learnerson learnability grounds are dispreferredby the sentence-parsing mechanism once additional representationshave been acquired. Putting it the other way around, certain linguistic representationsthat are preferred by the sentence-parsing mechanism are ones that would create problems of learnability if learnersinitially adopted them. Therefore, children's initial representationis on~ that adults disprefer, but once children add additional representationsthemselves, they respond using the same parsing principles as adults do. Here is the argumentin a bit more detail. First, recall the relevant operating principle of the parser, Crain and Steedman's (1985) Principle of Parsimony: Principleof Parsimony If there is a reading that carriesfewer unsatisfiedpresuppositionsor entailments than any other, then, other criteria of plausibility being equal, that reading will be adopted as most plausibleby the hearer, and the presuppositionsin question will be incorporated in his or her model. (p. 333) Crain and Hamburger (1992) suggest that the Principle of Parsimony is ultimately motivated by the need to minimize cognitive effort. To conserveeffort, unnecessary extensions to the mental model are avoided whenever possible. The advantage of this "least effort" strategy for ambiguity resolution is to reduce the risk of making commitments that will need to be changedlater on. In other words, the parser is a "minimal commitment" device. Adopting the minimal commitment strategy, whenever the parser confronts a structurally ambiguousphraseor sentencefragment, it begins to construct discourse representationscorrespondingto all its structural analyses. Being unable to maintain multiple discourse representationsin working memory for long, however, it must rapidly abandonall but one representation. The decision about which representation to keep and which to discard is made by considering the sets of circumstancesthat make the sentencetrue on each discourserepresentation. Sometimes, these -alternative setsof circumstances(correspondingto the alternative discourserepresentations ) form a subsetjsuperset-relation; that is, the set of circumstancesthat make the sentence true on one discourse representation is a superset of the circumstancesthat make it true on another representation. Whenever this state of affairs holds, the parserselectsthe discourserepresentationthat makesthe sentencetrue in the largest set of circumstances .1 To illustrate this operating characteristicof the parser, we turn back to sentence(I ), with its alternative interpretations (a) and (b). (1) The big elephantis the only one playing the guitar. a. The only thing playing the guitar is the big elephant. b. The only elephantplaying the guitar is the big elephant. It is immediately apparent that pentence(I ) is true on interpretation (a) only in a limited set of circumstances , as comparedto interpretation (b). That is, (a) is true only

Principlesand Preferences

117

in circumstances where the only guitar -playing animal is ~he big elephant . On the other hand, (b) is true in circumstances in which the big elephant is the only elephant playing the guitar , regardless of whether or not other animals are playing the guitar as well . In short , the circumstances corresponding to the (a) interpretation of (1) constitute a subset of the circumstances that make (1) true on the (b) interpretation . That is, the (a) reading is the subset reading , and the (b) reading is the superset reading In formal terms, the subsetjsuperset distinction boils down to an entailment relation between the alternative readings of the ambiguity . The relevant notion of entail ment can be stated as follows : A reading Q of a sentence entails another reading R if and only if every circumstance in which Q is true is also a circumstance in which R is true . To say that the (a) reading of (1) entails the (b) reading , then, amounts to the claim that every circumstance in which "the only thing playing the guitar is the big elephant " is true is also one in which it is true that " the only elephant playing the guitar is the big elephant ." This relation among circumstances does indeed hold ; therefore , the (a) reading entails the (b) reading . The converse relation does not hold , however ; if the big elephant is the only elephant playing the guitar , it may still not be the only thing playing the guitar . Therefore , the (b) reading does not entail the (a) reading . Now we can state the subsetjsuperset relations among the readings of certain ambig uous sentences: If reading Q entails reading R, but R does not entail Q , then Q is the subset reading and R is the superset reading . By this definition , (a) is the subset reading of (I ), and (b) is its superset reading . The upshot is this : the minimal commitment strategy (the Principle of Parsimony ) dictates that the parser will favor the reading that is true in the broadest set of circumstances. The interpretation that is true in the broadest set of circumstances is the superset reading . That is why people prefer the (b) interpretation of sentences like (1): (b) is the superset reading . The preference for the superset reading of an ambiguous sentence plays an important role in the design of contexts for the truth value judgment task (see chapter 33). To see which reading you prefer , we propose the following test : Try to imagine a context in which someone tells you (1), The big elephant is the only one playing the guitar , but in which that statement is false. Now , write down the aspect of your mentally constructed situation that makes the statement false. As predicted by the Princi ple of Parsimony , when adults perform this task, they generally write down things like A little elephant was playing the guitar , too. Few adults write things like Someone elsewas playing the guitar too (Crain , Ni , and Conway 1994). 14.2 The Semantic SubsetPrinciple Our assumptions about the operating principles of the LAD are quite different from our assumptions about the operating principles of the parser. The principles of the LAD must enable learners to converge successfully on the target grammar on the basis of the available evidence . Presumably , the evidence available to learners consists of (sentence, meaning ) pairs- that is, sentences presented in circumstances that make them true . Sometimes, more than one interpretation of a sentence is made available

118

Chapter 14

by Universal Grammar.2 To further complicate matters, these alternatives may form a subsetjsupersetrelation; that is, the circumstancesthat make the sentencetrue on one interpretation may be a proper subset of the circumstances that make it true on

another interpretation. In such cases , a semantic subset problem arises if the target languageincludesthe subsetreading, but not the supersetreading. To avoid semantic subsetproblems, the interpretive options for sentencesmust be ordered in the LAD by a principle instructing learnersto initially choosethe representationthat is true in the smallest set of circumstances .3 This is called the SemanticSubsetPrinciple(Crain 1992, 1993; Crain, Ni , and Conway 1994; Crain and Philip 1993). SemanticSubsetPrinciple Supposethat the interpretive component of Universal Grammarmakestwo interpretations , A and B, available for a sentence, S. If so, then see if S is true in a narrower range of circumstances on interpretation A than on interpretation

B. If so, then A will be hypothesizedbefore B in the courseof language development. The SemanticSubsetPrinciple establisheschildren's initial semantichypotheses; these are hypothesesthat can be falsified on the basisof positive evidencefrom the input.4 Just as children lack negative syntactic evidence- evidence about the ungrammat icality of sentence forms - so it seems likely that they lack the kind of evidence that

would be neededto reject incorrect hypothesesabout what sentencesmay and may not mean. Therefore , children who commit semantic overgeneration -

assigning sen-

tence meanings beyond those assigned by adults- would not recover from their errors. Although there is little empirical evidence on the matter, we think it highly unlikely that children expunge semantic errors on the basis of experience .

To avoid semanticerrors in the first place, the SemanticSubsetPrinciple arranges grammatical options to ensure that learners initially hypothesize the interpretation that makes a sentence true in the smallest set of circumstances . In this way , learners

are assuredof formulating falsifiable hypotheses. To make sentencestrue in the narrowest possible sets of circumstancesamounts to making the maximal commitments about

the entities

and events

in the domain

of discourse . In short , the LAD

is a

"ma'ximal commitment" component of the languageapparatus. This brings us back to sentence(1), Thebig elephantis the only oneplaying theguitar. Circumstancescorresponding to the alternative interpretations of the sentenceare depicted in figure 14.1. It is important to note that these are simply some of the circumstances that would make sentence (1) true on the alternative interpretations ; these

circumstanceswould not necessarilybe ones we would use if our goal was to investigat~ children's awarenessof the alternative meaningsof (1). The (a) reading of (1) is depicted on the left of the figure. The circumstancesthat make (1) true on this reading are the ones without the "Ghostbuster" bar through them. It should be clear that these circumstancescorrespond to the maximal commitment interpretation. On this interpretation, the sentenceis false if anything other than the big elephant is playing a guitar .

The (b) reading correspondsto the circumstanceson the right of the figure. This set of circumstances is consistent with the minimal commitment interpretation . On

this interpretation, the contrast set consistsjust of elephants. The illustrations show that the maximal commitment interpretation (a) makes sentence (1) true in a narrower set of circumstances than does the minimal commitment interpretation (b). In fact, on

Principles and Preferences

119

~

ii ~

Ii1:f

7!f~

~ ~

n MaximalCommitment

'::el:

Minimal Commitment

Figure 14.1 Setsof circumstancescorrespondingto the alternative readingsof sentence(1)

the maximal commitment interpretation, (1) is true in only one of the circumstances depicted here.s For the sentenceto be true on the (b) reading, however, all that is required is that no elephantbesidesthe big elephant is playing a guitar. The activities of the octopus and the crane are not pertinent on this reading.6 The crucial observation here is that the (a) interpretation of (1) is true in a subsetof the circumstancescorrespondingto the (b) interpretation. For learners, then, sentences like (1) representa potential semanticsubsetproblem. The circumstancesin which the two readings of (1) are true fall into a subsetjsupersetrelation. The SemanticSubset Principle rescueschildren by compelling them to initially hypothesize the falsifiable interpretation (a), becausethis interpretation makesfewer commitmentsabout who is playing a guitar (cf. Hornstein and Lightfoot 1981). For empirical results, see Crain, Ni , and Conway 1994. 14.3 Conclusion

To recap, the principles of the LAD must be responsive to demandsof leamability. To achievelearnability, learners' hypothesesmust be constrainedso as to guarantee that linguistic representationsthat are not derived in the target language will not

120

Chapter 14

be formulated or, if formulated, can be disconnrmed by readily available evidence. Therefore, it is expectedthat children will adopt the "maximal commitment" reading of (1), reading (a). In contrast to the LAD , the sentence-parsing mechanismselects among the competing linguistic representationsthat are derived in a language. Selection is basedon considerationsof simplicity. Therefore, adults prefer the "minimal commitment" reading of (1), reading (b). This explains why adults and children give (1) different interpretations, letting the Modularity Matching Model off the hook. This chapter also featured two aspectsof children's linguistic behavior that substantiate our claim that children are interesting subjectsfor testing hypothesesconcerning the inner workings of Universal Grammar. First, children are a good sourceof data bearing on the InnatenessHypothesis, and second, child subjectsin the studies cited here produced nonadult responsesthat can be used as evidence germane to hypothesesthat have been advancedin current linguistic theory.

Chapter 15 PerformanceErrors

According to the Modularity Matching Model, the languageapparatusof both children and adults is organized into modules, and the inner workings of these modules are essentially the same for children as they are for adults. In chapter 12, we explained why we add the caveat "essentially the same" to describe the relation between the language-processing subcomponentsof children and adults. There is one important difference between children and adults : children have access to a module

that may be unavailable to adults, the LAD . In the previous chapter, we considered how children's nonadult responsesmight arise from the operating characteristicsof the LAD .

According to the Modularity Matching Model, another viable source of children's nonadult linguistic behavior is the theory of Universal Grammaritself. It is consistent with the theory for children to sometimeshypothesize grammatical representations that are not characteristicof their target language (see chapter 5 for examples). Parametersare one sourceof suchdeviations; another possible sourceis maturation. On the Modularity Matching Model, however, the internal representationsthat children assign to sentencesmust be compatible with the principles of Universal Grammar. This is necessaryin light of the problems of learnability that would otherwise arise. Essentially, children's grammarscan differ from the target grammar only in ways that adult grammars can differ from each other. This is the Continuity Hypothesis. Put simply, the Continuity Hypothesis requires children's grammaticalhypothesesto be constrained by Universal Grammar (see Crain 1991).

In certain cases , however, children produce errors that cannot be readily explained by appealing to the LAD , linguistic theory, or maturation. On the face of it, the existence of such errors might seemto compel us to reject the Modularity Matching Model and consider adopting the Competing Factors Model. The situation is not quite as grim as it might appear, however. As long as errors arise from nonlin~ istic aSDects of the tasks children .

are expected to perform , then these factors cannot count

... against the Modularity Matching Model . Moreover , these factors make clear one basic

L

researchstrategy of the model, which is to eliminate or reduce as far as possible the nonlinguistic demandsof a task. In this chapter, we discussthe nature of errors that sometimesmask children's linguistic knowledge, and we discussways to eliminate these sources of children 's non adult responses .

We discussfour sourcesof children's nonadult behavior: (a) artifacts in experimental design, (b) extralinguistic knowledge that develops over time, (c) the complexity of nonlinguistic processesthat are needed to provide correct answers in a psycholinguistic task, and (d) performancefactors that influencethe linguistic behavior of both children

and adults . It should be noted that the existence

of these factors in no way

122 Chapter15 challengesthe fundamentalprecepts of the Modularity Matching Model . However, these factors are important to the investigation of child language, so they should be afforded the proper respect; we will want to eliminate them or at least reduce them as much as possible in designing tasks to assesschildren's linguistic competence, because, left unchecked, these factors produce underestimatesof this competence . It is ~ important to emphasize , however, that these factors do not compete with linguistic knowledge; they merely stand in the way of progress toward an understanding of children's mastery of linguistic principles. 15.1 Extraneous Demandsof the Task Flaws in experimentaldesign are one source of children's errors. This sourceof nonadult linguistic behavior clearly needsto be eliminated whenever possible. Errors are likely to occur, for example, in experimental situations that force children to violate one kind of linguistic principle or another. In thesecircumstances , children are forced to choosewhich kind of principle to violate. If children choose to violate a syntactic principle, say, in order to provide a pragmatically felicitous response, they should not be said to lack syntactic knowledge. When placed in situations that are pragmatically felicitous, children should adhere to syntactic principles and should reject sentences that do not conform to them. Typically, syntactic and pragmatic principles are at odds only in infelicitous experimentalcircumstances ; in ordinary circumstances , they do not compete. In sum, this sourceof errors in children's behavior is artifactual and should not serve as the basis for a model of the language apparatusas it functions in most real-life circumstances . Children's performanceerrors may stem from the nonlinguistic demandsof an experimental task. Tasksthat use pictures, for example, assumethat children are able to interpret pictures in the sameway as adults do. We find this assumptionhighly suspect. In severalinstances, we have given alternative tasks to children with the same linguistic materials (e.g., Crain, Thornton, and Murasugi 1987; Miyamoto and Crain 1991). The results from the picture versions of the experiments are consistently poorer than the results from the versions that involve vignettes acted out by the experimenter using toys and other props. We have the subjective impression that children are less proficient than adults are at "parsing" the contents of pictures. Moreover, children dislike the picture task, as comparedto tasksbasedon stories. Pictures are limited becauseonly "snapshots"- not ongoing records- of events can be depictedgraphically. This constrainsthe kinds of verbs that can be usedin test sentencesand the kinds of predicatesthat canbe investigated. For example, individuallevel predicates(which refer to properties that persist over time) such as being intelligent, or being a car-driver, are difficult to portray in a drawing. Therefore, errors that children makein interpreting sentencesin responseto pictures could inaccurately reflect their grammars; instead, errors could result from their lack of a nonlinguistic skill, one that takessometime to master. 15.2 CognitiveComplexity Besidespicture tasks, other tasks can causenonlinguistic errors. Many psychological tasks require building and executing. a variety of cognitive algorithms, only some of which are linguistic. In some circumstances , the complexity of the nonlinguistic com-

Performance Errors

Figure 15.1 Array used to test children the secondstriped ball ."

's

123

interpretation of prenorninalmodifiers. Childrenwere instructed to "Point to

ponents of the task may exceedthe child's cognitive ability . This can happen, for example, if the child must formulate a highly complex responseplan in order to "act out" the meaning of a sentencewith the props available in the experimental workspace. In such circumstances , analyses that have been made within the language module may be obscuredby the complexity of nonlinguistic processes . As with experimental design flaws, errors that derive from this sourceshould not be weighed in assessingchildren's linguistic knowledge or in modeling their language-processing system. In this case, the reasonis that the componentsof the languageapparatusare not permitted to contribute' to the child's behavior as they would normally. Even consistent

nonadult

responses

in such circumstances

cannot

be construed

as evidence

of children's nonadult languageprocessing(e.g., within the syntax). If the goal is to investigate child syntax , then it is important to strip away extraneous sources of

errors such as those due to responseplanning. This is accomplishedby reducing the complexity of the cognitive algorithm (or "plan") the child is required to formulate in the task. Only then can it be determined whether or not the child 's analysis of the

linguistic input is similar to that of an adult.l We can illustrate the point with an actual error committed by children that arose

from the complexity of the responseplan, and not from the children's lack of grammatical knowledge. The error appearedin studies by Matthei (1982) and Roeper (1972), who found that many children failed to correctly execute instructions con-

taining sequencesof prenominal modifiers, including an ordinal, suchas "Point to the secondstriped ball." Having been trained to count objects from left to right, many children who were given the instruction to "Point to the secondstriped ball" in an array like that in figure 15.1 pointed to the ball in secondposition (the object in the array that is secondand stripedand a ball). Adults point to the secondof the striped balls. This nonadult responsewas interpreted by Matthei as evidence that children adopted a "flat" syntactic representation, rather than the correct adult representation. Based on the conviction that children assign the same syntactic representations as adults

do , there

is an alternative

account

of children

' s errors , however

. Children

could

have difficulty forming the cognitive algorithm that is needed to correctly execute instructions in which an ordinal selects and counts members of one subset of objects

in the array, while disregarding the members of another subset of objects. If this nonlinguistic account of children's errors is correct, then children's performance should improve if aspectsof the cognitive algorithm could be formulated in advance, by allowing children to partition the objects into subsetsbefore they were placed in

124

Chapter 15

the array. The pretrial exercisecan be quite simple; children can simply be asked to hand sets of objects to the experimenter, one set at a time. Hamburger and Crain (1984) found that children's performancedramatically improved once they had performed this pretrial task. Becausethe IIerror" of interpretation was almost entirely eliminated, syntax was exonerated as the culprit in evoking nonadult responsesin earlier studies. 15.3 Performance Errors In sentenceprocessing, the linguistic knowledge of both children and adults can be obscuredby performanceerrorssuch as slips of the tongue, repetitions of words, and mispronunciations. Even more relevant to the assessmentof children's knowledge of syntax are errors causedby limitations in computational capacity1 for example, limitations in verbal working memory. Since both children and adults have restricted memory capacity, performanceerrors may arise in circumstancesthat place excessive demands on memory. The Modularity Matching Model assumesas the Null Hy pothesis, however, that the memory capacity of children matchesthat of adults.2 The finding that children and adults show similar patterns of behavior is a good diagnosticof performanceerrors. If limited working memory capacityis a contributing factor underlying an error, then the error can be expected to appearin testing both children and adults. Moreover, errors are anticipated only for constructions that require subjects to maintain linguistic material in the working memory buffer in unstructured form for an unusual amount of time (albeit only tens of milliseconds). Although it is generally believed that errors occur as a function of the length of a sentence, i'length effects" are probably a hallmark of performanceerrors only when the additional lexical material cannot be rapidly integrated into the syntactic representation that is under construction. This leads us to expect length effects with "flat" phrase markers rather than with "tall" ones; length effects are also associatedwith maintaining linguistic material (e.g., a "filler" wh-phrase) in the working memory buffer until an appropriate "gap" is identified. This is presumably why wh-questions involving extraction from object position are more difficult than ones involving extraction from subjectposition. More practical considerationsalso lead to errors. Even in circumstanceswhere the materials and experimental design are not at issue, errors may appear. Errors can be expectedat the beginning and end of an experimentalsessionwith children. Presumably, errors occur at the beginning of a sessionbecausechildren are attempting to figure out the task and to mentally calculatewhat construction or analysisyields the right results. At the end of the task, children may become tired and their attention may begin to wander. Errors arising from these sourcesshould be sporadic, however, appearing only a relatively small proportion of the time. As a rule of thumb, we expect such errors to accountfor 10% or fewer of children's responsesin an experiment. Children's responsesshould represent their competencein the main; only a limited number of errors due to "warm-up" and "fatigue" effects should be manifested. 15.4 ParsingPreferences Another influence on language processing by both children and adults is parsing preferences . Theseare assumedto derive, at least in part, from the architectureof the

PerformanceErrors

125

component of the language-processing apparatusthat accesseslinguistic principles (syntactic and semantic) and resolvesambiguities that arise when more than one representationis consistentwith the input (both the test sentenceand the discoursecontext). Basically, a parsing preferenceis exhibited when the architectureof the system favors one linguistic representationover others. Given the assumptionof the Modularity Matching Model, we should expect the same preferencesto appear in both children and adults (seechapter 12; Crain and Fodor 1984, 1993). To give a concrete example, the " A void Pronoun" principle (Chomsky 1981) can be thought of as a parsing preference. The tendency to avoid pronouns may in fact be just one of the manifestationsof the more general strategy to use "reducedforms" wherever possible (see lasriik 1990).3 In pro-drop languages, such as Italian, the "Avoid Pronoun" strategy reflectsthe preferenceto use null subjectsexcept when an overt pronoun is neededfor emphasisor contrast. In languagessuch as English, by contrast, the adult grammar excludesthe pro-drop option, so the parsing preference to drop subjectNPs cannot be followed, at least according to the Modularity Matching Model . There is evidence, however, that pro-drop is a viable option in the grammar of young English-speakingchildren (Hyams 1986; Rizzi 1993/ 1994; Roeper and Weissenborn 1990).4 According to the Modularity Matching Model, null subjects should not occur in children's productions as soon as the pro-drop parameter has been set to the "minus," non-pro-drop value. Adopting a Competing Factors stance, Bloom (1990, 1993) offers a performance account of the early stage of Englishspeaking children's grammars, the stage at which there are missing subject NPs. Bloom contends that children at this stage have the correct parametersetting (minus pro-drop) but override this for performance reasons. This is a further illustration of the key difference between the Modularity Matching Model and the Competing Factors Model. The difference concernsunambiguous sentences : on the Competing FactorsModel, unambiguoussentencescan be influencedby parsing preferencesand other performancefactors; on the Modularity Matching Model, they cannot (Crain and W exler, forthcoming). So far, we have discussedchildren's errors of performance and possible sources of their nonadult responses . We presented two new kinds of circumstanceswhere children and adults could be expected to perform differently, but where we would not therefore infer that children differ from adults in linguistic knowledge per se; the differencesin behavior could be ascribed(a) to experimental artifacts or (b) to nonlinguistic d'emandsof the ta"sk. For the moment, then, we are able to retain the Mod ularity Matching Model even in the face of certain circumstantialevidenceagainstit . In the remainderof the chapter, we delve further into possible sourcesof differences in the. linguistic behavior of children and adults, and we examine other nonadult responsesby children that are ultimately consistent with the Modularity Matching Model . In the paragraphsthat follow , we considerwhether there is evidencethat children's linguistic behavior is under the influence of parsing preferencesand/or performance factors in ways that the linguistic behavior of adults is not. This would be compelling evidence against the Modularity Matching Model and in favor of the Competing Factors Model. We examine instancesof children's nonadult linguistic behavior. In each case, we conclude that the Modularity Matching Model standsup to the challenge, becausethere are other sourcesof children's nonadult responsesin psycholinguistic tasks. For example, such responsesarise in certain instancesbecausechildren

126

Chapter 15

are languagelearnersand do not yet have the full repertoire of grammatical knowledge that adults do. 15.5 Repetitionof the Auxiliary in YesjNoQuestions Sometimes, children's nonadult linguistic behavior seemsto meet severalcriteria for a performanceerror, but also seemsto resist such an explanation. One such study was conducted by Crain and Nakayama (1987), who attempted to investigate children's adherenceto the structure-dependenceconstraint. The task was designed to elicit yes/no questionswith embeddedrelatives like "Is the boy who is being kissedby his mother happy?" In this study, several of the younger children produced some ungrammatical questions with an extra auxiliary verb , such as (I ).

(1) Is the boy who is being kissedby his mother is happy? In a follow -up study, Nakayama(1987) showed that, in addition to the fact that the errors were made predominantly by younger children, the majority of errors of this type occurred in questions with a long relative clause(containing a transitive verb and its complement); the number of errors was reducedin the child subjects' production of questions with relative clauses containing- an intransitive verb . Also . the~e responsesoccurred more often when children attempted to ask questionscontaining a relative clausewith an object gap than when they attempted to ask ones with a subject gap. This could also be consideredto be a "length effect," since the distancebetween the "filler" and the "gap" is greater in object gap relative clauses.Sucha length effect is also reported in the literature on other populations with notable working memory deficits: aphasicpatients (Grodzinsky 1990) and children with reading disabilities

(Bar -Shalom , Crain , and Shankweiler

1993 ).

In sum, children's productions included the insertion of lexical material, a property that is not characteristicof performancemistakes. On the other hand, the addition of material by children was influenced by phraselength, a property of sentencesthat is correlated with performancemistakes. Our view of the matter is this. Children are not inserting linguistic material to make processingeasier. Rather, they are responding to increasesin processingload by reverting to a grammaticalrepresentationthat is consistent with Universal Grammar but is not part of their current grammatical

system (cf. Lebeaux1988). Further evidenceof this camefrom an elicited production study where children were attempting to produce bound variable questions such as (2), but instead produced nonadult forms such as (3). Although (3) has completely grammatical counterparts in Italian , this use of a null subject is not grarrlmatical in

adult English (seeThornton and Crain 1989; Thornton 1990). (2 )

Who thinks he has the best smile ?

(3)

Who thinks

has the best smile ?

Children's deviant questionslike (3) appeared, however, only in their initial attempts to produce questions. This "warm-up" effect is known to limit performance, just as is the effect of "fatigue." We contend that children's searchspaceof grammaticalrepresentations is sometimes influenced by performance factors, but that this happens becausethey are still in the throes of languagelearning, where these alternative representations are available and may emerge when sentences impose processing de-

PerformanceErrors

127

mands. It should be kept in mind, however, that processing demandsdo not cause children to adopt grammatical analyses that are not made available by Universal Grammar. Moreover, the sameprocessingdemandsare experiencedby adults. Adults do not respond in the sameway as children do, however, presumablybecauseadults no longer have accessto earlier grammatical representations. For adults, then, processing difficulty must be evaluated using more sensitive experimental techniques, such as the measurementof eye movements (see, e.g., Frazier and Rayner 1982; Tanenhausand Trueswe111995 ; Ni , Crain, and Shankweiler1996).

Chapter 16 Methodological Preliminaries

Everyexperimentbegins.with two hypotheses : the experimental hypothesisandthe null hypothesis. The experimental hypothesis will be labeledHI . In the experimentsthat we will discuss, the experimentalhypothesis will be that children's grammaticalcompetenceincludesthe principles of the theory of Universal Grammar. The experimental hypothesis is contrasted with an opposing hypothesis, called the null hypothesisand

labeledHo. The null hypothesiswill be that childrenlack the principleof Universal Grammarunder investigation. This use of null hypothesis , which we put in lowercase,

describesthe expectedoutcomeof an experimentif the experimentalhypothesisis not confirmed; it should not be confused with the uppercaseterm Null Hypothesis , which describesthe set of working assumptionsthat follow from one's specificmodel of the languageapparatus. In chapter 2, we introduced the notion of constraints. Many of the experimentswe discuss are designed to investigate children's knowledge of a constraint. As mentioned in chapter 2, the absenceof a constraint entails that children's grammarswill permit meaningsthat the target grammardoes not permit. For example, one linguistic constraint prohibits He thinks the Jokerhas the bestsmile from having the meaning that the Joker thinks he, himself, has the best smile. The constraint does not prohibit the interpretation according to which someoneelse- say, Robocop- thinks that the Joker has the best smile. In testing children's knowledge of the constraint, then, the null hypothesis (i.e., the hypothesis that children lack the constraint) is that such a sentence has both the interpretation that the (adult ) grammar allows and the inter -

pretation that is ruled out by the constraint. In short, the null hypothesis is that such

a sentence is ambiguousfor children,but not for adults. The experimental hypothesis is that children's grammarsrender the test sentencesunambiguous, as adults' gram-

marsdo. Returningto our example , on the experimental hypothesis , children'should not permit the sentence He thinks the Joker has the best smile to mean that the Joker thinks he , himself , has the best smile .

16.1 Avoiding Type I Errors

When designing an experiment, it is important to make sure that the design guards against concluding that the experimentalhypothesis is. correct (that the child knows the constraint) when, as a matter of fact, the null hypothesis is correct (the child lacks the constraint). One of the main goals of any experimentis to avoid conclusionsof this kind, which are called type I errors. To avoid type I errors, the experimental design

mustconstitutea conservative test of the experimental hypothesis , HI . A test is conservative to the extent that it reduces the risk of concluding that children adhere to

130

Chapter 16

a constraint when in fact the constraint is missing from their grammar; that is, a good experiment defends the null hypothesis, Ha, by stacking the cards against the experimental hypothesis, wherever possible. To be conservative , care must be taken to avoid tasks that present test sentences in

situations where subjects will give correct answers for the wrong reasons. Type I errors contrast with type II errors, which occur when valid hypotheses are rejected.

Obviously, makingthis type of error canhurt one's career , but it is lessdamagingto

the field of inquiry . -

We can illustrate the choice of the null hypothesis and how type I errors can be avoided using the truth value judgment task. Recall that this task can be used to investigate whether children's grammarscontain linguistic constraints on the meanings

that canbe assignedto sentences . Whenchildren's knowledgeof constraintsis being tested, the null hypothesis is that children lack the constraint. By contrast, the experimental hypothesis is that children know the constraint. According to the null hypothesis, children should be able to assign the kind of meaning that the constraint rules out. Therefore, sentencesgoverned by the constraint should be ambiguous for children, whereas they will be unambiguous for adults. The truth value judgment task is able to pit the null hypothesis against the experimental hypothesis. Taking the null hypothesis as the starting point, the experimenter designs a context to ensurethat the meaning ruled out by the constraint is available for each test sentence. In the context , certain aspects of the story corre -

spond to the meaning of the sentencethat is illicit for adults, and other aspectscorrespond to the alternative meaning- the one that both children and adults license. On one meaning , the sentence is true, and on the other meaning , it is false. The

children's task is to indicate whether they accept or reject the test sentence. They simply respond by saying "Yes" or "No" depending on whether they judge the sen-

tence to be true or false .

But which meaning should be associated with the IIYes" response (i .e., true in the context ), and which should be associated with the IINo " response (i .e., false in the con -

text)? Following the counsel to Ilbe conservative," the meaning that eludes the constraint should correspond to the " No " response. Only this meaning should be accessed

by children who know the constraint. Associating the I'No" responsewith this meaning ensuresthat the test of the experimental hypothesis is conservative; that is, it reducesthe risk of m~king a type I error. Becausesubjectshave a bias to say I'Yes" (i.e., to accessthe interpretation of an ambiguous sentencethat makesthe speaker's statementtrue), the experimentalhypothesis- that subjectsknow the principle under investigation - should be associated with the I'No " response. If the correct answer

correspondsto a IINo" response, then a correct responseis more compelling evidence for the experimental hypothesis than it would be if it were associatedwith a I'Yes" answer

.

There are other ways to stack the cards against the experimental hypothesis, in order to avoid type I errors . One is to associate the final event in the context with

the meaning that is ruled out by the constraint. As just noted, this meaningis already associated with the "Yes" response; the sentence is true on this meaning . If this

meaning is also associatedwith the final event, then it should be relatively easy for children to access , if their grammarsdo not rule it out. Therefore, adopting this as a research strategy further lessens the likelihood of committing type I errors .

Methodological

Preliminaries

131

16.2 GeneralFeaturesof ExperimentalTasks We conclude part I by pointing out several other features of experiments that are important to keep in mind when working with children. Some of these featuresmay seemquite banal, but we have found them to be important for the successof a study. Foremostamong them is the child's comfort in the experimentalworkplace. With any kind of experiment involving young children, a key ingredient of successfuldata collection is to make sure that children enjoy the time they spendin the lab. In our view, higher priority should be given to the child's enjoyment than to one's own experimental goals. A child who does not enjoy the time spent in one experiment will not be a willing participant in future experiments. The "fun factor" is especially important in elicited production studies, sincechildren are being askedto expressthemselvesusing constructionsthat are not always totally familiar to them. If children are comfortable in the experimental setting, however, they want to participate over and over in these'Igames," for sessionslasting 30 minutes or more. Attention to children's well-being also enhancesparents' and teachers' confidencein the researchprogram, fostering the kind of productive working relationship that is most conducive to research. So that children will find an experiment optimally interesting, it is crucial for the experimentersto know the children well. Generally, the experimenter should have interacted extensively with each child participating in the study in the classroom setting and on the playground (or at the child's home). These interactions take place in the weeks before children are invited to I'play the game." If the experiment requires taking the child to an unfamiliar researchroom, it is often worthwhile to invite the child to preview the room and the IIgame." The child can be given a short introduction to the puppet and the game. This practice sessionshould be fun, so that the child will want to return ~o play in the real testing session. For a child who is hesitant to leave the familiar classroomsetting, it is sometimeshelpful to invite the teacherto accompanythe child to a practice session. Eachchild should receive training on the task. It is best to have a separatetraining sessionfor each child, using stories and sentencesthat are closely related to those that will be used in the real session. On occasion, we hold group training sessions , as part of the child care facility's summerprogram, for example. Children trained in thesesessionsare briefly retrained individually before the main testing sessionsbegin. Children who are unable to understandthe task, or who are unwilling to participate in the study, are excusedfrom further participation. Usually only a small percentage of children (less than 10%) do not participate, but a record should be kept on these children, with any relevant information. Our experienceindicatesthat it is often valuable to recruit a friend or colleagueto help with the experiment. This allows a much greater degreeof experimentalcontrol. Usually, one experimentermanipulatestoys and prop~, and the secondexperimenter plays the role of a puppet. The main reasonfor including a puppet is that children ate often more willing to interact with a puppet than with an adult (even though they know that an adult is posing as the puppet character). This means they are more likely to be verbal in an elicited production task and are more willing to offer corrections in a truth value judgment task. Including a puppet also makesthe elicited production task and the truth value judgment task similar, so that children associatethe two tasks.

132

Chapter 16

There

is

sion

a

more

mentarily

the

jokes

,

child

gets

prepares

the

the

child

'

s

does

not

her

the

,

make

the

the

closer

ward

the

adapted

to

will

better

.

Or

items

if

it

ups

if

gets

it

is

in

also

,

way

16

. 3

In

Ie .

k

"

)

,

be

en

,

the

the

puppet

the

may

-

. 1

keep

then

of

and

trials

with

puppet

amount

certain

the

puppet

also

child

.

monitor

scorekeeping

it

and

to

,

Sometimes

children

and

for

. g

. ,

" If

-

every

troll

the

of

.

,

crossed

over

taste

eat

two

do

push

,

-

just

designs

the

bridge

,

it

reminders

where

a

be

example

to

experimental

example

can

can

and

-

-

the

make

puppet

re

re

reward

For

volunteer

rewards

certain

for

child

to

can

that

the

items

reminder

puppet

system

even

designs

the

the

that

the

the

for

on

the

say

feed

food

pay

as

as

to

The

lito

food

serve

to

for

,

)

will

reluctant

.

child

or

( plastic

want

fun

salt

the

wrong

use

reminder

puppet

thing

item

are

more

the

wrong

we

they

the

)

suggest

is

so

which

because

give

reward

the

children

be

( pretend

experimental

( e

,

If

child

says

choose

.

little

it

the

it

admonition

a

.

have

when

to

if

on

often

may

want

we

suggest

and

show

the

to

,

facility

protocols

We

.

It

can

experimental

then

feed

Kermit

have

are

own

written

aides

"

with

to

list

down

typical

"

to

see

of

.

Some

the

children

of

truth

and

administrators

,

simply

a

It

must

research

interacting

about

have

about

although

It

children

.

into

may

researchers

protocols

.

video

that

and

the

The

methods

stories

they

the

These

working

.

fact

disappear

teachers

for

be

trial

experimental

questions

protocols

the

with

will

experimental

witnessing

the

answer

:

rapport

you

underscores

these

experience

talk

and

establishing

where

and

their

to

,

its

not

a

.

discuss

.

facility

using

their

relate

meetings

will

of

studies

experiments

we

of

care

video

and

later

volunteer

attend

importance

methodology

teachers

conducting

techniques

task

child

in

them

.

to

a

participating

hear

puppet

the

them

for

production

the

experimental

to

and

tips

experimental

elicited

at

enjoy

mysterious

practical

the

enough

to

; monstrates

few

of

the

emphasize

thoroughly

research

a

both

administrators

useful

a

so

task

and

seem

it

one

for

cannot

teachers

room

reminder

the

and

relevant

judgment

may

the

for

we

to

then

they

conditionals

section

are

We

limited

child

Tips

this

value

, "

along

the

the

helps

game

story

of

trial

.

Practical

points

a

II

child

one

the

-

tells

the

puppet

the

-

en

who

up

The

the

role

do

reminder

allowed

can

certain

are

coo

,

only

for

simplify

sentences

as

child

but

,

of

puppet

that

make

.

to

the

is

-

mo

experimenter

the

from

result

that

part

attends

feed

is

experimenter

right

a

to

child

thing

the

wrong

the

it

administering

the

is

The

serve

ask

but

,

give

than

make

can

this

,

love

more

puppet

,

children

and

wrong

puppet

During

playing

for

,

.

the

subject

Children

which

said

.

The

ses

this

entertains

props

changes

An

,

also

stories

role

.

and

character

away

.

the

,

Instead

interesting

puppet

trial

the

s

audiocassettes

fun

reminders

and

puppet

as

'

child

.

makes

story

.

thing

. "

and

well

story

change

on

game

right

attention

wards

these

so

The

next

puppet

the

an

This

a

the

puts

the

experimenter

,

and

as

the

child

the

equipment

To

Each

the

,

,

,

.

up

quietly

of

on

.

set

from

sit

role

so

for

trials

on

hand

and

however

to

just

experimenter

needed

progress

to

the

,

,

time

away

child

assuming

shy

puppet

takes

attention

first

are

a

often

the

is

between

in

entertain

saying

,

the

that

is

free

audio

the

while

focused

notetaking

to

up

,

attention

With

,

using

It

ask

trials

mixed

trial

.

' s

to

between

props

a

for

child

have

interludes

Once

the

experimenter

easily

scenarios

with

the

would

between

joys

reason

for

alone

gages

a

important

takes

working

d

more

enjoyable

our

research

.

sometimes

amount

to

com

-

133

Preliminaries Methodological man

courtesy

the

teacher

area

. Others

ters

in

It

:

stories

not

are

later

on

into

the

It

is

you

" ),

to

.3 . 1

to

lion

available

ent

themselves

puter

flow

way

. And

to

have

and

time

children

for

example

elephant cream on stories

,

it

needs ,

be

correctly

)

tape

to

re

,

listen . It

a

working

to

puppet

tape

to

is

very

has

been

the

child

before devote

their

lead

the

child

into

the

room

( IIA

my

has

full

role

moment

the the

for

of

puppet

child

the

ex

to

in

tran case

the

-

see

later ,

enters

events

through ,

come

catalogued the

tell

to

or

-

the

room

,

the

.

,

ice

base

might cream we

the

characters

,

on

have

the

of

to

the

what

a

grounds

child

would

that against

world events

.

is events

If

the need

in

or

view

not

It

the

that

act

.

the

.

on

belief

and

that or

hay

base

that

they

,

that

an

but

not

ice

their

Recall

., So

said

,

responses

Model .

their

common

characters

sentence

Matching

novel

heavily

among

concern

not

rarely

create

a

good

do

story

be

peanuts

story a

are

to

children

is

-

some

children

plausible

a

ses

Children same

to

( so

in

. It

lean

seems

eat

cause

-

simple

resolved

so

children

reject

Modularity

-

com for

because

the

relations

that

of

pres

experimental

potential

semantically

elephants the

the

,

the

.

toys

experience

feature

that

sentences

on

that

argued

implausible

assume

the

and

perhaps

own

stories

gives

of the

gets

should

all also

implausible

speculate

' s in

all

props

front

short

for

-

child

Goldilocks

to

be

that

interest

the

layout

and

assemble

should

characters

characters seem

It

to

appealing

.

action

' s

of

to

planned

child

certain

responses depict

.

some

to

helpful

in

people

for

characters

missing

some

ideas

sitting

time are

stories

the

and

comes

near

responding

their that

I knowledge

in

often

while

story

include

is

people

world

, one

of

to

how

of

is

to

,

of

ones

easily

Sometimes

, it

it

good

capture

about

. Many

will

a

it

more

.

abstract

needed

all

because

comes

inspiration

the

combinations

sentences

containing

that

can

combinations

the

needs

, Robocop

the

. However general

a

child

equipment

accidents

( i .e .,

and the

by the made

cause

team

the

tasks

to

combinations

ate

-

- quality on

being

brought

to

in

. When of

to

time

Superman

of

reject

a

inspiration

in

ingredients

stories

will

the

for

frustration

ideas

unusual

that

is

is

is

interested

interesting

stories

some

unusual

odd

knowledge

From

spawns

,

to

action

up

if

Using

name

experimental

up

,

preconceived

bothered

high

attention

about

not

playing

searching

Making

fantastic

mix

. child

in

. When

same

the or

using

experimenters

child

can

the

. Sometimes

enough

the

funny

idea

use

several

' s

tape

will

as

the ' s

mind

experience

leads

are

At

When

audiotape

her

, inevitably

) .

the

experimenter

keeping

But

.

there

sion

charac

.2

child

the the

for

.

frequently

that

is

,"

toys

There

activities

Stories

not

experiment

informing

as

bothered

properly

work

.

the

slipped

stories

do

the

child

,

curious

microphone

that

experimenters

remind

duty

others

stories

,

outside

monsters

usually

a

they

the

so

session

that

to

up

than

the

both

Constructing

Making

child

the

their

are

working

that

child

not

they

that

is

and

mention

has

is

to

casting

focus

that

so

room

with

are . If

explanation

down

recorder

guarantee

name

not

every to

table

equipment

research

that

and

puppet

16

tape

should

' s

taped

with

Children

the

the

the

experimental

scription child

be

the

important

perimenter

. on

with

interacting

successful

each or

,

experimenters

collected

that

into to

interview

classroom

weapons

session

the

sitting

check

brought

to the

toy

every

being

should

to

to

using

allows

satisfied

. Wires

attention

a

data

usually

inserted

: not

audiotape

microphone

important

permission

returned

subtle

. This

the

and

they

is

to

on

corder

for

has

.

equipment

and

teacher

child

more

essential

reliable

the

the

are

the

is

asking

that

is that

correct ,

on

, the

134

Chapter 16

Modularity

Matching

world

knowledge

relate

to

in

the

test

making

For

thumb

construction

acters

; corresponding

alities " theme

." For

" This within

all

do before

to

not

tell

are

storyline

" trials

not

- up

stories

. The

filler

best

into

resealable

brought of

way

At

handle

keep

. of up

the

the

toys

the

bag . Having

a random

presentation

picks

bag of

of energy enerp , y . Another

you

say

."

" No

the

to

world and

its

we , for

the

toys

one

the

test

can

be

trials

for

the

the

?" also in

we

story .

attest

to to

also

new

to

that

children

-

for

trials

,

the

example

understanding

test

a

as new

' s responses

their

also

make

answer

ensure

reason

the

seen

to

is

correct

, to

plots

are

try

the

-

say ,

." The

, as long

. Children

complexity

can

having

before

wrong

puppet

across

a different

' s picnic

. If the

" Yes " trials

char -

common

out . Children

heard

trials

order

experimenter

that

they

the

of

have

when

each

have

the

. However

satisfied

the

all

remember

found

example

usually

to

; they

. As

for

number

should

turned

have

structure

same

a dinosaur

, as

happened similar

about

twice

test

be

let

see

, tell

child

this

,

to

the

of

the

.

avoid

is

asking story

children

that

the

the

experimenter

get

. For

if

they

this to

want

is a yes / no

play

another

up

to

do

stories

finished ,

. At them

of

" fly "

simply

inspect

the

have

a lot

a little

physical

another

story

, children or

to the

ensuring

child

to

question game

long

wants

who

use

for

the

on

a child

. The

children to

in

way

up

is

begin

burden

making

getting

a chance

children ?" Because

is

them child

at the

watch

story by

sort the

step

that

experiment

means

provides

they

away

to

to

clear

toys

to

unfair

is a convenient

of this

has

the

toys

an

children

story

design

the

still , this

another

put next

the

then

after

is

, before

puts

, who

them

. Often to

finding

. If it becomes

help

is the

next

in

, encourage

the

interesting

sit

delay

stories

out , ready

storyteller

handle

to

particular

laid

the puppet child .

session

like

to

is no

narration

choose

hard

to to

of the

child

out

be

disorganized

often

set

for

all

, there

the

if

together

should

. A

the

order

it

way

the part entertain

looks

find

Instead

this

to

the

that

like

. These

during

, children

pitfall

"Would

the

items

should

the

up , the

set

same

recognize

are

things

reminder

in not story

how

trials

experimental

possibilities

and

asked

sentence

same

, each

story

with

filler

trial

toys

, offer

of

of

. In

playing trials to

end

selection

of

four

items the

actions

story

to

the of

four

you

are

test

contain

same

tell

to

all

room

into

the

the

bags

experimental

a story

the

interspersed

really

each

the

each

is identical

should

the

start

a test

children

different

experiments

I'What

plastic

pick

between

these

" This is a story , be identical .

our

the

into

the not

knowledge

general

are

of Stories

to

the experimenter periods between

and

they

general

that

all

, that

of

giving

trials

Presentation

The

overrides

, which

a minimum

perform

a story

be

for

should

repeat

correctly

question

for

often

plot

should

enjoy

stories

, children

or

responding

follow

ning

you

be

seem

seeing will

is ' /No ," then

- they

16 . 3 .2

toys

substituted

sentences

because

, as the

they

items

should

story " or , of course

if

characters

" Filler

their

should

, however will

videotape

if the

use

context

events

mismatch

stories

stories

enjoy

you

even

are

the

, and

demonstration

test

. The

characters

example

Children

quick

. The

, however

is a Halloween the stories will

a story

specific

several

there

so on . It is important that

or

be

that

investigated

design

. So

not

match

should

suggest

being

basic

; and

the

, there

, we

same

stories

should

the

specific

.

experiment

of

of

encounter

, they

about

context

any

, understanding children

sentences

decisions

associated

a rule

Model

; because

see

can

another

:

Methodological Preliminaries

135

story : " Let's do another story ," " I have another story that I know you will like ." Have

a positive attitude, and the session will be more enjoyable for the child and for you

.

As the toys and props are brought out of the bag for a particular trial, the experimenter introduces them and checks with the child to see if the characters are familiar . Either the child or the puppet can be asked, " Do you know what this is?" If the child has his own name for a character , it is advisable

to use that name in the

story and in the test sentence . Or , if the child is not familiar

with

a character ' s name ,

it may be better to use a more descriptive name. For example, if a child has never heard of " the Incredible Hulk ," then it may be better to call him " the strong guy ."

As the toys are introduced, they can be set up in the workspace, ready for the start of the story. Use toys that stand upright- ones with a baseor with large feet. Toys that keep falling down are a distraction and waste precious time, particularly if the child wants to take on the challengeof getting them to stand up. Toys that fall down can also impede the momentum of a story that is underway. Besides, the experimenter needsboth hands to manipulate other toy charactersthat are "talking" in the story .

The experimenter manipulating the toys is the "voice" for all the charactersthat talk in the story. To avoid any confusion about which characteris talking, it is often a good idea to move a characterup and down slightly as it speaks. Experimentersare often tempted to narrate the story. But in fact, the stories are much more dynamic if the charactersdo the talking. Having the charactersdo the talking has another important advantage: the stories don't take so long to complete. For example, as the narrator , you would

have to say , " And then the Incredible

Hulk came into the room ,

carrying a pizza, and he said, 'Would you like some pizza?' " Instead, simply move the Incredible Hulk to center stage, with his pizza, and say, "Would you like some pizza?"

It is frequently useful to write down the order in which events are to take place in each story- particularly at the start of an experiment, before the stories become ingrained in memory.For example, a recent experiment investigating children's comprehensionof questionslike "What didn't every rabbit buy?" included a story about three rabbits going to a spare parts store, where spare noses, tails, eyes, ears, and so

on, were for sale. Three actions took placein the story. The following notes servedas a reminder

of the events and their order :

Sentence:

"What didn 't every rabbit buy ?"

1. every rabbit buys a gold heart 2. no rabbit buys spareelephantears 3. 2 of the 3 rabbits buy a spare tail

At the end of a story, the charactersand props should be arranged so that the alternative interpretations are as clear as possible. We will make this point in more detail in part III, when we discussthe truth value judgment task. There, we will note other aspects of each story that are critical to the experiment . To anticipate , one crit -

ical feature of the experiment is the lead-in, that is, the linguistic material that precedes the target sentence. The lead-in has several purposes . First , it lets the child know that the story is over . Second, it draws the child 's attention to the puppet , rather than to the experimenter who acted out the story . Finally I it establishes any

136

Chapter 16

linguistic antecedents that may be needed for the test sentence to be felicitous on the intended reading (e.g ., if it contains a pronoun ). A typical lead-in , then, would be: "Sammy . I know what happened in that story . That was a story about Grover and Big Bird ." The stories should be told exactly the same way to each child subject (as near as possible). This takes some effort . You should rehearse the stories aloud , either to yourself (before , and as you drive to the children 's school , for example) or to friends . This practice step irons out potential problems and ensures that you will present the stories smoothly during testing . It is always a good idea to test older children before moving to younger subjects. Older children are more patient if you happen to fumble and frequently will tell you what is confusing about a story . Each session with a child should last no longer than 20 to 30 minutes . If children enjoy an experimental task, the time will pass quickly for them and they will want to come back for more . The opposite is also true . If children do not enjoy a task, they will be reluctant to participate in further studies. 16.3.3 Handling the Data As soon as a child 's session is finished , label the audiotape and its container , identify ing the child 's name and age, the testing date, and the topic under investigation . Even by the end of a single session, it is surprisingly difficult to remember which children were tested and the order in which they were tested. This step also prevents careless mistakes such as taping over another child 's session or losing a child 's data. If possible, the audiotapes should be transcribed the same day . This task is much easier if the session is fresh in your mind . The task is also easier if it is done in small doses. The experimenter playing the role of the puppet can take notes, when possible, in addition to audiotaping . A score sheet is needed to record the child 's name, age, birth date, and the name of the child care facility where testing took place; it also pro vides a ready tally of the child 's " Yes" and " No " responses. For experiments using the truth value judgment task, it is essential to have a score sheet so that the puppet knows what target sentence to utter at the end of the story . It is also useful to record any comments of interest that a child may make on particular trials . This helps the experimenter pinpoint the place on the audiotape later . A sample score sheet is illus trated in figure 16.1. Every response by the child should be accepted as a good response, whether or not it was what the experimenter expected or wanted to occur . The experimenter should always be upbeat, so that children do not feel that they said or did anything wrong . Experiments can be, and should be, positive experiences for children . And for you , the experimenter . 16.4 Conclusion

It is easy to get poor results from children. Interpretable results, whether they conform to the experimental hypothesis or not, are obtained only by thinking through all aspectsof the task, by piloting and repiloting (which requiresconstructing entirely new stories), and by practicing. Only through experiencecan the "art" of being an experimentalistbe mastered.

Methodological Preliminaries

Scope of Negation Child'sName : School : TestDate:

137

Experiment

Sammy SunnyHill 11/ 17/ 96

Age: 4;6 Birthdate :5/ 9/ 92

Targets 1. What didn't everyrabbitbuy?

Response ears

2. What didn't everytroll eat?

bananas

3. What didn't everymermaidtakehome? flowers 4. What didn't everyhorsejump over?

chair

Fillers : 5. What did the pig roll in?

mud

6. What did PapaBearride on?

dinosaur

7. What did the spaceman crashinto?

wall

Comments: -paidattention well -said"Everyhorsedidn't jumpoverthefence " (meaning 'noteveryhorse ') Figure16.1 A samplescoresheet

The remainder of the book will provide guidelines for constructing stories for experiments using the elicited production task and the truth value judgment task. We cannot empha"size enough how much effort goes into the design of a successfulexperiment- one in which children base their responseson their grammatical knowledge and not on strategiesthat they devise becausethey are confusedor frustrated. This concludes part I of the book. In part II, we consider the methodological desideratafor eliciting sentencesfrom children.

PART II The Elicited Production Task

Chapter 17 Elicited Production

More to

and rely

more

on

mental

, it

experimental

research elicited

the

context to

transcripts Despite has

the

one

able

in

be

virtues

transcripts

used

to

of

elicit

Elicited

Elicited

sparsely

at

is

in

call

' s experience

In

ever

addition

a

or

years

in

leaves

child

's

tion

, by The

herent

to

sur

-

than can

controlling

that

problem

,

arise

when

in

at -

examining

spontaneous

data

. The

with

production

are

elicited

some

of be

sometimes

avail

production

effort

-

the

task results

in

-

can

are

not

order

longitudinal repeated elicited in

collect

a

course

production

of

often

. For

not

they the

in . It

children

chil

-

enables produce

also

experimenter

single

true

of to

sample

week

established few

a more

.

~s

sponta

-

months

phenomenon may

with weeks of

a

' s grammar

cover

linguistic

many offers

child

that

stages

every

gather session

children

files a

or

, it

of

search

be

overcomes thing

the

searches

of

to

experimental

about

grammatical can

every

the a

have large

one

, if

rarely

obstacle

that

conclusions

important

development

technique studies

allows

solid

sufficiently

sessions

this

only

because

only

overcomes

within

frequently

that

scarce

occur

constructions

technique

is

are

in are

.1

structure draw

interest productions

forms

technique linguistic

. This

possibility

elicitation

comprehension

constructions

. Researchers

to the

linguistic

to

productions

theoretical

' s spontaneous

linguistic

targeted

time

' s spontaneous of

children

speech

collected

children

complex

production

transcripts

open

of

these

involving

the

of

phenomena

production

data

point

speech

searches

spontaneous

elicited

can

particular

This

speech

transcripts

elicited

sample data

neous

is

, productions

difficulties

technique

, but

linguistic

particular

their

, the

Sufficient at

for

gather , in

data

the ,

. The

to

, if

to

. Many

Presumably

that

researchers

preferable

, however

all .

situations

robust

suitable

are

Production

represented

appear

rarely

the

. Longitudinal

time

. This

more

benefits

, a frequent

production

of

of

children

thing

-

still

.

production

a period

one

experi

common

be

obvious

many

' s spontaneous

over

production instances

dren

elicited

. For

are

productions

elicited

to

come

are

.

Why

many

the

, cases

meaning

from seem

has

number

. Less

sentences

. There

in

sentences

would

most

' s intended

children

data

instantaneous

17 . 1

of over

elicit

development

. Fewer

of

eliminates

' s spontaneous

advantage

language

speech

to

contexts , this

a child

children

child

method

, perhaps

importantly

interpret of

many

controlled

. Most

tempting

attempt

production in

carefully

on

' s comprehension

that

elicited

methods in

literature ' s spontaneous

children

investigations the

the

children

of

, because

other

that

of

investigations

prising

be

seems

transcripts

be

.

missed

elicited

. A

produc

-

. the

direct

limitations way

in to

assess

142

Chapter I7

children 's emerging linguistic competence . A comprehension task that relies on "Yes" versus "No 'I' responses by children must take steps to ensure that nonadult patterns of responses by children truly reflect their grammatical knowledge and are not an artifact of the task. One general advantage of production data is that they reveal the child 's grammar without the need to make inferences from l' Yes" and 'IN 0" responses, as in a comprehension measure such as the truth value judgment task. Production tasks tap children 's grammars more directly than comprehension tasks. Correct combi nations of words do not come about by accident. As long as the possibility of imitation is excluded , the successful production of even a single target utterance in an appropriate context is a compelling argument for competence with the relevant structure . Productions are revealing about children 's grammars because they contain so much information about children 's linguistic competence, including (a) word order , (b) which words are grouped together , (c) the appearance of " displaced" groups of words in one position of a sentence and superficially "missing " words in another position , and so on. If a particular sequence of words appears in children 's speech, the sequence is very probably generated by their grammars . It is unlikely that children accidentally put words together in the ways that they do without consulting their mental grammars. It is even more unlikely that children accidentally put words together in exactly the same way as adults do, in the same contexts that adults do, unless they have access to the same grammatical principles as adults do . Therefore , children 's adultlike productions permit us to infer that children and adults have similar grammars, as expected on the Modularity Matching Model . 17.2 What 's the Problem? Despite the advantages of the elicited production paradigm , it is easy to understand why the technique has for the most part remained on the shelf. As Ferreiro et al. (1976) remark, the problem is identifying a uniquely appropriate context for the lin guistic phenomenon of interest : Experiments in which children are asked to produce sentences seem on the whole to be free of the hazards of comprehension studies. However , once again, we come up against an obstacle that is due to the nature of language itself ; how to construct a situation that will obligatorily give rise to a certain sentence pattern ? No such situations exist : thanks to the very rules that make language what it is, perfectly adequate and grammatically correct descriptions in many different forms can be given for any event or situation . (p. 231) In this part of the book , we will provide instructions for devising contexts that are uniquely felicitous for several linguistic constructions , including ones that have significant bearing on learnability in the absence of linguistic experience . The trick will be to ensure that the contexts used to elicit sentences from children are not amenable to alternative ways of describing the context . We 'will prove that it is in fact not im possible to construct uniquely felicitous contexts . This chapter contains some general remarks about the elicited production task and a review of previous research (also see Thornton 1996). First, though , we consider the place of production data in the broader framework of the Modularity Matching Model .

Elicited Production

143

17.3 Productionand Modularity Matching The expectationsof the Modularity Matching Model extend to children's productive abilities. As noted in part I, the Modularity Matching Model makestwo fundamental assumptions. The first is that the architectureof the mental faculty for languag~ processingis modular. The secondis that the language-processingsystem of children is essentiallvthe sameas that of adults; more specifically, the principles within any submodule of the language apparatusare the samefor both children and adults. Maximizing the similarities between the cognitive systemsof children and adults helps to explain children's universal mastery of language- the fact that all children success -

fully convergeon an adult grammardespiteimpoverishedinput anda greatlatitude in experiencewith language. To the extent that the cognitive mechanismsof children and adults are similar, the problem of learnability does not arise. On the Modularity Matching Model, the similarities between children and adults - in grammaticalknowledge and in the processingmechanismsused to accessthat knowledge- clearly apply to the language production system. There will be differencesin the productions of children and adults, and these will have to be explained. As illustrated in chapter5, children's nonadult productions are a major source of evidence for the Modularity Matching Model and against one version of the Competing Factors Model, the Input Matching Model. For example, many children produce negative questions with two auxiliary verbs, such as "What did he didn't eat ?'! (Thornton

1993 ; Guasti , Thornton

, and Wexler

1995 ). Since such sentences are

not grammaticalin the target language, it cannot be argued that children are mimicking the input, parroting back phrasesthat adults produce. Rather, such productions exposea differencebetween the child's grammaticalknowledge and that of an adult. On the Modularity Matching Model, the explanations for observed differences between children and adults are quite limited. The Modularity Matching Model requires nonadult constructions to be compatible with Universal Grammar and learnable from positive evidence. Chapter 5 examined a number of children's nonadult

productionsthat canbe explainedby invokingprin~iplesof the the,ory of Universal Grammar. Chapter 15 discussedthe kinds of errors that can arise owing to the nature of the performancesystem. Finally, chapter 13 showed that nonadult responsesby children sometimesreflect properties of the learning mechanism: the LAD . The principles of the LAD include the SemanticSubset Principle, which determines which hypothesis children initially formulate for sentencesthat are ambiguousfor adults. Once children have achievedthe full grammatical repertoire, they choose among the different ways of expressing a message using the same parsing strategies as

adults. According to the model describedin chapters 11 and 12, both children and adults adopt "least effort" (minimal commitment) strategiesin resolving ambiguities, based on the complexity of the discourserepresentationsassociatedwith the alternative readings of the sentence. It is reasonableto assumethat least effort strategies are also employed in production. When there are severalways to expressthe same message , the "simplest" meansof expressionis chosen. For example, children favor questions with extraction from subject position , as in (I ), over ones w.ith extraction from object position , as in (2)! when both question forms express the same message.

We will return to this preferencein section 17.4. (1) Which bug got steppedon by the elephant? (2) Which bug did the elephantstep on?

144

Chapter 17

In order to assessthe full linguistic competence of children using the elicited production task, the experimentermust devise situations that demandlinguistic constructions that correspondto specificcomplex discourserepresentations,perhapssituations that children only rarely witness. This is one of two main design featuresof an elicited production task. The other is the construction of contexts that are uniquely felicitous for the structure under investigation . This enables researchers to evoke lin -

guistic expressionsthat children could otherwise avoid, in favor of "simpler" means of expression .

17.4 Design Featuresof the Task

The elicited production method is appropriate when the question is the sentenceform that children associatewith a particular meaning. As in the truth value judgment task, a situation

is acted out with

toys and props

in front

of the child . The situation

is

associatedwith a specific sentencemeaning. In addition to acting out a context that correspondsto ! this meaning, the experimenter produces a "lead-in" statement. The lead-in elicits the sentence form that children associate with the meaning (in their

grammars) that corresponds to the situation that has been acted out by the experimenter. There are three observablephenomena: the situation, the lead-in, and the child's productions: (3) Input (situation, lead-in) Output (sentence) The input side of the equation in (3) is under the control of the experimenter . The

elicited production task enablesthe experimenterto control the meaning that is to be associatedwith the child's utterance, by presenting a particular scenarioon eachtrial of the experiment . Then , the lead-in instructs the child to report the situation to a third party , often a puppet who was not able to see the story that was acted out .

As a concrete example, suppose that the research question is whether or not children can produce passive sentenceswith a "by phrase." After careful pilot work, the experimentersettles on a design that embedsthe passivesentenceinside a question . The child is instructed to pose questionsto a puppet. In the context presented to children , the question that best conveys the intended message (for an adult ) is a 'Jpassive question with

, on by the -, ele-- -a by phrase, such as '~ hich bug got stepped phant?" (Crain, Thornton, and Murasugi 1987). In order to elicit this passivequestion, the experimenter devises a situation that includes two bugs, an elephant, and one other animal say- a-walrus. The basicplot of the story is that the walrus accidentally steps on onebug ~and the elephantaccidentally steps on the other. (The story would

be embellished to make it more interesting for children .) After the story , the ex-

perimenter addresseschildren with the following lead-in: (4) Experimenter In that story, the elephant stepped on one of the bugs, right? Ask the puppet which bug. Even at this point, there are severalpossible ways for children to form the question. The child could ask an object extraction question rather than a passive question , for

example, "Which bug did the elephant step on?" The child could even reproducethe experimenter's lead-in: "The elephant stepped on a bug. Which one?" The best way to find out if children will respond with these alternative questionsis to conduct pilot

Elicited Production

145

studies. It turns out that in the present situation , a few children ask object extraction

questions, but most .3- and 4-year-old children consistently produce get passives'with a by phrase, suchas "Which bug got steppedon by the elephant?" It is important to note that children do not ask the simple question fragment "Which bug?" in responseto the situation;lead-in we have described. This question is infelicitous becausetwo bugs were stepped on. The by phraseis necessaryto differentiatebetweenthe bugs that were steppedon. Apparently, 3- and 4-year-old children are well aware of such pragmatic felicity conditions, at least enough so that they can easily be coaxed to avoid violating them. This underscoresthe view that young children have even complex semantic/pragmatic principles under their belts. This shows how the elicited production task can enable the experimenter - - -- - examole - - - --- - . L to evoke sentencescorresponding to syntactic structures that occur only- rarely in children's spontaneousspeech(and possibly in adult speechas well). A context that is uniquely felicitous for certain complex sentencesmight be quite exotic in day-to-day conv,ersation. Moreover, ordinary contexts might not be uniquely felicitous for a

given construction , such that children may avoid using the construction and choose

an alternative, simpler meansof expression. By presenting situations that are uniquely felicitous for the construction, elicited production can deepen understanding of the full extent of children's grammaticalknowledge. The techniqueof elicited production works well for children about 3 years old and

older. With effort, it canbe usedwith many childrenas young as 21. For children younger than this, however, it is difficult to maintain the appropriate degree of ex-

perimentalcontrol to reliablyevokeconsistentproductions ; for thesechildren, some compromise between elicited and spontaneousproduction data is necessary. Alter natively, comprehension techniques can be used, such as the preferential looking paradigm (see, e.g., Gleitman 1990; Hirsh-Pasekand Golinkoff 1996; Naigles 1990). Three kinds of experimentsusing the elicited production techniqueare describedin part II. Two of the three experimental designs examine children's knowledge of linguistic constraints on the form of sentences . Both of these designs demonstratethat children fail to produce certain linguistic forms. On the Modularity Matching Model, this failure is expected becausechildren know the relevant constraints. This means that the experimentalhypothesis is supported by "null" findings: the absenceof some behavior that might be expectedto occur. It is important to point out the pitfalls in inte~preting negative findings. The problem is that there are often alternative explanations- in addition to the experimentalhypothesis- for the absenceof specific behaviors. The experiment could be flawed, there might not be enough subjects or trials, parsing strategies could militate against the production of certain linguistic forms , and so forlh .2

Fortunately, most of the obvious alternative explanations of the findings can be handled in the experimentswe describein part II. Chapter 17 introduces one way in which the elicited vroduction task can use the absenceof a certain kind of production to argue for children's knowledge of linguistic constraints. The particular constraint ~

at issue is the one that prevents contraction of the verbal elements want and to to

form wannain certain linguistic environments. As noted in chapters2 and 3, the ban against wanna contraction applies in questions like (5); as (6) illustrates, however, contraction is permitted in other linguistic environments. (5)

a. b.

Who does Arnold *Who does Arnold

want to cook breakfast ? wanna cook breakfast ?

146

Chapter 17 (6 )

To

study

a .

Who

does

Arnold

want

b .

Who

does

Arnold

wanna

children

through

the

Elicitation

of

structions

that

In a

in

(5 )

another

study

;

the

* ls

Suppose of

answer

,

the

like

adult

(8 )

The

do .

questions

like

occur

So

far

assess

,

hard

speech

will

children

' s

can

,

is

a

in

explain constructions

and

ones

useful

tool Model

successfully

in

this ,

produce ,

with

the

do

such

Does

the to to

input

pro

-

. Presum

The

-

use

that

upon

show

-

not

in

The

is

compatible

.

occur

children

On is

them

's

Universal production the

Modu

that

forms to

to

Model

of

18

( 7 ) .

establish

to

elicited

linguistic applying

to

expected

chapter

the

like

from

chapter that

make

task

principles

.

like

Matching

core

in

to

task

constructions

hypothesis

conditions

the

-

predicted

questions

are

regularity

will

are

suffice

Modularity

linguistic

en attempt

production

of

that

questions

forms must

clear

who

nonadult

other

linguistic

produce

elicited

. The

with we

not

the

it

the

children

,

of

make in

nonadult

prohibits

heavily

Grammar limiting

' s

tempting

children

should

that

acquisition

of

experimental

same

5

constructions

as

in

the

(8 ) .

encounter

They

uses .

draw

,

is

linguistic

is

out

acquisition

Universal the

.

that

produce

regard

,

children

( 7 ) .

It

expect

their

chapter

that

the

absence

adults

, 19

( 7 ) .

like ?

to

( 7 )

turns

language

constraint

that

that

in

they

it

fact

language

the

in

forms the

linguistic of

reason

that

constraints

produce

models

(6 )

example

in

in

questions

to

case

of

in

chapter

forms

hypothesis

not

Model

and

contract

. For

illustrated

-

?

concerned of

fact

to

Matching

contexts

the

has

produce

good

utterances

models

knowledge

especially ,

some

no

nonadult ,

know

certain

pressed

Grammar

priate adults

to

discussion

to

many

observation

,

are

bald

present

is

con their

.

in

nonadult

occur

corresponding

the the

fact

questions

Matching

in

produce

that

children

according

larity

(8 )

the

children

Input

rod

described

certain

experimental is

sleeping

mimic ,

according

argument

the

simply

This

of

there

question

is

of

,

in

?

the

such

the

Moreover

.

since

measuring

(6 ) .

which

constraint

,

like

contract

to

the

them ones

against

willingness know

no

not

bald

matter

that

who

of

however

that

of

man

not

vironment

to

the

is

a

is

and

to rod

constraint

should

support

( 7 ),

discussion

children

as

grounds

form

Is

,

there

(5 )

willingness

measuring

they

absence

that

sleeping

not

,

encourages like

. Their that

different the

form

who

the

questions

a

a

evaluated

design

productions on

be

to

linguistic

do

"

can

experimenter

' s

as

evidence

adherence

man

such

" No

ably

task

the

(5 )

serves

? ?

questions

children

. This

for for

the

both

establishes

anticipates

of

,

produce

convincing

breakfast breakfast

' constraint

to

experimental

' s

children

absence

( 7 ),

of

kind

(6 )

like be

hypothesis

( 7 )

is

questions would

children

experimental

duce

like

cook cook

the ,

contraction

type

of

speech

to design

permit of

not

adherence

questions

productions but

' s

experimental

to

in as

-

children

apply

appro

to

.

This ends our general remarks on the elicited pr~duction technique. The next section begins with a discussionof previous researchand briefly sketchessome variations in the applications of the task, used to addressparticular issues. The section ends with an outline of factors that should be taken into account when planning an elicited production experiment. The chapters that follow present several elicitation experiments; in the course of these chapters, we discusssolutions to problems that one may encounterin using this methodology.

Elicited Production

147

17.5 PreviousResearch The elicited production task has been in service in one form or another for over 30 years. An early and much celebrateduse of the task was an experiment by Berko (1958) investigating children's morphological knowledge. The aim of this study was to assesschildren's ability to create new linguistic forms, ones that follow from the application of rules given by the theory of Universal Grammar. To this end, Berko introduced children to novel words, which they could not possibly have heard before. Children between the agesof 4 and 7 were shown pictures of an object, such as a cartoon bird, and were told, "This is a 'wug.' " They were then shown a secondpicture with two tokens of the sametype of object. They were told, "Now there are two of them." To' elicit the target responses , children were asked to complete the experimenter's carrier phrase, "There are two ." The carrier phraseelicited the correct form, "wugs," from most children. This was interpreted as evidence that these children had internalized the grammaticalprocess(e.g., a rule) for supplying the plural ending to nouns. Becausechildren produced the plural forms of novel words, it was inferred that they were not basing their productions on the input. Similar results were found for a number of other grammatical morphemes, including tense marking and possessivemarking. In other words, the findings were taken as evidence against the Input Matching Model . In any event, the experimental results clearly argue against the view that children's productions of real plural nouns are the result of imitation. Another early use of the elicited production task was reported by Bellugi (1971), who used it to probe for children's knowledge of subject-auxiliary inversion in positive and negative questions. One experimenterused a puppet who respondedwith the "quavering voice of an elderly puppet." The lead-in devised to elicit wh-questions was an indirect question. Questions like the following were elicited at various times throughout the experimentalsession: (9)

Experimenter Adam, ask the Old Lady what she'll do next. Adam 0 ld Lady, what will you do now? Old Lady I'll fly to the moon.

(10)

Experimenter Adam, ask the Old Lady why she can't sit down. Adam Old Lady, why you can't sit down? Old Lady You haven't given me a chair.

Notice that the lead-in preservedthe word order of the declarativecounterpart to the wh-question, giving the child no clues about subject-auxiliary inversion. Therefore, the consistentproduction of inverted subjectsand auxiliary verbs by children would have provided compelling evidencethat this processis representedin their grammars. As the examplesshow, Adam failed to invert the subjectNP and the auxiliary verb in negative questions, but be did invert these elements in positive wh-questions (see Pinker 1984 for discussionand for further empirical details). Having fallen out of fashion for some years, the elicited production task has recently resurfacedin studies of other aspectsof children's grammars, in the areas of both syntax and semantics.Here is a partial list of the grammaticalproperties and syntactic structures that have been studied: structure-dependence(Crain and Nakayama 1987; Nakayama 1987); the wanna contraction and that-trace paradigms (Thornton 1990; Crain 1991); object agreementin Italian (McKee and Emiliani 1992); passives (Crain, Thornton, and Murasugi 1987); subject-auxiliary inversion (Erreich 1984;

148

Chapter 17

Sarma

1991

);

negation

in

( Schaeffer );

( Crain

Guasti

et

1994

);

, al

and

.

modi6ed

the

task

with

a

The makes

about

task

into production

and

rej

.6

the an

this

target

lies

accessing

and

We

effectively children

the

1996

,

reasons

as

truth

" Point

to

. . ./ / ) /

with

toys

the

value

value

that

of interact

out that

the

establish

variations

acted

truth

really

they

statement

the

are sometimes

which

( such

turns

to

there must

/ most in

scenarios

using

;

Cairns

) .

game

a

in 1990

and

task

correct

clauses ( Labelle

common

something

judgment

judgment

children

.

puppet

task

are

,

accepting

.

considerations

skill

of

all

a

children

important the

know

because

conducting

involving is

that

a

use

in

a

because words

child

a

a

too

that

snail

or

game

,

In

so

,

a

high

experience

issue

that

children experiment

addition

often at

require

elicitation

.

they

well

they

successful

structure

subjects

shy

about

the

The

to

talk

.

solicits

research

strategy

" -

the

volve

the

puppet

.

In

the

difficulty

To

.

It

they

child

,

course

experimenter

,

planned

. not

experimenter

that

won it

' t

is

tries just

an the

and

it

When to

and -

the

the

child

talk

to

will

level

of

difficulty

is

only

,

in

therefore

critical

comply

isn

't

in

with

their

is

the

The child

kids

out

.

position

have

This

to kind to

tell

a

reason

of perform

corne

should

the silences

be

know

' s

avoided

help

of ,

is

used

" Yau to

in

with

by

the

session structure that because

-

the

have

thinking

target

.

this

sometimes

experimental

with

child

ploy

children

the

up

the

to

communicate

quick

awkward

like

experimenter or

demands from

long

the

to

,

often

works puppets

advantage

puppet

. -

explains

would

additional to

that

experimenter she

-

helps " grown

these

enlists

One

the

puppet

, the

what

the suc

also

. Because

experimenter

deviations

struggling .

to

experiment .

for

talk

in quite

time

. Another

find

interest

puppet

every

shell

' s proven

of

shell

its

Whatever

production

its

of

.

child

has

choice

question

a

- ups the

right

puppet

child

- ups

out

the

the

grown

calls

a

cannot

structure

him for

will

dilemma

elicited

child

The

ask

the

grown

into

to

she

that

probably a

.

of

broken

from

target

frustrate waits

,

her

essential

of

accessing

important

or

- ups

captures

shy

withdraws

information

is

puppet

which

be

verbal

has

resolve

the

be

that

for

quickly to

close

grown

dilemma

puppet

child

,

too

to

a

puppets

to

dinosaur

poses

of

pretends

puppet

gets

newborn

this

use

that children

experimenter is

the

puppet

encouraging

often

was

to

right

.

. However a

( Thornton

relative

( Eisenberg

al

about

used the

do

studies

special

of

producing

found

in

well

the

in by / or

Using

cessful

ask

Part

to

have .

are

for

et

procedure

In

typically

French

);

Italian

.

game

"

is

in

in

to

this .

clauses

goals

children

task

);

production

puppet

Adopting

involve .

experimenters

up

direct

utterances

verbal

therefore

elicited

puppet the

1982

in

children

the

to .

experiments

involvement

We

used

,

1995 and

questions

Crain

features

The

in

Wexler

Dutch

Preliminaries

quite

requests

question

production

Experimental

be

direct

to

scenarios

elicited

ecting

.

involving

can

component

Elicitation

for

be

Thornton

and

in

sentences

experimental

of

directed even

,

,

- phrases

relative

many

particular

strategy

wh

infinitival

( Crain

techniques

. Children be

can

of

Thornton

scrambling

and );

have

,

object

bare

1990

studies

accomplish the

can

versus

properties

these

);

( Hamburger

Emiliani

production

puppet

full

quanti6cation of

to

task

and

( Guasti

1996

English

control

universal

adopt

they

to

);

elicited

be

of

,

1995

questions

( Guasti

in

McKee

all

several

English

properties

clauses

Although

17

);

relative

Italian

in questions

1996

1995

or

negation

Italian

that ,

occur

it

is

when they

can

Elicited Production

149

make the child feel as if he is being tested. If the child does not say anything, it is often best just to move on to the next trial. Sometimes, repeating part of the story will prompt the child to make another attempt at producing the target structure. If the experimenterdecidesto repeat part of a story, however, it should be clear to the child that there is a good reason for doing this. Here the puppet can come to the rescueand say something like, "That story was very hard for me. Could you do that one again?" This savesthe child from feeling as if he failed to perform successfully for the experimenter. If a child does not produce the target structure on the second attempt, it is best to proceedto the next trial. Becauseit is important that children should not feel as if they have failed, it is worthwhile to offer confidence-building statementssuchas "You're doing a great job. This is a hard game for the snail, isn't it? But you're really helping him." Depending on how a particular child reacts, it is also sometimesuseful to insert a "fun" filler trial before moving on to the next test trial. Severalof thesefiller trials should be prepared aheadof time. The fillers can be used to elicit an unrelated structure, preferably one that children will find easy to access . A successfulexperiencewith the filler item servesto renew the child's confidenceand interest in the game. Suchdeviations from the experimenter's planned sessionare essentialto a successfulsession. Failing to take all of the necessarystepsto make a sessionenjoyable for children also jeopardizesthe chancesof other experimentersto work with them. The drawback to accommodatingto the child's comfort and deviating from the planned sessionis that some children will receivemore filler trials and will hear parts of the test stories more often than other children do. Researchersmay find this disquieting, feeling that they would thereby be abandoningproper researchprotocol. In elicited production studies, however, there is little reasonto believe that such differences compromise the experiment. Unless repeating a story or adding a filler item " cues" children about the target structure, their productions are firm evidenceof their grammaticalknowledge. The difficulties in gathering and reporting results are offset by the richnessof production data.

Chapter 18 Eliciting Relative Clauses

This chapter highlights the importance of satisfying the felicity conditions of a target construction when the goal is to elicit sentences from children . It also illustrates how satisfying the felicity conditions of a construction can result in circumstances that are uniquely appropriate for that construction . Alternative ways of expressing the same message are often eliminated once the felicity conditions are met . Failure to appreciate the pragmatic conditions that make sentences appropriate to the conversational context can lead to mistaken conclusions about the syntactic analyses children assign to them . We begin with some general remarks about felicity conditions . Then we consider the conditions that are specifically associated with sentences that contain a restrictive relative clause. As we discussed in chapter 8, failure to attend to the conditions for felicitous usage of the restrictive relative clause has led some researchers to claim that children have a nonadult representation for relative clauses. Such claims are not compatible with the Modularity Matching Model , nor are they consistent with the empirical data that are forthcoming once the felicity conditions are satisfied.

18.1 Felicity Conditions An essential part of designing an elicited production experiment is to understand -the pragmatic conditions that are both necessaryand sufficient for producing the structure under investigation. Identifying the properties of the uniquely appropriate situation for a target construction usually requiresinvesting considerabletime in conducting pilot experimentswith adults and older children, before turning to younger children. The point cannot be overemphasized : pilot work is essentialif the experiment is to achieveits goal. In fact, pilot work often takesmore time than running the actual experiment. Adults' intuitions about the conditions necessaryfor producing a particular structure are frequently dull becausethey have the ability to 'accommodatepragmatic infelicities. Adults can often fill in the missing inferential stepsin responseto speakers who fail to establishthe presuppositionsthat underlie their statements. The downside of having the ability to readily accommodatepresuppositionalfailures is that adults, including linguists, find it difficult to identify those aspectsof pragmatic context that are neededto elicit a particular sentencestructure. Young children 3, 4, and 5 years old do not have the same capacity as adults to understand sentencesproduced in infelicitous circumstances . Apparently, children cannot accommodatepresupposiHonalfailures (see, e.g., Hamburger and Crain 1982). Presumably, this is becausethey do not shareadults' generalknowledge of the world,

152 Chapter18 and not becausethey lack certain linguistic principles (Lewis 1979; Chierchia 1995). This meansthat in order to successfullyelicit a structure from a child, the pragmatics must be exactly right . If an experimental context does not fulfill the presuppositions for an utterance, children are likely to have difficulty producing it . They probably won't understand what is being asked of them, and they will quickly become frustrated with the experiment and with the experimenter. The next section presentsthe acquisition literature and reviews some of the structural properties of the restrictive relative clause. This will be relevant for the ensuing discussion. 18.2 NonadultRelativeClauses The early literature on the acquisition of relative clausesmade two claims: first, that children and adults assign different syntactic representationsto relative clauses, and second, that children's misanalysisresults from either (a) the absenceof certain rules from their grammars or (b) misattachmentof the relative clausein the phrase structure that they assign. We will considerboth claims. The proposalswe considerall stem from experimental evidence that seemingly demonstrateschildren's consistent misunderstanding of a certain relative-clauseconstruction (seeSheldon 1974; de Villiers et al. 1979; Tavakolian 1978, 1981; Goodluck and Tavakolian 1982). The construction that was particularly problematic for children was the as relative, so named becausethe relative clause'modifies the noun in object position of the main clause (i.e., 0 ), and the relative clauseitself contains a superficially empty subject (i.e., 5) . (1) The dog pushesthe sheepthat jumps over the pig. The experimentsthat evoked nonadult responsesfrom children typically used the researchmethodology of the figure manipulation (or act-out) task. In these studies, the experimenter presented children with a sentenceand instructed them to "make it happen" using toys and props that were present in the experimental workspace. These studies revealedthat children consistently acted out as sentenceslike (1) in a nonadult fashion. Many children who were asked to act out the meaning of (1) had the dog push the sheepand then jump over the pig. It is the secondevent that made children's responsesdifferent from those of adults. For adults, (1) assertsthat the sheep, not the dog, jumps over the pig. Children's nonadult enactmentof events led different researchersto different conclusions. For example, Tavakolian (1978, 1981) concludedthat children had assigned a nonadult structure to (1). On her account, the structure that children assignedto (1) closely parallelsthe structure underlying the sentencein (2), which has two conjoined clauses ; it assertsthat the dog pushesthe sheepand then it, the dog, jumps over the pig. Hence, T avakolian dubbed the account the conjoined -clauseanalysisof relative clauses. (2) The dog pushesthe sheepand jumps over the pig. For children, then, Tavakolian's account claimed that the structures underlying sentenceswith a relative clause, such as (I ), have conjoined clauses (i.e., IPs), as in (3).

Eliciting Relative Clauses

(3)

153

IP

IP

.----------

IP

- --------- - - - ~

the dog pushes the sheep

--.-.-------- ------- - ------- " - - - - - -

and

jumps over the pig

In (3), thesheepand jumps overthe pig are not in the sameIP. The diagram in (4) gives the adult syntactic representationcorresponding to sentence (1). In contrast to the representation attributed to children , the adult repre-

sentation of (1) has, as a constituent, sheepthat jumps overthepig.

.

(4 )

According to Tavakolian (1978, 1981), children do not have accessto the (recursive) rules needed to produce the appropriate hierarchical structure for a relative clause. Specifically, sheclaimedthat children's grammarsdo not have recursionwithin the NP ; that is, they are unable to generate one NP inside another NP . Notice that in

the adult structure, an entire clause(containing its own NPs) is generatedbelow the NP containing the sheep . According to the conjoined-clauseanalysis, the inability to form recursive NP structures forces children to produce nonhierarchical, "flat" structures in interpreting sentences like (1).

At the time this "flat structure" hypothesis was formulated, phrase markers were viewed as the product of phrase structure rules. Children's intermediate grammars were assumedto comprise a subsetof the rules and principles that characterizedlater stages. In children's early grammars, the rules that provide recursion within the NP were thought to be unavailable to them. This is depicted in figure 18.1. It was thought that the first recursiverule a child acquiresrepeatsthe category IP, but does not embed it under another IP. Given that this was the only recursive rule available, children were believed to generate a structure like that in (3) for a sentence with a restrictive

relative

clause .

154

Chapter 18

Later

G ram mar

" more

rules "

Early Grammar "few rules "

Figure18.1 Earlyandlatergrammars in a rule-basedsystem

There is another class of explanations for children's reputed conjoined-clause analysis of relative clauses. One alternative account, by Roeper (1982), proposes that children 's nonadult responses are a function of the operating characteristics of the

parserand do not reflect the absenceof grammaticalprinciples. On Roeper's account, a specificoperating characteristic(or strategy) of the parser- namely, its tendency to attach constituents in a way that generates the fewest nodes- encourages the misattachment of relative clauses. This is known as the minimal attachment strategy (Frazier and Fodor 1978 ; Frazier 1978 ).

Although adults too are guided by the minimal attachment strategy, it is applied differently by children and adults, according to Roeper. Adults restrict its application to the resolution of structural ambiguities, whereas children also invoke it in constructing the phrasal analysisof unambiguoussentences . Application of the strategy in (1) results in a nonadult structure with the relative clausepositioned high in the phrase marker, as illustrated in (5). As compared to the adult structure in (4); the structure

in (5) contains fewer nodes (i .e., there are no branches between

the CP con - .

taining the relative -clause and the topmost IP). (5 )

IP -

-

- - - - - - - ----- - - -

IP

6 the dog

- - - - - - - -- - - - -

CP

/ / / "" '" " " NP

-

------------- --------------" ~

VP

that jumps over the pig

/ // /~" """ V

NP

misattached

// """" " /

pushes

~

relative cIa use

the sheep

By relegating the responsibility for children's errors to a performancecomponent of the language apparatus, Roeper's account exonerates the syntax from bearing the

Eliciting Relative Clauses

155

responsibility for children's errors. Therefore, the account is able to maintain that children's syntax is equivalent to that of adults. Roeper's accountis clearly inconsistentwith the Modularity Matching Model; it is an embodiment of the Competing Factors Model . It is a fundamental assumptionof the Modularity Matching Model that the grammar has primacy over the parser. According to this model, the parser cannot influence the representation generated by the grammar for unambiguoussentences . A study by Goodluck and Tavakolian (1982) led them to conclude, like Roeper,

that children's nonadvlt responsesare performanceerrors and do not reflect a lack of linguistic competence. Goodluck and T avakolian contended that children made nonadult responseswhen their linguistic processing system was overtaxed. They found that children ' s nonadult

responses decreased when the sentence contained

fewer ani -

mate nouns (e.g., The dog pushedthe sheepthat jumpedover the pig vs. The dogpushed the sheepthat jumpedover thefence). Apparently, the presenceof three animate nouns in the sentenceburdened the processing system. Similar claims relegating children's nonadult comprehensionof structures to the processingcomponent have been made for control structures by Tavakolian (1981) and by McDaniel, Cairns, and Hsu (1991 ).1

It should be clear by now that all such performanceaccountsof children's "errors" are difficult to reconcilewith the Modularity Matching Model . If adults are not overtaxed by the presence of three animate NPs in a sentence, then children should not

be, if the Modularity Matching Model is correct. Therefore, we must look elsewhere for the source of children 's nonadult behavior . Before we broach this topic , how -

ever, we would like to raise two theoretical problems faced by previous analysesof children 's nonadult behavior . The first , relating to changes in linguistic theory , is of

concern to the conjoined-clause analysis. The second is learnability, that is, how children's grammarsor language-processingsystemschange, to becomeequivalent to those of adults .

18.3 Problemswith the Conjoined ClauseAnalysis The conjoined -clause analysis proposes that children lack recursion in the NP . This analysis may have made intuitive sense when it was initially formulated , within a lin -

guistic theory that viewed grammarsas sets of phrase structure rules. However, the proposal that children lack recursion in the NP makes less sensewithin current incarnationsof linguistic theory. Recently it has been proposed that a general schema within Universal Grammar generatesthe phrasalstructure assignedto sentences . According to the schema , phrasesare hierarchically structured proj ections of their heads. The so-called X' (read X-bar) schemais usually written as shown in (6), where X and Y are variablesranging over any linguistic constituent: Noun (N), Verb (V), Preposition (P), Complernentizer(C), Inflection (1), and so forth . This rule schemaprovides the building blocks for projecting hierarchicalstructure from the headsof phrases(X-level constituents) to phrases(the maximal projections of heads, i.e., XP-Ievel constituents) (see Haegeman 1994 and Radford 1988 for further discussion).

(6) XP - + Specifier, X' X'

- + X ' , YP

X'

- + X , YP

156

Chapter 18

Relative

Clauses

(with Hierarchical Structu re )

Conjoined Clauses

Figure 18.2

The relationship between the conjoined-clauseand relative-clauseinterpretations

The claim that children

lack recursion

in the NP now would

amount

to the claim

that children have incomplete mastery of X' theory. In the adult system, there would be a fully general X' schemathat projects in the sameway for every maximal projection, constructing CPs, IPs, NPs, PPs, and so on. In the child's system, by contrast, phrasal structure would presumably construct projections in one way for CPs, IPs, PPs, and so on, but in a different way for NPs . This seems inconsistent with the spirit of Universal Grammar . Children 's analyses of sentences with a relative clause would be aberrant

.

This brings us to the secondproblem. Figure 18.2 depicts the situation: Children first assign conjoined -clause interpretations to sentences (even ones that have a dif -

ferent meaning for adults). Later, children add to their linguistic repertoire, somehow learning to assign correct structures and their associatedmeaning to sentenceswith relative

clauses .

What is wrong with this picture? The problem for children would be to find evidence from the environment that will compel them to abandonthe conjoined-clause analysis. Adopting this analysisresults in different meaningsfor sentenceslike (1) for children and adults. At first glance, it may seemthat this posesno problem. Perhaps children simply abandon the meaning that they assign to such sentenceswhen they encounter situations that make it clear that adults assign a different meaning . This

won't work in general, however. Becausemany, perhapsmost, sentenceshave more than one meaning, it follows that learnerswho adopted this strategy would be forced to discard many legitimate sentencemeanings. It is more likely that children would add the new interpretation and keep the old one, leaving children with two meanings for a sentence that has only one meaning for adults . That is, children would find sen-

tenceslike (1) ambiguous, whereasadults would find them unambiguous. This leaves the problem of unlearning the offending sentencemeaningunresolved. This problem of unlearning must be confronted by any proposal that would have children assign incorrect meanings to sentences. As noted earlier, learners must identify their errors solely on the basis of positive evidence, in the form of

Eliciting Relative Clauses

(

sentence

is

,

meaning

evidence

to

)

that

sentences

.

a

grammar

of

-

would

Finally

is

,

it

clause

of

not

to

available

be

s

'

s

parsing

mechanisms

differ

children

'

s

parsing

mechanisms

change

These

deliberations

can

18

. 4

lead

be

found

.

Presuppositions

What

wrong

standing

of

supposed

aspects

of

tional

,

presupposed

at

,

found

and

to

Crain

(

suppositions

of

context

.

and

. ,

that

and

Iv

Crain

ing

to

is

the

put

be

fence

to

child

tion

Many

tion

done

the

.

was

event

is

sequence

of

sentence

( first

,

That

,

the

only

restrictive

.

and

The

the

relative

First

if

was

.

second

clause

licking

,

the

:

jumping

of

the

event

event

mentioned

-

before

the

is

example

This

con

,

second

because

a

conflicts

in

the

-

the

with

by

-

con

in

dictated

-

the

pre

as

mentioned

)

is

-

.

clause

.

In

func

context

the

are

events

the

.

presupposi

In

event

events

over

the

relative

,

-

context

place

.

example

which

order

the

restrict

the

second

clause

present

in

The

to

restrictive

took

in

main

some

in

.

Hamburger

jumped

the

fence

clause

in

clause

the

the

that

,

-

sentence

As

is

present

contained

before

there

present

relative

the

.

rates

the

this

relative

context

error

zebra

the

the

pre

experimental

in

for

the

zebras

high

the

over

clause

of

the

jumped

,

in

have

Hamburger

because

clause

presupposition

conceptual

is

children

.

in

that

relative

set

in

,

,

only

one

order

-

are

sentence

clauses

satisfied

zebra

zebras

in

second

the

-

these

conversa

the

experiments

felicitous

the

in

pre

that

of

relative

.

information

event

events

many

not

'

is

place

many

are

the

-

the

that

use

relative

two

contained

taken

or

propositions

the

restrictive

by

,

In

the

resulted

the

,

,

arose

of

;

the

in

to

assert

assumed

-

in

attend

in

If

under

tested

not

established

then

were

from

a

information

licking

' s

' s

sentences

misunderstandings

least

met

does

are

.

presuppositions

at

have

is

.

licked

interesting

of

of

s

described

have

the

children

indicates

been

,

clause

not

presupposition

order

two

that

particularly

this

how

.

were

it

what

restrictive

the

zebra

also

the

ask

children

children

if

sentence

point

lion

be

action

should

supposition

has

was

.

to

)

periments

the

to

If

.

contribution

clauses

uttered

'

that

particular

clause

that

led

circumvented

presupposition

clause

should

,

to

the

in

relative

~

.

of

be

that

with

out

fence

view

senseless

case

The

semantic

e

relative

,

children

restrictive

The

are

testing

our

have

relative

there

one

requires

prior

supposition

a

relative

sentence

conceptual

,

identify

ceptually

jumping

it

a

sentence

the

.

-

those

explanation

established

are

pointed

language

of

ses

restrictive

over

felicitous

confronts

we

can

term

of

case

that

Crain

jumped

used

~

the

this

.

addition

already

worst

argued

the

then

infelicitous

The

sentences

Consider

Hamburger

.

previously

the

cla

)

,

propositions

utterance

misunderstand

1982

adult

,

mechanisms

alternative

In

In

are

speaker

been

relative

been

also

leamability

be

.

These

should

in

?

sentence

the

and

the

presumably

.

experiments

can

the

not

best

Restrictive

by

preceding

have

awkward

an

grammars

performance

adults

assigned

( in

;

mechanisms

of

clauses

the

meaning

,

whether

circumstance

.

sentence

context

. e

A

of

true

of

adult

problems

relative

.

be

those

analysis

be

clauses

leamability

to

clause

whose

evidence

data

of

the

children

relative

"

responses

ask

presuppositions

to

,

semantic

comprehension

with

bear

Therefore

problem

-

also

Clauses

content

entences

.

linguistic

to

the

previous

sentences

)

primary

from

conjoined

cannot

with

nonadult

to

then

Relative

in

presuppositional

( i

,

circumstances

be

us

so

of

went

felicitous

~

If

1

"

the

'

(

sentences

the

that

children

the

sentences

negative

in

noted

attribute

abandon

conjoined

like

analysis

children

errors

to

to

access

should

that

needed

clauses

require

evidence

accounts

is

assigned

relative

conjoined

)

What

meaning

containing

imposed

sort

pairs

the

157

the

the

pre

relative

-

158

Chapter 18

clauseshould have occurred before the assertion contained in the main clause. This kind of conflict between order of mention and conceptualorder will prove to be important in our discussionof children's acquisition of relative clauses. 18.5 Context Two studies by Hamburger and Crain (1982) have led to a different conclusionfrom previous studies about children's grammatical competence. Taking into account the different presuppositionsof relative clauses , Hamburger and Crain identified a different source of children's nonadult behavior, and one that is more consistent with the Modularity Matching Model. The findings from their studies indicate that children's errors in previous experimentswere neither the result of a lack of syntactic knowledge nor due to a misattachmentof constituents within the phrase marker. Instead, children's errors were an experimental artifact. They arose becausechildren were tested in circumstancesthat were infelicitous for the sentencesthat were presentedto them. In experimental investigations using new methodological procedures, it has been amply demonstratedthat children master the relative clausebefore their third birthday. The new procedureswere motivated by the observation, outlined in the previous section, that sentenceswith relative clausesoften bear presuppositionsthat were not met in earlier studies. The typically poor performance by young children in these experimentswas attributed to their ignorance of the linguistic properties of relative clause constructions. But suppose children did know the linguistic properties but were also aware of the associatedpresuppositions. Such children might be unable to relate their correct understanding of the sentencestructures to the inappropriate circumstancesprovided by the experiment. In other words, children were put in a position that required them to ignore either their knowledge of the syntax of the relative-clause construction or to ignore their knowledge of its pragmatics. Adult subjectsmay be able to "see through" the unnaturalnessof an experimental task to the intentions of the experimenter, but it seemsunrealistic to expect this of young children. Following this line of reasoning, Hamburger and Crain made the apparently minor changeof adding more sheepto the acting-out situation for sentence(1). This simple change resulted in a much higher percentageof correct responsesby children even younger than the ones tested in earlier research. The most frequent remaining IIerror" was made by the oldest children, who failed to act out the event describedbv ~ the relative clause. Hamburger and Crain called this the assertion -only response, because these children acted out only the assertionin (I )- that the dog pushesthe sheepand not its presupposition- that there is one sheep that is distinguished from the others by the fact that it alonejumps over the pig. Hamburger and Crain argued, that the assertion-only responseis not really an error, but is precisely the kind of response that is compatiblewith .perfect comprehensionof the test sentences . Responsesof this type did not appear in previous studies presumably becausethese studies failed to meet the presuppositionsof the restrictive relative clause.

18.6 Elicitationof RelativeClauses Themost compellingevidencethat young childrendo not lackthe structuralknowledgeunderlyingrelative-clauseformationcomesfrom a secondexperimentin the

Eliciting Relative Clauses

161

In the past few years, several other unaccommodatingfindings have been reinterpreted as reflecting the influence of task factors, and not lack of structural competence. When questions about syntactic knowledge are askedin a different way, by adopting tasks that control as far as possible for the pragmatic demandsof comprehension, children have been found to succeed(see Crain and Fodor 1993; Crain 1991). We conclude that the failure in previous researchto control for nonsyntactic factors led to underestimatesof children's grammatical capabilitiesin some cases. As a consequenceof the new findings, the timetable for the acquisition of syntax has been brought more in line with the expectationsof the Modularity Matching Model, which anticipates rapid acquisition of grammaticalknowledge without numerous intermediatestagesof successiveapproximation toward the target grammar.

Chapter 19 Asking Questions : The "Ask /Tell " Problem

In elicitation experiments designed to evoke question structures, experimentersare

often plaguedwith what is commonlycalledthe ask/tell problem . Characteristically , in situations designed to elicit a question, children respond with an answer instead. For example, if the experimenter instructs a child IIAsk Kermit what he's eating ," the

child responds by telling the experimenter what Kermit is eating: IIAn ice cream." This problem can be seenas another instanceof failure to satisfy felicity conditions. A J

.

.

.

In ....., general, the problem is more acute with younger children and is encountered freA quently with children . in the 3-year-old range. This is to be expected, since younger

learners are less able to accommodate to infelicity than older ones, who have more

world experience. In order to think about how to avoid the ask/tell problem, we can begin by distinguishing felicity conditions at two levels: at the level of the game, and

at the levelof individualexperimental stimuli. The first step is to satisfy the felicity conditions for the experiment as a whole. That is, the experiment should be designedto involve children in a game in which it makes sense to ask questions . One way is to introduce children to a puppet they have

never seen before and know nothing about. This ploy is used successfullyin Crain and Nakayama's (1987) experiment testing structure-dependence , to be describedin chapter 20. In this experiment, children were asked to pose questions to the Star Wars characterJabbathe Hutt . Jabbawas from another planet and had never been to Earth before. The experimenter began the sessionby "wondering" about Jabbathe Hut ! to set up a lot of unknowns . The experimenter 's patter went something like this : (1)

I wonder if Jabba the Hutt can walk or talk . . . I don 't know . Look at his

body, it's got funny bumps on it . Maybe he's part animal. Could be! I don't know. He looks quite friendly. Maybe he likes people. I wonder if he gets hungry. I have a hamburgerhere. I wonder if he likes hamburgers. I don't know if they have hamburgerson the planet he comesfrom. Maybe you could help me find out. Having set the stage for the experiment, the experimentercan proceed to elicit individual questions. For many children, provided the game is set up in the right way, there is no danger that the ask/tell problem will arise. There are children who have more stringent requirement~ on question asking, however. For these children, the felicity conditions for asking questions must be satisfiedfor each target question in addition to the more global level of the game. Eachtarget question must request information that the child does not already know. If the child knows the answer to the target question, the ask/tell problem reemerges, and the child provides the answer. Continuing with the Jabbathe Hutt theme, the experimentermight elicit a yes/no question, for example, by saying, IIAsk him if they have hamburgerson his planet."

164

Chapter 19

This lead-in would evoke the question "Do you have hamburgerson your planet?" In this case, the global setup of the experimentalso satisfiesthe felicity conditions for an individual target item. In asking whether or not there are hamburgerson Jabbathe Hutt 's planet, the child is genuinely requestingnew information. It is not always simple to set up individual target questions to request new information , however . Let us suppose that among the target structures for the ex-

periment are wh-questions with third person singular subjects. Now, it is no longer possible to have the child ask Jabbathe Hutt a question about himself or his planet becausethese questionswould all require a secondperson pronoun, taking the form "Do you . . ." or 'What do you . . ." . To elicit questions with a third person subject (i .e., 'What does he . . ." ), the ex:perimenter needs to ask the child to ask Jabba the

Hutt about someoneelse. This charactermight be part of a scenariostagedwith toy props- say, one in which Fido eats a cookie but not a bone. Following this scenario, the experimentermight say, I'Ask Jabbathe Hutt what Fido ate." Now, the difficulty is that the question does not request new information. The child already knows th'at Fido ate a cookie and not a bone because this event was acted out with toys . The

danger is that the child will reverf: to answering, not asking the appropriate question. There are two ways to rescue the experiment . One is to do a little more work at

the global level. Having askedJabbathe Hutt some questionsabout himself, the experimenter can suggestfinding out if Jabbaunderstandsabout events that take place on Earth. For example, the experi:menter and the child could find out whether or not Jabba(as a creature from another planet) can pay attention and understand things about Earth, by quizzing him about the events that took place. In this way, it at least makes sense to ask Jabba a question about an event or character that doesn't relate

directly to him. This step doesn't solve the problem of the child's knowing the answer to the target question, however. This problem can be solved by modifying the story. PerhapsFido is given a bone and a cookie, and he takes them behind a wall to hide them . The child can't see behind the wall , but Jabba, the Hutt can. Fido eats the cookie

and buries the bone for later. In this situation, it would make perfect sensefor the child to ask Jabba the Hutt , 'IWhat did Fido eat?"

For a child with an extreme ask/tell problem, it might be necessaryto set up every target question so that the answer is a mystery to the child . For most children , this

last step isn't necessary.As long as the experiment starts off with questionsthe child doesn't know the answer to , the child gets into the question game. At this point , it is

usually possibleto switch to stories like the first Fido sceneabove, without a problem.

Chapter 20 Structure-Dependence

As

Chomsky

( 1965

guistic

)

principles

dependence

to

refer

to

Given

.

The

tation

is

is

in

) .

an

The

straint

.

tions

The

is

particular

not

is

other

to

kinds

linguistic

To

Theoretical

see

iary

move

The

hypothesis

a

NP

verb

is

is

the

as

.

,

.

We

then

knowledge

the

the

in

structure

more

of

The

current

the

experiment

by

/ no

question

questions

and

the

Crain

and

constraint

-

will

basic

/ no

-

ques

constraint

more

yes

form

con

yes

presenting

- dependence

-

about

movement

of

briefly

and

to

head

forming

by

the

structure

the

formation

.

in

chapter

to

the

progresses

the

turn

of

,

-

Crain

hypotheses

formulations

entertain

expec

entertain

show

as

on

theory

begin

to

to

This

by

adherence

known

Particular

conform

.

conducted

will

of

Matching

children

we

theory

will

- independent

constraint

linguistic

We

' s

As

structure

children

.

,

hypotheses

constraint

however

hypotheses

chapter

of

the

development

children

the

-

without

relevant

Nakayama

,

.

Background

how

subject

of

,

.

' s

this

questions

a

,

Modularity

children

language

in

of

structure

would

to

the

that

of

kinds

/ no

violate

change

of

background

children

. 1

to

constructions

theoretical

testing

out

important

continue

the

application

formulation

what

doubt

concerns

yes

the

using

a

words

fundamental

anticipates

stage

the

stated

hypothesis

to

investigated

in

turns

- independent

-

-

."

is

reported

be

of

lin

structure

with

string

the

The

to

ordered

fundamental

Model

examining

inversion

,

an

constraint

every

on

.

contrasts

structure

be

constraint

formation

operations

" leftmost

to

experiment

frameworks

questions

a

and

taken

at

by

- auxiliary

/ no

"

experiment

constraint

linguistic

Such

first

Matching

( 1987

subject

also

operations

evaluated

to

- dependence

it

.dent

dependence

"

general

syntactic

apply .

as

a

grammar

hypothesis

would

structure

Modularity

Nakayama

possible

structure

,

- depen

all

such

the

is

in

- dependent

that

Grammar

structure

- dependence

hypothesize

structure

relations

that

Model

no

A

hierarchical

linear

Universal

20

.

hypothesis

reference

structure

requires

notions

independent

,

children

constraint

structural

yes

notes

that

yes

the

/ no

question

man

moved

is

past

auxiliary

is

)

The

( 2

)

Is

the

man

who

is

( 3

)

' fols

the

man

who

beating

yes

,

however

is

/ no

.

Suppose

,

NP

a

beating

,

a

( 2

relative

is

is

)

is

a

) .

it

( 3

( 1 ) .

question

is

Here

,

not

the

possible

,

the

auxil

-

to

) ) .

.

mean

?

mean

?

compatible

hypothesized

/ no

that

( see

mean

in

yes

Notice

clause

donkey

( 3

child

declarative

forming

giving

donkey

in

the

the

In

donkey

a

question

.

the

beating

consider

donkey

inside

( 1

ungrammatical

a

subject

that

who

formed

beating

the

verb

man

is

who

with

a

that

structure

yes

- independent

/ no

questions

are

166

Chapter 20

formed is

by

moving

considered

local

;

the

" first

simply

word

4

archical

structure

nonlocal movement

. It

as

moves

"

auxiliary

verb

phonetic

to

is

is

a

sentence

taken

into

movement

of

to

stream

of

- initial account

word

8

the

The

man

who

1

2

3

Is

the

man

Is

make

theory

the

man

a donkey

is mean

6

8

5 beating

who

are

head ,

of it

clear

,

is

beating

the

must

7

in

(4 ) .

When

turns

that

hier

out

proves

to

to be

be

local

.

9

a donkey

is mean

?

a donkey

mean

?

by

if

the

the

some

working

movement

) . Because as

introduce

linguists

- auxiliary

Complementizer (5 )

formed CP

of to

we to

looks

subject

movement

the

shown movement

language apparently

1

According

questions

moves

. If is

1

this

.l

name

position

movement

stream

is beating

r

Comp

as this

phonetic

4

who

1

( the

,

this

.

(4 )

T a

,

however

- initial ,

position ,

in

sentence words

of

the

auxiliary

subject inversion

.

the

The

the

from

verb

moves

verb

movement

is

the

head

of

subject

switch more

linguistic

framework

( the

across

auxiliary

current

generative

auxiliary

verb

and

machinery

within

, IP )

( in

positions accurately

yes

to

/ no

Comp

SpecIP

) to

hence

the

;

described

as

CP --------~Comp ' IP -----------//--/--Infl Spec ' " ' " " NP ( l ~ f ) l AdjP ~ ,::man ::who := --~~ ---a-donkey -- ~ 6 the beating mea Infl

next

to

Comp

higher

Phrase

,

however

head

( CP

position

) . This

;

the

head

of

the

phrase

in

movement

the

process

Inflectional

structure is

Phrase

,

depicted

Comp

, the

in

( IP ) , head

Infl

,

of

(5 ) .

Movement of Infl to Comp is subject to the headmovementconstraint. According to this constraint, a head can only move locally, to the next higher head. If the auxiliary verb in the relative clauseis moved, as in (3), it will have to move past the headsof other phrasalprojections: the Comp position in the relative clause, and the head noun of the relative clause (man). This nonlocal movement violates the head movement constraint (Travis 1984) and results in a violation of the Empty CategoryPrinciple (Chomsky 1986). The movement that results in ungrammaticality is shown in (6). The intervening headsof phrasesare circled in the diagram.

Structure-Dependence

(6)

167

CP

/ / ./""'" " " Camp '

lnfl

'

/ / / "-'" " " NP

Infl

AdjP

/ / / " """" Det

N '

Q is

mean

/ // ""~ the

N '

CP

/ / / """'" " "

1 SpecCamp ' who

t @ beatinga donkey Chomsky (1971, 1975) observedthat the issueof children's yes/no questionsis interesting, becausethe structure-dependenthypothesis for yes/no question formation appearsto be more complex from a pretheoretical standpoint than the alternative, structure-independent hypothesis. Moreover, as Chomsky also observed, both hypothesesare compatible with much of the input that children receive, namely, oneclausesentences . For example, in questionslike (7a- b) both the structure-dependent hypothesis of Infl-to-Comp movement and the structure-independent hypothesis "Move the first auxiliary verb" yield the right results. (7) a. Will Bill play the saxophone? b. Is the sky blue? If children's initial hypothesesare limited to structure-dependentones, then they are expectedto use Infl-to-Comp movement as their initial hypothesis about the formation of yes/no questions. This precludesany kind of movement that would violate the headmovement constraint (or "shortestmove" or whatever takesthe placeof the head movement constraint in later formulations of the theory). In particular, ,it precludes hypothesesbasedon the linear position of an auxiliary verb in a sentence.This is what it meansto say that children are restricted to hypothesesthat are structure-dependent. Regardlessof the precise formulation of the structure-dependenceconstraint, the central issue in investigations of Universal Grammar is whether or not children formulate a constraint that can be characterizedonly in structure-dependentterms. The consequencesof proposing that children entertain structure-independent hypotheses would not be trivial . For one thing, a leamability problem would arise.

168

Chapter 20

Suppose, on the basis of simple yes/no questions compatible with both structure dependent and structure -independent hypotheses , that children chose the structure independent hypothesis . Eventually , on the basis of hearing complex examples containing relative clauses like Is the man who is beating a donkey mean? in the input , they figure out the structure -dependent rule of Infl -to -Comp movement and add it to their grammar . The problem is that adding the structure -dependent rule to the grammar does. not entail purging the incorrect structure -independent one. Thus, we might expect ungrammatical yes/ no questions to coexist alongside grammatical ones. Unlike simple yes/no questions like (7a- b), which are compatible with a structure J . ~- ~~- ---- dependent hypothesis and a structure -independent hypothesis , examples like (1), which contain a relative clause, can distinguish the two hypotheses . Children who adhere to the structure -dependence constraint (and the head movement constraint ,) will produce questions like (2). On the other hand, children who can entertain a structure independent hypothesis might readily produce questions like (3). Structures like (1) containing a relative clause, then, are relevant for testing children 's adherence to the structure -dependence constraint . Weare now in a position to layout the experimental procedure for testing whether or not children adopt a structure -independent or a structure -dependent hy pothesis in their initial formulation of yes/ no questions . The null hypothesis is that children adopt structure -independent hypotheses . The experimental hypothesis is that children adopt only structure -dependent hypotheses . The specific grammatical and behavioral consequences of the null (Ho) and experimental (HI ) hypotheses can be summarized as follows : .

Ho: Children can entertain structure -independent hypotheses . They move the first Infl in the linear string to Comp , even if it appears inside a relative clause. Expectedresults: Children initially ask nonadult yes/no questions with sentences containing relative clauses.

.

HI : Children can only consider structure -dependent hypotheses . They always move the Infl from the main clause to Comp , regardless of its posi tion in the string . Expectedresults: Children initially ask adultlike yes/ no questions, even with sentences containing relative clauses.

According to the null hypothesis , children are not restricted to hypotheses that are structure -dependent . On this view , children might produce some sentences that are grammatical and happen to adhere to a structure -dependent hypothesis , and also some ungrammatical sentences that are not compatible with such a hypothesis . In other words , on the null hypothesis , we might expect a random pattern of errors .

20.2 ExperimentalDesign The first step in testing children's mastery of Infl-to-Comp movement in yes/no questions containing a relative clauseis to identify the situations that evoke similar questions from adults. In these situations, we can observe whether children produce question forms that are compatible with the structure-dependenceconstraint, or whether they produce question forms that violate the constraint (as shown by ""sentence ). The design can be summarizedas follows:

Structure -Dependence

(8)

Input Output or Output

169

(situation, lead-in) (sentencel) (*sentence2 ) (structure-dependentviolation)

There is the possibility, of course, that children could come up with question forms that are ungrammatical but do not violate the structure-dependenceconstraint. So, we might add or Output

(*sentenCe3 )

Becausethe questions produced by children are produced in the same situations as adults' questions , we can infer that they have the same meaning . This will be true

even if children's questionshave a different form from those of adults. This brings us to the experiment by Crain and Nakayamaon structure-dependence . Thirty children between the age of 3;2 and 5;11 (mean 4;7) were tested in the experiment . Yes/ no questions containing relative clauses were elicited from the children

during the courseof a game in which they posed questionsto Jabbathe Hutt , a creature from Star Wars. Jabba the Hutt was played by one of the experimenters . The second experimenter orchestrated the game. The experiment followed the usual pro cedure of testing children individually in a research room at the children 's school .

The experimenter in charge of orchestrating the game and manipulating the experimental situations explained to the child that Jabbathe Hutt was from another planet- that humansdon't know about life on his planet, and he doesn't know about life on Earth. (This preliminary conversationwas instrumental to avoiding the ask/tell problem discussedin chapter 19.) The gamebeganby finding out about life on Jabba's planet. This part of the game servedas a "warm-up" and to capturethe child's interest in the game. Later, the focus of the game switched . Jabba the Hutt was shown some pictures , and the child 's task was to find out whether he understood the events de-

picted there by asking him relevant questions, as directed by the experimenter. This second part of the game evoked the pretest questions and the target yes/ no questions that contained

relative

clauses .

In order to elicit yes/no questions, the experimenter showed the child a picture ,

and they discussedit together. This discussionintroduced the characteristicsof the oicture that were relevant and the lexical items that would be needed for the target ...

question.2 If the picture in figure 20.1 had been used, for example, the experimenter would have pointed out that it shows one man who is beating his donkey and another man who isn't beating his donkeys.3 The lead-in for eliciting the relevant yes/ no question embeddedthe relevant statementin a carrier, "Ask Jabbaif .. ." . A sample lead-in is given in (9). In this situation, the child has the opportunity to obey, or violate, the structure-dependenceconstraint. (9)

Protocolfor eliciting yes/ no questions Experimenter Ask Jabba if the man who is beating a donkey is mean. Child Is the man who is beating a donkey mean? or *Is the man who

beating a donkey is mean? After the child formed the question, Jabba was shown the picture and responded "Yes" or "No ." (Since Jabba came from another planet, he would not necessarily know whether beating a donkey was a mean thing to do.) If Jabbaresponded correctly, the child fed him a frog, his favorite food.

171 Table20.1 Correctandincorrectresponses by group

Responses 81 87 168

Group -1 Group 2 Total h. .

1.

Grammatical 31(38%) 70(80%) 101(60%)

Ungrammatical 50(62%) 17(20%) 67(40%)

The boy who is being kissedby his mother is happy. The ball that the girl is sitting on is big.

20.3 Results The children were split into two groups for purposes of data analysis. Group 1 included the younger children, ranging in age from 3;2 to 4;7 (mean 4;3). Group 2 included the older children, ranging in age from 4;7 to 5;11 (mean 5;3). Both groups of children performed well on the pretest sentences , correctly forming grammaticalyes/ no questions. The results from the test sentenceswere surprising, however. Among children's productions were a sizable number of ungrammatical questions. In fact, 62% of the younger children's questions were ungrammatical. The older children produced fewer errors; still, 20% of their productions were not the expected "adult" yes/no questions. Table 20.1 summarizesthe results. Clearly, a thorough analysis of children's errors was necessarybefore any conclusions about the experimental and null hypotheses could be drawn. Unless most of the nonadult responsescould be shown to be consistent with the Modularity Matching Model, the high incidenceof nonadult responsesin the experiment could be troublesomefor the model. Crain and Nakayama begin their discussion of children's errors by pointing out that if children did not have innate knowledge of the structure-dependenceconstraint, and their ungrammatical productions representedmisgeneralizationsof a structureindependenthypothesis, they might be expectedto make random errors acrossthe test items. This did not happen, however. Certain test sentencesresultedin more nonadult questions than others. For example, for group 1, items (IDe), (lOg), and (10h) were most problematic, evoking at most 20% grammatical sentences . On the other hand, more than half of children's responsesto the other items were grammatical. This brings us to the crucial question. Did any of the nonadult question forms producedby the children violate the structure-dependenceconstraint? That is, were there any questions in which the auxiliary verb from the main clausewas left in its base position, and the one from the relative clausemoved to Comp, as in (II )? Or were children's nonadult questionsungrammaticalfor other reasons? (11)

Is the man who beating the donkey is mean?

There were no questions like (11). However, close examination of the errors produced by the children raised new questions. In the nonadult question type produced most frequently by the children, an auxiliary verb appearedin Comp, but there was no gap, either in the relative clauseor in the main clause. An example of this kind of error is given in (12). Theseerrors are labeleddouble-Infi errors in table 20.2. (12)

Is the man who is beating a donkey is mean?

172

Chapter 20

These questions did not make it possible to teaseapart a structure-dependenthypothesisfrom a structure-independentone becauseit was not possible to tell which, if any, of the auxiliary verbs had "moved" or, at least, which auxiliary verb was "related" to the one in CompoIn the 1970s, errors of this sort were termed copying-withoutdeletion errors (Mayer , Erreich, and Valian 1978). Then , the idea was that the child

had inserted a copy of the auxiliary verb in the Comp position, but had forgotten to delete the original one. A variant of this analysis would be to say that the auxiliary verb in (12) has beenmoved to Comp, and the trace of movement is "spelled out.'" As Crain and Nakayama point out , there is yet another analysis of questions like (12).

They could be formed by simply prefixing the auxiliary is to the declarative. This would be much like forming questionsby affixing ka in Japaneseor prefixing est-ceque in French. It is not clear whether children producing questionsof this sort would be applying a question formation J"strategy" basedon a structure-dependenthypothesis. There was one more error type, termed restart errors by Crain and Nakayama. In thesequestions, children began with a well-formed fragment of a question and followed it with a secondquestioncontaining a proform. Thesequestionshave the flavor of performanceerrors. The child gets underway with the question, appearsto forget what shewas saying, and recovers. An exampleis givep in (13). (13) Is the boy that is watching Mickey Mouse, is hehappy? Errors made by the children are summarized in table 20 .2.

As noted, the double-Infl errors could not help to distinguish a' structure-dependent hypothesis from a structure-independentone sinceit was not po~sible to examinethe "origins" of the auxiliary verb in Comp. Sinceas many as 62% of the errors made by the younger group of children were of this kind , Crain and Nakayama were not able

to argue that the structure-dependenceconstraint was in place. They could only make a weaker argument for innate knowledge of the constraint: that children do not make errors like (3) that blatantly violate the structure-dependenceconstraint. Naturally, Crain and Nakayama had not anticipated errors like (12) that would fail to disentangle structure-dependent from structure-independent hypotheses. To investigate the structure of double -Infl questions like (12), they conducted a second

experiment. Ten of the children who had produced questionslike (12) in the first experiment participated in the follow -up experiment. Again an elicited production task was used to further probe for the sourceof children's nonadult productions. The sentences used in this experiment introduced a different auxiliary verb in the

relative clauseand the main clauseof the lead-in sentences , as exemplifiedin (14). (14)

a. The boy who is happy canseeMickey Mouse. b. The boy who canseeMickey Mouse is happy.

Structure-Dependence

173

Statementslike (14a- b) containing two different auxiliary verbs could be used to test the structure-dependenceconstraint, while investigating the structure of children's questions. Let us take (14a) as an example. In responseto the lead-in, children could potentially produce questionswith the structuresillustrated in (15a- d). Only the first yes/no question, (15a), is correct" however. (15)

Protocolfor elicitingyes/no questionswith fwo J'auxiliary" verbs Experimenter : Ask Jabbaif the boy who is happy can seeMickey Mouse. a. Child Canthe boy who is happy seeMickey Mouse? b. Child *Is the boy who happy canseeMickey Mouse? c. Child *Can the boy who is happy canseeMickey Mouse? d. Child *Is the boy who is happy canseeMickey Mouse?

The question in (15a) is the grammatical question form that does not violate the structure-dependenceconstraint. The questionin (ISb) reflectsa structure-independent hypothesis becausethe auxiliary verb from the relative clausehas incorrectly been moved to Comp. The questionsin (15c) and (lSd ) reveal the derivation of the doubleInfl question structure. In (15c), the auxiliary verb that is doubled is the modal can from the main clause. Although the question is ungrammatical, the question form supports a structure-dependenthypothesis. The child clearly knows that the auxiliary verb in the main clauseraisesto CompoFurther, (lSc) would suggestthat the child is not forming questionsby prefixing is in sentence-initial position. On the other hand, questionslike (lSd) do not teaseapart the prefix analysisof the double-Infl questions from a structure-independent analysis. That is, questionslike (lSd) could have been formed by prefixing is to the declarativesentence,or they could have beenformed by moving (or copying) the auxiliary verb in the relative clauseto Camp and spelling out the trace (or failing to delete the copied auxiliary). The second, follow -up experiment began with elicitation of two simple yes/no questionswith the modal can. There were four target sentences : two like (14a), with the modal in the main clause, and two like (14b), with the modal in the relative clause. Two targets contained can1and two were designedto elicit should. Children apparently had more difficulty producing questionsin the secondexperiment than in the first. Crain and Nakayamasuggest this may be becausepictures are not ideally suited to eliciting questionswith modal verbs. In order to make sure that the child would produce a modal verb, the description of the pictures was more abstract. Intuitively , it is harder to depict or convey the idea that someoneis able or obliged to perform some action than to depict or convey the idea that someoneis performing ... - some action. These -problems could be alleviated to some extent if the scenarioswere acted out with toys and props. Children produced a greater number of errors in the secondexperiment than they had in the first. For the 10 children who -participated, the error rate increasedfrom 65% nonadult questionsto 79%, Altogether, the 10 children produced34 yes/no questions containing relative clauses. Of these! 7 were grammatical, 9 could not be analyzed becausethey were question fragments, and 7 were !'restart" errors. None of the remaining I I responseswere questions; like (15a), that clearly violated the structuredependenceconstraint; in other words, there were 1I double-auxiliary questions that were open to interpretation. The results are summarizedin table 20.3. Of the questionswith is as the first auxiliary verb, 6 were like (15d); that .is, they could not

174

Chapter 20

Table 20.3 Analysis of double-Infl errors by sentencetype Rel.-clauseInfl Sentencetype is/modal . modal/ is

Main-clause

moved/copied

Prefix

Infl moved/copied -

6 0

-.1) 2 -

distinguish between the prefix analysis and a structure-independent error (e.g., Is the boy who is happycanseeMickey Mouse?). In another 3 questions, the main-clauseauxiliary verb had clearly been moved/copied, in accord with the structure-dependence constraint. Of the questionswith the modal in the relative clause, 2 could be analyzed either as being structure-dependentor as exhibiting the prefix analysis(e.g., Is the boy who can seeMickey Mouse is happy?) but none unambiguously supported a structureindependenthypothesis (e.g., Can the boy who canseeMickey Mouse is happy?). To summarizethe results: The children did not produce any questionsthat clearly exhibited a structure-independentoperation; that is, no child produced questionslike Is the boy who happycanseeMickey Mouse? or Can the boy who canseeMickey Mouse is happy?Although analysisof a few of the questionswas not obvious, children's nonadult utteranceswere best analyzed either as prefix errors or as instancesof moving/ copying and spelling out the trace/failing to delete. Overall, the experiment invites the inferencethat children abide by the structure-dependenceconstraint. In a further experiment, Nakayama (1987) showed that the double-Infl errors appearedin questionsthat placed more demandson children's processingsystems. The auxiliary verb was doubled more in questions in which the relative clausewas long and contained a transitive verb than in ones with short relative clausesand intransitive verbs. Likewise, more errors appearedin relative clauseswith an object gap than in those with a subject gap. All of thesefactors suggesta performanceexplanation of the nonadult questions. As we discussedin chapter 4, the puzzle is that the doubleInfl errors are errors of insertion, not deletion, which is characteristicof performance errors. Our suggestion is that children respond to increasesin processingload by reverting to an earlier grammar (seeLebeaux1988). Thus, although the children tested in this experiment produced yes/no questions with doubling of the auxiliary only when processingdemandswere strenuous, we assumethat at an earlier stage, the doubling would have been consistent with their grammars. As it happens, the auxiliarydoubling error has been found to be consistent in another syntactic environment: questionscontaining negation (Thornton 1993; Guasti, Thornton, and Wexler 1995). This suggeststhat we are correct in thinking that this is an error that is compatible with Universal Grammar. 20.4 Conclusion

The experiments describedin this chapter have shown how the elicited production techniquecan be used to test a constraint that rules out certain sentenceforms. In this case, the technique was used to seeif children produce yes/no questionsthat violate the structure-dependenceconstraint or the headmovement constraint. An experimental hypothesis was set up and tested in the first experiment. However, some children produced an unanticipatedform of nonadult question that made it impossible to dis-

Structure-Dependence

175

entangle the experimental and null hypotheses. The nonadult questions of these children were further examinedin a follow -up experiment also using the elicited production technique. This experimentprobed the sourceof the double-Infl error in order to provide evidenceeither for or against the experimentalhypothesis. The conclusion was that children's questions provide no evidence that can be uncontroversially explained as violating the structure-dependenceconstraint. A few questions could be analyzed as violating the structure-dependenceconstraint, but they could equally well be explained as prefix errors. Finally, we have assumedthat becausechildren do not produce questionsthat violate the structure-dependenceconstraint, the constraint is in place in their grammars. At this point, we should raise the issue of principles versus preferences.Do children fail to produce questions with the auxiliary verb from the relative clause in Camp simply becausethey prefer to move the auxiliary verb from the main clause? Or does their failure to move the auxiliary ~ verb from the relative clauseindicate that their innate knowledge of the constraint prevents the illicit movement? It is not possible to prove that the ungrammaticalquestionsfail to appearbecausethey are prohibited by the. structure-dependence constraint. On the other hand, it seemsvery unlikely that ~ children are merely exhibiting a preferencefor moving the auxiliary verb from the main clause. Given the high number of ungrammatical questions that children produced in this experiment, we might expect the structure-independent error to have appeared. And it did not. In describing an experiment on wanna contraction in the next chapter, we will show how an experimentalcontrol condition can eliminate this kind of preferences -versus-principles problem.

Chapter21 Wanna Contraction

In this chapter , we investigate the constraint that blocks contraction of the verbal elements want and to in certain wh -questions , first discussed in chapter 2 . Children ' s knowledge

of this constraint

is examined

by ' eliciting

questions

from the " wanna con -

traction " paradigm . There are two important reasons for presenting this experiment . First , it provides a very clear comparison of the Input Matching Model and the Modularity Matching Model . The two models make different predictions about the outcome of the experiment . Second , it demonstrates that an elicited production experiment experimental principle principle

can clearly

adjudicate

control

that provides

between

principles

a convincing

and preferences . It includes

test of children ' s knowledge

of Universal Grammar ; we are able to disentangle from their preference for a particular response .

children ' s knowledge

an of a of a

21 .1 Forming Wh - Questions The phenomena investigated in this chapter concern the formation of wh -questions . For the sake of this discussion , we assume that wh -questions are formed by move ment of a wh -phrase from its position

at an underlying

level of representation

to its

surface position . A further assumption is that an empty category , which we abbre viate as t (for " trace " ) is left behind as a record of wh -movement . The constraint of Universal Grammar that is the focus of our discussion prohibits contraction of two elements if a trace of wh -movement intervenes between them .! In English , the con straint

on contraction

restriction

can be contracted ronments

across a wh -trace (i .e., a trace of wh -movement ) plays out as a

on the environments in which

in which

sequences like want to, supposed to, have to

to roanna , sposta, hafta . In French , the constraint liaison

is possible

(see Selkirk

English version of the phenomenon .2 First , let us examine the facts that the constraint questions infinitival

attempts

focus on the

to explain . Compare

wh -

in which the wh -phrase is extracted from object position of an embedded clause with those in which the wh -phrase is extracted from su~ject position .

In object extraction extraction

affects the envi -

1972 ). Here , we will

questions

l.ike (I ), wanna contraction

questions like (2), it is not .

(1)

Object extraction a. Who do you want to kiss t ? b . Who do you wanna kiss t7

(2 )

Subject extraction a. Who do you want t to kiss Bill ? b . *Who do you wanna kiss Bill ?

is permitted , but in subject

178

Chapter 21

(3) Declaratives a. I want to kiss Bill. b. I wanna kiss Bill. The theory of Universal Grammar explains the facts as follows. In (I ), the wh-phrase who is moved from object position, and a trace (t) is left behind in this position. When the trace is in object position, it does not interfere with contraction of want to to wanna. This is not the casewith (2), however. The wh-phraseis moved from subject position of the lower clause (cf. You want who to kiss Bill?), leaving a wh-trace between want and to. This is the environment in which the constraint rules out contraction. In declarativeslike (3), there are no tracesof wh-movement. The constraint is therefore irrelevant and contraction is not inhibited. Consider, first, the predictions of the Input Matching Model for acquisition. On this model, children rely heavily on the input to formulate grammatical hypotheses. Questions like (la ) and (lb ) provide ampleevidencethat contraction is possiblewhen the lexical items want and to are adjacentin the string of words. In fact, children hear contraction frequently and should therefore develop a preferencefor contraction. This preferencefor contraction should extend to subject extraction questions, sinceon this model no constraint inhibits contraction in this syntactic environment. According to this model, children monitor the input - and reachthe conclusionthat contraction is not possible in subject extraction questions. Therefore, contraction in subject extraction questionsshould taper off gradually. Until children reach the conclusion that the extraction site is relevant,. however.. they would be expected to contract as often in subject extraction questions as they do in object extraction questions. Note that there could be parsing pressuresto encourage contraction in object extraction questions, however. The literature on aphasicsand poor readershas establishedthat object extraction questions are more difficult than subject extraction questions. If this is the case, there might be more pressureto contract in object extraction questions than in subject extraction questions. Either way, on the Input Matching Model, children are expectedto produce numerousillicit contractions, like (2b), which might persist for some time, until they realize that a cell in the paradigm is missing from the languagethat they hear, and formulate the correct grammaticalhypothesis. Universal Grammar takes a quite different approach. Proponents of Universal Grammarassumethat the constraint is innately specifiedprecisely becausemastery of it would require negative evidence; adult wh~questionslike (la ), (lb ), and (2a) inform the child which questionsare permitted by the grammar, but do not inform the child that contraction is impossible when the wh-phrase is moved from subject position (i.e., in (2b)). Since negative evidence is thought not to be available, the solution is simple: the knowledge is innate. On the Modularity Matching Model, then, if the constraint is part of Universal Grammar, children should be steadfastin their adherence to it . Children should not mistakenly contract when they produce subject extraction questions, regardlessof any conflicting parsing preference. As soon as they can be tested, all children should be found to avoid contracting want and to in their subject extraction questions. To test children's knowledge of the constraint, we examine the hypotheses that children entertain about contraction by examining their production of questionsfrom the wannacontraction paradigm. Our hypotheseswill be as follows:

Wanna

.

Contraction

179

Ho: Children will allow contraction in subject extraction questions (as often as in object extraction questions ).

. HI: Childrenwill not producesubject"extractionquestionswith contraction . 21.2 Design The experiment discussed in this section includes an essential control that enables

us to disentanglechildren's knowledge of the constraint from their preferencesfor a particular sentenceform. We are able to do this by taking advantageof an interplay between the syntactic constraint and a parsing preferenceto produce reducedforms. Before we proceedto the experimentaldesign, it is instructive to review the design of the structure-dependenceexperiment, to seewhy that design is not appropriatefor studying children's knowledge of the constraint on contraction acrosswh-trace. The design used in the structure-dependenceexperiment is representedin (4). (4)

Input Output or Output

(situation, lead-in) (sentencel) (*sentencez )

In reporting that experiment, we noted that children produced many questions that were ungrammatical but that, crucially, none of their questions clearly violated the structure-dependenceconstraint. From this, we concludedthat children obey the constraint. Supposewe apply the samelogic to the constraint on contraction. Using the design in (4), we would present children with the appropriate situation for a target subject extraction question, and we would observe whether they produced both uncontracted and contracted variants in their questions or whether they instead con-

sistently produced the grammatical questions without contraction. This implementation of the design is illustrated in (5). (5)

Input Output or both Output Output

(situation, lead-in) (sentencel. . . want to . . .) (sentencel. . . want to . . .) (*sentence2 . . . wanna. . .)

Approaching wanna contraction in the sameway as structure-dependencewould lead us to conclude from children's avoidance of the ungrammatical sentencewith wanna that they are demonstrating knowledge of the constraint blocking contraction . There is an important difference between the two experiments , however . In the

structure-dependenceexperiment, questions that violate the structure-dependence constraint (i.e., the head movement constraint) are always ungrammatical. In the present experiment , the form of contraction (wanna) that is ruled out by the constraint

in subject extraction questions is permitted in different syntactic environments~ for e.xample, in object extraction questions. When extraction takes place from object position, both wannaand want to are permitted. This raisesthe following dilemma. Children can chooseeither to contract or not to contract in object extraction questions.-Supposethey consistently choosenot to contract, for whatever reason; call it factor F. Factor F would presumably also apply to children 's subject extraction questions . If so, then children would not contract in

subject extraction questions even if this were possible in their grammars. Therefore, simply demonstrating that children do not contract in subject extraction questions

180

Chapter 21

would not be convincing evidence that they do so becausethey are obeying a constraint. They might simply prefer to avoid contraction. It must be demonstratedthat there is no factor, F, pressuring children to avoid contraction whenever possible. In this experiment, then, the issueof preferencesversusprinciples is critical. How do we show that children's failure to contract in questions like (2b) reflects knowledge of the constraint? One possibility is to examine adults. Appealing to the Modularity Matching Model, we can say that if adults exhibit a preferenceto use reduced forms whenever possible, then children should exhibit this same preference. The tendency to use reduced forms flies in the face of factor F, which is the bias to use the unreducedform of want and to. Of course, this begs the question. Becausewe are trying to argue for the Modularity Matching Model, we cannot presupposeit . To make a more compelling casefor children's innate knowledge of the constraint, then, we would need to disprove the proposal that there is some factor, F, that compels children to avoid contraction. This is done by including object extraction questions as controls in the experiment. Since both contracted and uncontracted forms of (I ) are grammatical, eachindividual child's preferencefor contraction should be assessed . If it can be shown that children prefer to contract in 0bject extraction questionslike (lb ), but go out of their way to avoid contraction in subject extraction questionslwe would have strong evidencethat they are obeyinR the constraint. The design incorporating both subj~ct and object ext~action qu-esttons is shown in (6) and (7). (6)

Objectex-traction Input (situation, lead-in) Output (sentencel. . . want to . . .) and Output (sentencez. . .wanna. . .)

(7)

Subjectextraction Input (situation, lead-;n) Output (sentenCe3 ' . . w.ant to . . .) ??? Output (*sentence4 . Oc . wanna. . .)

The goal, then, is to demonstrate children's preference to contract in object extraction questions, where contraction is optional. Fortunately for this experiment, we already know that adults exhibit a parsing preferenceto use reduced forms (see lasnik 1990), which leads them to favor contraction in object extraction questions like (1). Following the Modularity Matching Model, we would expect children to show the samepreferencesas adults, so they too should manifest a tendency to use reduced forms whenever possible. Children should therefore favor wanna over want to in their object extraction questions. Given these observations, the first part of the oexperimentaldesign is to set up situations that elicit object extraction questions from children. This will establish whether or not children exhibit the samepreferenceas adults to use reducedforms, in this caseto produce wannarather than want to. Once children's preferenceto contract is established, we move to the secondpart of the experimental design, eliciting subject extraction questions from children. This will enable us to determine if children override their preferencefor contraction in their subject extraction questions. We have discussedthe importance of avoiding type I errors (improperly rejecting the null hypothesis) in conducting experiments. The null hypothesis in the present experiment is that children lack-the constraint on contraction acrossa wh-trace. The

181

Wanna Contraction

experimental

hypothesis

contract

in

" sures

that

encourage

to

help

,

questions

.

clues

,

21

. 3

of

These

The

tested

children

a

live

in

finding

ups

.

II

the

,

game

up

)

.

tested

errors

,

the

if

,

this

experi

some

if

-

aspect

the

puppet

style

Ask

on

he

; 5

)

of

speak

-

~

where

he

lived

to

and

the

elicit

on

the

him

child

'

child

.

s

help

invited

to

so

-

come

about

the

,

-

shy

had

questions

,

con

experimen

Ratty

shy

up

.

was

anything

less

-

who

other

solicited

warm

from

the

,

the

know

She

feeling

data

constraint

within

that

' t

The

the

Ratty

;

.

was

.

licited

named

didn

simple

; 5

.

experimenter

an

ongoing

for

pizza

often

makes

eat

that

do

and

find

Once

out

the

structures

was

what

if

child

at

issue

kind

you

wanna

over

the

individual

protocol

in

(

experimenter

.

8

)

the

items

of

plas

out

as

.

I

and

I

' ll

bet

he

have

to

,

wants

rat

to

,

then

.

)

the

what

eat

So

,

Rather

,

some

-

.

he

not

,

for

the

that

game

were

.

protocol

of

targets

puppet

the

some

drink

the

experimenter

the

in

,

?

general

the

shown

thirsty

in

by

with

(

does

uses

-

.

hungry

wanted

,

acted

protocol

In

.

they

was

fol

were

example

elicited

,

after

experimenter

protocol

wanted

elicited

Ratty

noted

repeated

,

that

substituting

.

experimenter

.

noted

Various

proceeded

there

Ratty

experiment

eat

,

the

?

pizza

discourse

to

? "

hungry

game

of

eat

pepperoni

"

probably

the

each

in

.

this

stories

eat

questions

looks

assessed

as

wanna

extraction

rat

controls

elicited

you

and

what

For

the

were

workspace

do

people

formed

' What

object

that

.

"

of

The

5

to

4

rat

example

questions

investigated

so

asked

.

a

proceeded

snack

the

him

Is

separate

for

for

,

and

explained

with

was

10

were

,

one

when

he

;

adherence

age

she

,

that

eliciting

What

and

Notice

,

These

The

questions

part

s

( mean

puppet

experimenter

on

in

for

Child

lowing

the

for

out

out

Ratty

drink

any

,

example

mimic

2

'

because

were

old

.

missed

Protocol

,

how

of

experimenter

worry

later

situation

laid

thing

play

; 5

by

up

began

contract

the

had

Similar

a

and

,

ages

5

a

drinks

questions

to

were

8

girl

extraction

Ratty

pizza

extraction

providing

.

set

food

a

to

played

session

Experimenter

as

to

paradigm

-

was

and

,

; 6

around

was

it

The

or

in

to

had

tempted

One

extraction

hand

I

for

.

subject

other

type

children

3

grown

Ratty

tendency

(

to

be

contraction

The

and

.

boy

object

that

the

the

designed

object

the

-

that

avoid

-

the

from

Ratty

. "

about

a

age

foods

experiment

' s

up

,

game

immersed

The

of

-

house

was

order

On

potential

assessing

revolved

role

favorite

out

the

child

to

between

wanna

game

The

playa

rat

was

.

mea

ensures

the

to

should

contraction

might

for

in

the

The

grown

her

his

to

the

.

the

at

him

with

allowed

avoid

This

not

seek

is

later

speech

will

should

experiment

switch

s

.

we

.

used

from

-

was

what

to

leading

Children

children

ranged

game

grown

to

26

were

questions

of

-

,

can

the

therefore

,

experiment

and

not

to

forms

children

14

ter

the

is

favored

.

way

'

and

errors

they

experimenter

children

manner

I

Experiment

these

text

be

begin

and

type

the

,

the

is

uncontracted

experiment

14

tic

staccato

to

The

The

in

a

leading

to

preferred

,

would

by

is

is

tempted

in

ing

errors

contraction

experiment

spoke

of

when

hypothesis

the

I

constraint

avoid

whenever

favored

importantly

about

mental

type

the

To

contract

not

contraction

More

.

to

is

where

know

questions

children

avoid

questions

children

ion

hypothesis

way

that

extract

experimental

of

is

subject

not

wants

include

to

use

(

i

. e

. ,

third

of

the

person

want

to

) ,

but

sequence

this

that

is

not

is

a

sequence

being

-

182

Chapter 21

that

can

the

possibility

be

contracted

of

experimenter

' s

instead

of

not

of

cases

,

a

this

,

for

menter

.

might

cavities

.

The

as

9

)

he

for

eliciting

subject

I

who

Child

Who

You

critical

like

puppet

in

choose

"

task

a

"

was

perimenter

in

10

of

lot

to

teeth

playing

Excuse

the

me

? "

ongoing

drink

In

most

the

conversa

and

because

in

eat

he

,

,

experi

' t

and

-

the

wouldn

workspace

the

,

-

want

to

protocol

)

.

a

idea

to

Complex

brush

his

teeth

for

him

.

Ask

?

you

And

some

gets

take

a

gets

to

to

eat

Child

a

cookie

Who

Ratty

do

'

you

are

'

s

,

going

to

eat

a

dog

the

.

a

to

he

cookie

N

the

-

child

that

Cookie

ow

.

and

,

a

first

who

and

,

take

cookie

rat

,

happen

things

the

,

and

ex

the

.

to

gets

do

Ask

want

Monster

baby

guys

Let

?

felicitous

' s

the

place

them

to

Ratty

questions

a

different

.

right

s

take

rat

got

and

by

among

extraction

there

these

cookie

could

be

"

,

-

the

Ratty

described

events

pro

and

workspace

actions

not

those

of

the

simple

child

that

in

prepared

the

the

distinguish

subject

to

explained

three

would

,

one

,

do

Cookie

of

things

,

eat

a

"

game

to

was

to

eliciting

walk

it

was

addition

involved

potential

?

protocol

in

placed

needed

different

who

to

,

one

want

this

complex

was

were

several

for

more

This

game

with

if

In

guys

teeth

protocol

this

was

do

a

.

toys

question

protocol

choose

In

up

that

full

,

Three

one

Experimenter

gets

your

complex

.

.

' Who

gets

brush

question

more

game

was

a

?

to

full

The

each

OK

questions

someone

questions

ask

"

that

)

his

wants

want

happen

fragment

(

9

choosing

The

realize

question

(

match

.

would

part

a

toothbrush

rat

extraction

would

to

as

had

extraction

the

you

to

one

what

bet

do

child

the

"

,

or

experimenter

or

? "

whether

!

subject

the

tocols

the

the

want

.

Ratty

encourage

you

:

Protocol

the

to

to

do

evaluating

? "

about

.

elicited

had

a

,

' What

clues

response

' What

for

happens

question

no

in

,

useful

,

brush

put

says

not

this

full

Ratty

probably

Experimenter

For

If

with

arises

simply

saying

also

after

experimenter

him

.

a

were

should

follows

to

by

elicits

,

child

that

child

is

repetition

example

say

proceeded

(

For

the

question

want

a

questions

Ratty

the

This

sequence

for

provides

difficulty

that

clarification

extraction

with

is

.

ask

it

potential

,

question

can

so

One

however

full

request

,

.

in

the

Ratty

Subject

tion

get

-

contract

role

wanna

contraction

lead

asking

children

the

to

one

nap

,

gets

to

these

and

one

one

,

rat

of

,

So

wants

Monster

the

guys

of

of

these

these

guys

.

?

!

21.4 Analysis of the Data In

this

experiment

,

experimenters

open

to

impartial

to

judgments

the

criticism

waveform

that

wh

-

phrase

-

the

Some

is

potential

position

s

adherence

children

relevant

not

deduce

and

and

to

' s

not

of

questions

sort

and

comparing

following

not

leaves

the

by

experi

those

were

independent

.

.

grammatical

.

do

of

digitized

judges

was

.

-

not

soliciting

with

children

string

the

is

them

contraction

the

the

avoided

five

on

experimenter

was

to

or

rests

This

the

this

presented

VP

.

from

whether

the

constraint

because

students

were

the

transcripts

scientific

criticism

the

program

could

initial

it

graduate

of

editing

judges

in

A

linguistics

.

a

ensure

.

four

'

of

that

outcome

experimenters

children

recording

the

from

using

the

of

subjective

ment

the

assessment

you

To

,

wannaf

183

Wanna Contraction

want to were omitted. The judges were thus presentedwith an artificial circumstance that was quite difficult to judge in isolation. For eachof the five children, eachjudge heard four instancesof what the experimenterhad perceived as wanna, and four that the experimenterhad perceivedas want to. The crucial instancesof want to were analyzed for agreementwith the experimenter's judgments. Of the total of 80 judgments made by the four judges, 91% (73/80) agreed with the experimenter's judgments, with 6 of the 7 judgments that did not agree coming from the productions of one child. Given this high level of agreement with the experimenter's judgments, the results were consideredto be valid. It may be helpful to preface our discussionof the findings by reviewing t~e expectations of the null and experimentalhypotheses. On the null hypothesis, children are assumednot to have any innate knowledge of the constraint! so we would expect them to contract equally often in subject and object extraction questions. How often children do contract in both question types would presumablydependon the parsing tendency to reduceforms. On the experimentalhypothesis, children are expectedto contract in object extraction questions, at whatever level the parsing preferencedictates. By contrast, children!s subject extraction questionsare expectednever to show contraction. The experiment is only convincing, however! if children exhibit a preferencefor contraction in object extraction questions. Both the null and experimental hypotheses depend on children producing both subject and object extraction questions, so only children who produced questionsof both types could be included in the analysis. In addition! only children who produced two or more questionsof each type were included, since these children all produced contracted object extraction questions. These criteria selectedthe 14 children whose data are shown in table 21.1. These childrens failure to contract in questionsextracting from subjectposition could not be taken as a preferenceto avoid contraction. Table 21 .1

Data of 14 children included in analysis Object Subj/Age

Want to

Subject Wanna

Other

Want to

Wanna

Other 1

KM

5 ;5

-

4

-

4

2

PM

5;1

-

3

-

3

-

SA

4 ; 10

-

4

-

4

1

TI

4 ;9

1

5

-

3

-

ST

4 ;8

-

5

-

4

-

IS

4;7

-

6

-

5

-

AM

4 ;6

-

2

-

4

-

BD

4 ;6

-

8

-

9

-

KE

4 ;4

-

2

-

5

-

GA

4 ;0

-

3

-

5

-

KL

3 ; 10

-

4

-

3

SR

3 ; 10

6

6

-

7

CA

3; 7

1

5

-

7

IE

3;6

-

3

-

5

8

60

0

68

-

-

-

-

Total

3

-

-

-

6

1

-

184

Chapter 21

Table21.2 Dataof 12childrenexcludedfrom analysis Object

Subj/Age

Want

MI 5;5 CA 5;4 SO 5;3 JI 5;1 KP 5;0 TT 3;9 KR 3;9 MO 3;9 RE 3;9 MC 3;7 MA 3;3 PI 2;10 Total

Subject to

Wanna

Other

Want

to

Wanna

Other

1 1 2 5 I -

1 1 1 1 3 I I

1 2 I 4 -

2 2 2 2 4 2 1 ,-

-

2 4 8 I -

10

9

8

15

0

15

Overall, the 14 children whose data are shown in table 21.1 exhibited a strong preferencefor contraction in the control object extraction questions. Of the 68 questions elicited from this group of children, 60 (88%) showed contraction. The preference for contraction was clear in the productions of 13 of the 14 children. Just one child (SR) did not show a strong preferencefor contraction. She failed to contract in 6 of her 12 object extraction questionsand was thus responsiblefor 6 of the total of 8 questions with no contraction. The 14 children produced a total of 74 subject extraction questions. Of these, 68 (92%) had no contraction. The 6 apparent violations of the constraint were produced by 3 children; 2 by KM , 1 by SA, and 3 by KL. The ungrammaticalinstancesof wannaproduced by KM and SA could perhapshave been misheardby the experimenter, but it is more difficult to analyzeKL's productions this way, sinceas many as half of her subject extraction questionsappearto have used the illegitimate form. Of course, no instanceof contraction is easily accountedfor by the theory of Universal Grammar. But it is also true ~hat no child's responseswere in accord with the null hypothesis. No child contracted in subject extraction questionsas often as in object extraction questions. Even subjectKL contractedin all of her object extraction questions; though she contracted in only half of her subject extraction questions. The data from the children who did not meet the criteria for inclusion in the data analysis are presentedin table 21.2. A glance at these data shows that they do not include any questionsthat violate the constraint on contraction acrosswh-trace. That is, there are no subject extraction questionsin which children used the illicit wanna contraction. Most of these children were eliminated becausethey did not produce enough questions. Others produced enough questionsbut did no.t show a preference to contract (e.g., subjectsTT and RE). Still others produced questionsthat were not relevant to testing the constraint. These "other" questions were irrelevant because the chil.d used an alternative means of expressionsuch as 'What would you like to eat?" instead of 'What do you wanna eat?"

Wanna Contraction

185

21.5 Conclusion The data from one child aside, the findings from the experiment on wanna contraction support the claim that the prohibition against contraction acrosswh-trace is an innate, universal constraint. Children did not make many errors initially , as the Input Matching Model would predict. On the contrary, there were almost no errors. The children whose data were not analyzed for adherenceto the constraint did not produce questionsthat violated the principle, a result that is compatible with the Modularity Matching Model . In summary, we have provided support for an account of the contraction facts that is compatible with Universal Grammar.

Chapter 22 Long -Distance Questions and the Medial -Wh

This chapter shows how elicited production can be used to investigate the sourceof a nonadult form. The nonadult form appearedin an experiment designedto elicit longdistancewh-questions from children. In t~e course of the experiment, some children consistently produced questions with an extra wh-phrase in the intermediate CP; hence the term medial-wh questions .! Examplesof children's medial-wh questionsare given in (1).2 (1) a. b. c. d.

Who do you think who Grover wants to hug? (TI 4;9) Who do you think who's in there, really really really? (AM 4;6) What do you think what Cookie Monster eats? (KM 5;5) What do you think what the baby drinks? (MA 3;3)

Our analysis of the error is that it is consistent with the principles of Universal Grammar. This meansthat it reflects a universal option that is appropriate for other languages, but that happensnot to be correct for English. Using the criteria laid out by the Modularity Matching Model, we elicit a variety of structuresto demonstrate that the nonadult form is the question form consistent with these children's grammars. It does not reflect a performanceerror. The medial-wh questions appearedin an experiment designed to elicit questions from the "that-trace" paradigm (seeChomsky 1981). In the Principles-and-Parameters framework of Universal Grammar, this paradigm is frequently cited as showing that the constraint known as the Empty CategoryPrinciple(ECP) is at work.3 The longdistance questions that make up the that-trace paradigm are given in (2) and (3). In this paradigm, the ECPconstrainsthe appearanceof the complementizerthat. (2) Objectextraction a. What do you think flies eat t7 b. What do you think that flies eat t? (3) Subjectextraction a. What do you think t eats flies? b. *What do you think that t eatsflies? (4) Declaratives a. I think flies eat garbage. b. I think that flies eat garbage. The questions in (2), extracting from object position of a tensed embeddedclause, are grammaticalwith or without the complementizer. In questionsin which the whphrase is moved from subject position, however, the question is grammatical only when the complementizer is omitted. When the complementizer is present, as in

188

Chapter 22

the ungrammaticalquestion in (3b), that is followed by a wh-trace- hencethe name "that-trace" paradigm. The declarativesin (4) are given for comparison. The ECP is irrelevant in the declarative sentencesbecausethey contain no empty categories (such as wh-traces). Hence, the sentencesare grammatical with or without the COffiplementizer. The syntactic literature offers several different accounts of why the complementizer in (3b) makes the sentence ungrammatical (see, e.g., Lasnik and Uriagereka 1988; Rizzi 1990). For the present, it will suffice to note that the ungrammaticality of (3b) is due to a putative constraint of Universal Grammar .

22.1 Children's Knowledgeof the Empty CategoryPrinciple As researchersworking within the framework of Universal Grammar, then, we offer as our experimental hypothesis that children have innate knowledge of the ECP. On this viewI we should not expect children to make errors and produce the ungrammatical question form in (3b). The null hypothesis is that the facts of the paradigm have to be learned from the input, in which casewe should expect children to make a number of errors before they realize that complementizersare grammaticalin object extraction questions but not in subject extraction questions. The hypotheses can be summarized

as follows

:

. Ho: Children will allow complementizersin subject extraction questions as often as in object extraction questions .

. H 1: Children will not produce subject extraction questions with complementizers

.

Ideally, the samelogic anddesignthat wereusedin the wannacontractionexperiment reported in chapter 21 could be used for testing children's knowledge of the constraint on that-trace. If we follow the samelogic, the argument would be that it is not sufficient to test the ECP by examining only children's production of subject extraction questions. If only subject extraction questionswere elicited, we might be able to demonstrate that children never produce subject extraction questions with complementizerslike (3b), but data of this kind would not answer the objection that children have a preferencefor omitting complementizersin their speechin general. To demonstratechildren's knowledge of the constraint, one would want to show that children prefer to use complementizerswhen the grammar allows them, but override the preferencewhen the grammar excludes them. This logic necessitatesincluding object extraction questions in the design. As the that-trace paradigm shows, object extraction questionslike (2) are grammaticalwith or without a complementizer. Here, we would hope to seea preferencefor use of the complementizer. In the ideal situation , then, questions extracting from object position could be used to assess each

child's prefe.rence for using complementizers. If this control condition showed that children frequently use complementizersin their object extraction questions, but fail to use them in their subject extraction questions, we would have empirical evidence strongly supporHngchildren's knowledge of the ECP. In the abstract, the logic of the experiment is sound, but unfortunately it runs up against a property of the parsing system that worked in our favor in the wanna con-

traction experiment: the tendency to reduce forms. In that experiment, the tendency to reduce forms was instrumental in inducing contraction in the control object extraction questions. In keeping with the Modularity Matching Model, the tendency

Long-Distance Questions was

operative

to

see

ject

for

both

extraction

reduce

forms Matching

work

indicated

knowledge

parsing

not to

emerged so

the

of

' out

in

it

present

we

ing

the the

have

,

the

range

one

it

children

,

the

not

that

,

- wh

- trace the

para

ECP to

from

why

ECP

,

use

.

and

-

. Part the

children

and

the

are

medial

realistic

appear to

that

the

of

elicited

adhere

-

from

data

thing

was

questions

not

com

complementizers

knowledge

structures

no

.

from

though

' s

overriding

absent

provide

For

' s

nonadult or

.

children

of

experi children

use

drop

ECP

cannot ECP

even

a

, we This

in

in

what

for

test

-

.

that

we

will

assume

(5 )

(5 )

and

Object

wh

,

( situation

,

lead

- in

( sentencel

. . . . e5 . . . that

m

Output

( sentence3

"

Subject

( sentence

,

Output

( sentenCe4

Output

( * sentences

? 77

Output

( sentence6

design

explores subject

- wh

,

, as

or well

their not as

' "

simply

children

to

they

appear

to

elicit

investigate .

The

-

knowledge subject their

basic

use

design

is

. . .) . . .) . . .)

eJ

)

. . .) . . .)

. . . wh

. . .)

children

use

extraction children

questions

from

tensed

medial

- wh embedded

complementizers

questions

compare

the in

- in

. . . that

object

should

with

whether clauses

should

is

' s

)

. wh

lead

whether and

we

paradigm

reveal

experiment

used

children

extraction

Input

or

medial

design testing

extraction

( sentence2

in

the

for

from

- phrases

Output

equally

- trace

of

questions

medial

experimental

succeed

goal

Output

simple

the

the

- distance

or

(6 )

forms

complex

(6 ) .

Input

This

more cannot

that

long and

in

the

paradigm

extraction

illustrated

bedded

to

that

grammars

the

-

. Pilot

are

children

-

Modu

of

are

questions

way

show

If

to

of

the

The

particular

children

.4

preference

eliciting

some

whether

determined

complementizers

will

to

that

experiment

comparing

contraction

object

that

is

their

. A

research in

reveal

in

wanna ECP

and

of

pilot

this

ob to

tendency

complementizers

knowledge

like

tendency .

same

that

show

that

of

the

demonstration

constraint

fact

remarkable innate

circumstances

well

the

knowledge

related

, by

what

,

,

the

result

to the

would in

complementizers

convincing

possible by

grammar

exhibit the

a

not

of

, we

complementizers

Design

Since

of

is

of

,

with

elicitation

' s

chapter

may

form

the

,

experiment using

adult

should

abide

as

the

insertion

case

is

not

well

that

above

it

in

provide

to

is

be

this

discussed find

-

children course

the

questions

as

well

not

to

present for

that

children

fact

their

say

may

vestigation

of

to

the

it

goal

design

.2

order

data

probing

in ,

can

ECP

in

the

preference

is ,

used

questions the

is

in

. In

a

problem

be

the

of

extraction explain

This

22

any

adults

that

is

not of

in

relevant

of

this

preference

subject

The omission

could

plementizers

as display

predicts

that

innate

digm

.

Model

design

well

adults

encourages

mental

would

as

and

questions

larity

a

children

children

189

' s the

appears clauses

. As long

and part

- distance

wanna in

our

If

children

medial

from paradigm

with produce

- wh

investigation

questions

contraction questions

.s

of

the .

infinitival

This em

medial

- wh

-

190

Chapter 22

questionswith infinitival clauses , then they should be expectedto produce questions like What do you want what to eat? and Who do you want who to brush your teeth? This

section of the investigation is summarized in (7) and (8).

(7)

or m

Objectextraction Input (situation, lead-in) Output (sentencel. . . want to . . .) Output (sentence2.. . wanna. . .) Output (sentence3. . . want . . . wh . . .to . . .)

or

Subjectextraction Input (situation, lead-in) Output (sentence4.' . want to . . .) Output ("'sentences. . . wanna. . .)

or

Output

(8)

(sentence6. . . want . . . wh . . . to . . .)

22.3 The Experiment

The protocols designedto elicit long-distancequestionsuse the samegamedescribed for the wanna contraction experiment . That is, questions were elicited from children

by having them find out information from the rat puppet, Ratty, who is too shy to communicate

with

grown

-ups .

Long-distance questionsfrom the that-trace paradigm were elicited by having the child invite Ratty to participate in a IIguessinggame." The experiment was designed as a guessing game so that Ratty could be asked his opinion in Questions of the form -

...

"What do you think . . .7" For the part of the game eliciting questionsextracting from subject position , the child had Ratty cover his eyes. The experimenter and the child then hid a series of items (say, a toy bear, a marble , and the toy character Grover ) in

various places(in a box, under a blanket, and in a yogurt carton). Ratty was then allowed to uncover his eyes, and the guessinggameproceeded. Here, one experimental context sufficesto elicit three long-distancequestions. (9) Pri?tocolfor elicitingsubjectextractionquestions Experimenter (in low voice to child, so that Ratty can't hear) We know where all the things are hidden, right? We know that there's a marble in the box , a bear under the blanket

and we know

that Grover

is under the

yogurt carton . Let's see if Ratty can guess where we hid them . Let's do the box first , OK ? We know that there'~ a marble in the boxJ but ask the rat what he thinks . Child

What do you think is in the box ?

Ratty

Can you rattle the box for me? Hmm , I think there's a marble in the

box .

Child You're right ! Experimenter Hey, he made a good guess. Now let's do the blanket .. . (game continuesin the sameway) Notice that the experimenter does not use long-distance questions. Embedded questionsare incorporated in the protocol, and the nnallead-in to the child is also an embeddedquestion followed by an elided long-distance question. Since the crucial

Long-Distance Questions

191

long -distance question is not given as overt input to the child , it cannot be said to give the child clues to the targeted structure . The protocol for object extraction questions follows similar lines. Again , Ratty is asked to cover his eves while the experimenter and the child set up the items for him to guess about . The experimenter might set out three characters in the workspace (say, Cookie Monster , a baby doll , and a Ninja Turtle ) and then suggest to the child that Ratty guess what Cookie Monster eats, what babies drink , and what Ninja Turtles like to eat. The protocol is shown in (10). (10)

Protocolfor eliciting object extraction questions Experimenter (in a low voice to child , so that Ratty can't hear) We know about all these guys , right ? We know that Cookie Monster eats . . . Child Cookies . Experimenter And babies drink . . . Child Milk . Experimenter And Ninja Turtles like . . . Chi ld Pizza. Experimenter Right ! Now let 's find out if Ratty knows . Let's do Cookie Monster first . We know that Cookie Monster eats (whispered ) cookies, but ask the rat what he thinks . Child What do you think Cookie Monster eats? Ratty Well , Cookie Monster is a monster , so I think he eats monsters . Child No ! Cookies ! Experimenter He's silly , isn't he? Cookie Monster eats cookies, just like you said. Let's do another one . . . (game continues in the same way )

Recall that an experiment eliciting questions must satisfy the felicity conditions that go along with a question game in order to avoid the ask/ tell problem . It is often desirable to have every question request new information . That is, children should not know the answer to the question ' they are posing to the puppet . This is not the case in the protocols in (9) and (10). The child and the experimenter conspire , sharing knowledge of the answers; but this is a situation that is quite natural when asking someone to guess something . Even though the child knows the answer to the question , these protocols have not been found to present a problem as long as the game starts off in the right way . (In the warm -up, as described earlier, children begin by asking Ratty his name, how old he is, and so on, questions to which they don't know the answer.) If a problem were to arise with a particular child , however , it would be simple to modify the game. To elicit subject extraction questions, for example, the experimenter could hide the objects and have both the child and Ratty cover their eyes. The experimenter could then have them both guess, in turn , what was hidden in each place. The protocol could be adjusted to proceed something like this : (11)

Adjusted protocol for eliciting subject extraction questions Experimenter I'm going to hide some things , and then you and the rat guess. Hide your eyes! (experimenter hides objects ) OK , you can come out now . There's something in this box , something under this blanket , and so.mething in the yogurt carton . Let' s do the box first . You guess first , and then the rat can have a turn . OK . What 's your guess? Child I think there's a Smurf in there .

192

Chapter 22

Experimenter OK, you had your turn. You think that there's a Smurf in the box, but ask the rat what he thinks. Child What do you think is in the box? Ratty A hamburger? Experimenter I'll show you what's really in the box. It's Grover! In addition, "Is it true .. ." questions were incorporated in the game. These questions were elicited in order to provide another angle on children's use of complementizers. In this and other factive structures, unlike in declaratives and wh-questions .& -, adults generally have a preferencefor inserting a complementizer.6 This seemeda likely structure to use, then, to determine whether children behave like adults in this regard. It also provided a check on their lexical choice of complementizerin another structure. To elicit this structure, children questionedRatty about various rumors that the experimenterhad heard. A protocol is given in (12). (12)

Protocolfor eliciting "Is it true" questions Experimenter I've heard that rats can seein the dark. Could you find out if it's true? Child Is it true that rats can seein the dark?

22.4 Results The experimentalfindings were that a few children askedmedial-wh questionsof the form in (13c- d). The majority of the children asked questions that did not deviate from the adult forms in (13a- b). (13)

a. b. c. d.

What do you think (that) Cookie Monster eats? (adult) Who do you think (*that) is in the box? (adult) What do you think what Cookie Monster eats? (child) Who do you think who is in the box? (child)

Some of the children who produced medial-wh questions also produced occasional "partial movement" questionssuch as (14). In these questions, the true wh-phrase is found in the intermediate CP, and a dummy ~ h-phrase(which marks the scopeof the intended question) appearsin the matrix SpecCP. This structure will not be discussed at any length here (for details, see McDaniel 1986; McDaniel, Chiu, and Maxfield 1995; Thornton 1990). (14) What do you think which boy ate the cookie? The results from the experiment eliciting long-distance questions with tensed embeddedclausesare shown in table 22.1.7 There were 2 children who consistently used a medial-wh (TI and MA ), 2 children who consistently used complementizers (MO and TT), and other children who sporadically used these forms. As table 22.1 indicates, there were children who used complementizersin subject extraction questions who appearedto be violating the ECP0For the moment, we can put explanation of this problematic result asideand investigate the medial-whoThis investigation will eventually come full circle and shed some light on whether or not children's use of complementizersin subjectextraction questionsviolates the ECP. The "Is it true . . .7" structure was designed to check children s use of complementizersoThe finding was that children were much like adults. The protocol evoked the target questionsfrom 15 of the 21 children, and 14 of the 15 produced complemen-

194

that

Chapter 22

-

trace

paradigm

extracting

.

from

out

a

15

,

)

a

.

b

.

*

difference

the

wh

the

ECP

trace

in

What

do

you

think

t

do

you

think

that

(

16

)

a

.

b

There

tizers

.

are

further

We

have

tracting

from

context

either

A

Notice

)

and

(

eats

ISb

in

to

)

'

.

that

wh

-

questions

with

-

?

Comp

is

filled

properly

s

:

grammatical

.

flies

being

children

ECP

are

?

eats

is

the

clause

one

flies

t

from

that

What

.

in

(

governed

medial

-

wh

ISb

,

questions

,

)

.

This

and

prevents

a

the

violation

same

of

observation

up

be

the

-

wh

speaking

from

obligatory

long

extracting

from

'

the

and

-

other

ment

the

.

.

Cen

t

novel

-

18

]

)

use

of

complemen

their

-

questions

in

ex

this

-

syntactic

,

I

such

spec

later

-

' The

,

. ,

was

in

form

,

) ,

aL

rnheas

me

Camp

thought

I

An

in

and

.

English

-

in

McCloskey

of

the

a

special

complementizer

agreement

Both

be

subject

example

and

of

McCloskey

[ cp

was

observed

reflects

)

object

suggesting

appearance

,

-

in

.

stage

Chung

the

which

.

from

of

Fur

wh

pattern

found

first

questions

aI

change

17

patterns

the

agreement

out

This

,

to

-

-

related

the

by

wh

head

(

of

.

medial

drop

intrinsically

place

this

given

to

.

According

the

-

instructive

the

found

to

In

to

spec

but

' hornme

man

man

only

a

a

1979

wh

-

object

-

question

.

aL

duirt

se

Camp

said

he

)

que

je

crois

I

believe

I

believe

,

from

understood

[ cp

aL

Camp

head

agreement

causes

takes

speaking

the

' s

from

and

demonstrating

the

1990

.

" ' quejqui

viendra

who

.

will

.

come

.

. 1

.

.

obligatoriness

in

the

posi

-

develop

to

clauses

change

subject

grammatical

complementizer

relative

come

a

place

children

subject

Rizzi

will

? '

extraction

-

relative

18

who

who

when

in

subject

(

-

causes

place

in

he

spec

English

agreement

of

said

,

in

takes

given

he

French

stage

head

example

L

go

. e

think

as

,

,

the

it

similar

taken

also

] ] 7

did

the

An

,

i

was

complementizers

.

exhibit

[ cp

alternation

)

s

; 9

he

is

(

'

4

appear

obligatory

Compo

ursceal

novel

complementizer

,

is

has

is

se

French

.

TI

in

not

questions

questions

(

complementizer

This

questions

wh

)

(

produced

each

Irish

movement

questions

which

parallels

In

qui

nonadult

-

position

languages

of

-

extraction

showing

specifier

distance

' Which

this

11

children

medial

are

.

form

its

Later

used

intermediate

object

)

s

wh

understood

;

;

?

initially

.

data

thuig

tion

3

questions

complementizers

declarative

head

extraction

form

MA

pants

and

a

subject

languages

that

the

the

In

wh

children

who

longitudinal

signals

17

(

do

questions

in

reveal

s

in

(

-

extraction

children

children

Irish

changes

?

their

produce

some

persist

and

'

complementizer

tween

box

in

medial

object

that

for

data

-

that

ants

complementizers

extraction

same

children

) ,

That

and

but

Crosslinguistic

1987

the

.

subject

object

medial

English

in

never

showed

and

speaking

s

had

children

clauses

questions

the

'

who

between

that

testing

subject

extraction

that

what

say

parallels

noted

of

to

think

they

.

-

found

you

did

infinitival

follow

both

do

Who

comparison

ther

to

ISa

position

respect

embedded

with

What

. 8

with

tensed

.

(

(

contrast

a

ungrammatical

subject

ensues

critical

of

but

between

-

applies

the

position

complementizer

(

The

Recall

subject

change

subject

-

from

que

extraction

of

a

qui

Long-Distance Questions

195

The complementizerqueand its alternating form qui both also function as wh-words in French. This fact is important , because we will claim that the medial -wh is also a

complementizer, even though it is more akin to a wh-phrasein appearance . Rizzi (1990) arguesthat the changein complementizerform observedin languages such as French signals satisfaction of the ECP by a proper head governor. In IQngdistance questions, for example, the wh-phrase moves through the intermediate SpecCP, leaving a wh-trace. This wh-trace passesagreementfeatures to the Comp, licensing it as a proper governor for the trace in subject position of the embedded clause. In languagessuch as Irish and French, the changein complementizerform that reflectsspec-head agreementis overt. In other languages, such as English, it is argued that the process is covert. In English, for example, the same spec-head agreement processlicensesthe complementizeras a proper head governor for the subject trace, but the complementizeris null (that is, silent). The representationin (19) illustrates that the null complementizer can bear agreement features as a result of spec-head agreement and is thus a proper head governor for the subject trace of the embedded clause .

(19)

Whoi do you think [cp ti [c e5Agri UPti will come to the party ]]]7

Rizzi' s theoretical framework provides a ready explanation for the structures used by English-speakingchildren. Children who produce medial-wh questionscan be said to be using a form of the complementizerthat signals that spec-head agreementhas taken place. In the grammars of these children, the change in the form of the comJplementizer . must be ......manifested overtly, as in French. Children who insert complementizers, instead of dropping them like adults, can also be said to be fulfilling spec-head agreement . The difference is that for these children the form of the com-

plementizer signaling spec-head agreementis that, whereasother children take it to be a wh-complementizer. Left to explain is why some children might initially expressovert spec-head agreement for both subject and object extraction questions and later , just for subject ex-

traction questions. Linguistic theory provides the answer. Adult English, like French, only requires spec-head agreementto take place in subject extraction questions, because this is the only syntactic environment where it is needed to avoid an ECP

violation . In object extraction questions, for example, the wh-trace is properly head. governed by the verb in the embeddedclause, and overt expression of spec-head agreementin the intermediateComp might be consideredIloverkill." After sometime during which they express spec-head agreement overtly in object extraction questions as well as subject extraction questions, children ascertain that it is only necessary

for the latter .9

Another question is why children do not use a medial-wh or that complementizer in long-distancequestionsextracting from the wannacontraction paradigm. After all, in subject extraction questions like Who do you want to come to the party?, there is a

trace in subjectposition of the embeddedclausethat must satisfy the ECP. According to Rizzi (1990), the subject trace is properly head-governed by the exceptional-Casemarking verb want in the matrix clause. Proper head government (and assignmentof Case) cannot take place over a CP, suggesting that the verb want projects an IP and there is no complementizerposition.lo (20)

Whoi do you want [IP ti to come to the party ]7

196

Chapter 22

If so, it is not surprisingthat childrendo not producemedial-wh or that complemen tizersin their questionsfrom the wannacontractionparadigm . Themedial-wh or that complementizer is not requiredasa properheadgovernorfor the subjecttrace. Moreover, thereis no Comppositionin the structure . Fromthe fact that it is movementthroughthe intermediateCP that triggersspecheadagreement andsubsequent useof a rnedial-wh or thatcomplementizer , it follows that childrendid not produceany instancesof medial-wh in their "Is it true ... 7" questions . Although this is a questionstructure , it doesnot involve extractionof a wh-phrasethroughthe intermediateCP. Of all the structureswe haveexamined , themedial-wh or nonadultthatcomplemen tizers appearonly in long-distancequestionsin which a wh-phrasemovesfrom a tensedembeddedclause . In this syntacticenvironment , we havesuggestedthat the medial-wh or complementizer is presentto expressspec-headagreement andthe fact that the ECPis satisfied . In conclusion , the nonadultthatcomplernentizers in children's subjectextractionquestionsdo not violatethe ECP, contraryto appearances . Although we were not able to carry out a definitive experimentthat by itself demonstrated children's knowledgeof the ECP, we haveassembled an arrayof facts that leadto the sameconclusion . This wasonly possiblebecause we wereableto use elicitedproductionto gatherdatacrucialto the investigation . 22.6 Principles versus Preferences The factsfit very neatlywith a grammatical explanationof children's nonadultuseof complementizers andof themedial-whoBut how canwe be surethat the sourceof this difficultyis not languageprocessing ?It is worth reviewingthe argumentsin detail. Earlier,we notedthat childrenandadultsalikehavea parsingpreference to reduce forms. This resultsin a preferenceto contractwantto to wannaand a preferenceto omit complementizers in many syntacticenvironments .II Why, then, would a few childrenoverridethis parsingpreference andinsertcomplementizers , if not to satisfy the grammar ? .If the medial-wh and insertedcomplementizers are indicativeof a grammaticalprocess , they shouldbe obligatory, not optional. Let us takethis point up Hrst.I2 As table 22.1 shows, the experimentrevealed2 childrenwith a grammarthat producesmedial-wh questionsand 2 childrenwho alwaysused complementizers . That is, thesechildrenalwaysexpressed spec-headagreement by filling the Camp, or SpecCP .13Let us beginby examiningthe datafrom TI, a child who producedmedialwh questions . All of her subjectextractionquestionshad a filled intermediateCP (11/ 11 = 100%). In herobjectextractionquestions , theintermediate CPwasfilled 3/5 times(= 60%). Thefact that a medial-wh did not appearin all of her objectextraction questionsis not a puzzle, given the theory outlinedabove. Like TI, the other child, MA, useda medial-wh in all of his subjectextractionquestions(4/ 4 == 100%) but not in all of his objectextractionquestions(3/ 5 ==60%). Two other children, KK and SR (who were testedlater), also initially filled the intermediateCP very consistentlyin their questions ; in later sessions , however, the frequencyof occurrencedroppedoff. Could this be an indicationthat the nonadult questionsare performanceerrors? We can illustratethe changein frequencywith KK's data6In the first session , 6/ 7 (== 86%) of her questionshad a filled CP; in the secondsession , 16/24 (== 66%); and in the third session , only 5/12 (==42%).14The

Long-Distance Questions change

between

two

the

sessions

dency

to

such

use

as

in

an

a

of

forms

the

,

drop

.

In

Grammar

-

compatible

As

cally

table

.

22

Such

since

. 1

errors

.

We

grammar

.

ment

eliciting

many

trials

In

fact

.

)

a

.

subject

that

medial

subject

In

.

)

a

If

the

( 21

his

.

early

This

- wh

of

child

is

questions

as

can

the

that

in

- trace

appeared

tensed

long

in

-

of

however

a

child

the

earlier

similar

experi

-

questions

on

bed

?

- year

of

him

,

at

- old

.

children

In

this

causing

since

it

.

not

ad

situation

him

is

being

He

to

-

,

it

produce

known

whether

.

they

found

occur

The

long

A

clear

with

a

experiment

that

and

.

a

at

an

uncomfortable

that

.

short

as

to

- wh

the

4

,

is

clause

,

?

and

sentences

both

embedded

However

( such

for

appeared

,

paradigm

stress

medial

asked

errors

appear

.

revert

clearly

3

grammar

performance

of

out

being

speculation

session

performance

controls

was

was

a

-

particularly

) .

for

what

times

marbles

who

designed

by

medial

.

.

medial

questions

medial

of

a

Universal

sporadi

,

occasionally

produced

in

male

experiment

confused

What

do

you

think

what

What

do

you

think

really

( MA

digm

shown

undergraduate

- wh

.

in

complementizer

of

were

blue

from

b

clear

are

or

in

who

pushed

medial

-

adult

subset

of

eliciting

length

was

questions

,

illustration

long

-

apparently

not

provided

extrac

comes

from

-

MA

' s

.

( 22

nev

who

one

got

from

place

subjects

was

had

they

a

characterize

that

who

vestige

the

grammar

with

beginning

may

who

,

The

took

adult

) ,

adults

guess

questions

factor

( 1988

even

guess

questions

a

pro

agreement

a

keeping

insertions

they

an

- head

explanation

the

they

particular

distance

at

not

did

produced

.

effects

- wh

did

characteristic

items

data

a

- wh

Another

tion

and

children

somewhat

possible

some

lebeaux

in

being

"

but

questions

an

in

of

of

represent

performance

Who

participate

spec

,

-

by

course

aware

for

also

medial

a

Who

was

to

the

.

- up

following

the

a

to

deletions

) ,

his

rule

-

signif

explained

the

became

could

are

be

in

ten

mature

can

that

the

her

factors

could

children

the

used

questions

of

these

be

since

in

performance

change

may

reformulate

open

" warm

among

Two

b

mitted

,

rapid

It

productions

seem

that

,

to

memory

complementizers

children

do

as

session

,

to

' s

some

crossover

( 21

asked

,

noted

a

questions

or

,

) .

,

decrease

.

uses

suggested

hand

the

attributed

that

1981

dramatic

that

be

unlikely

other

them

particularly

unlikely

could

the

children

appear

of

On

was

seems

highly

- whs

these

shows

already

beginning

is

explanation

often

have

This

,

It

Comp

- distance

medial

sporadic

they

we

.

caused

the

sum

It

.

Hamburger

long

which

in

transition

a

time

( see

sessions

apart

in

.

short

variety

third

days

element

grammar

question

is

overt

such

a

So

10

limitations

developing

ducing

and

only

memory

icantly

second

were

197

- wh

involving

~ r

in

( 23

) ,

with - w

( 23

where

no h

)

3

; 3

were

a

performance

medial

error

from

questions

produced

- wh

that

box

?

really

and

the

a

a

,

infinitival

from

TI

in

really

( MA

,

3

really

; 3

)

really

what

' s

wanna

long

short

we

might

clauses

also

.

As

contraction

question

it

earlier

paradigm

from

question

expect

noted

from

the

the

.

wanna

that

a

.

Who

.

What

do

do

you

you

want

think

to

help

what

you

pigs

eat

say

the

?

( TI

cookie

,

4

; 9

there

?

The

a

appear

in

medial

- wh

contrast

is

contraction

- trace

?

)

to

,

paradigm

.

b

in

)

extraction

appeared

in

,

' s

( TI

,

4

; 9

)

para

with

-

a

198

Chapter 22

It is also true that performanceerrors are more likely to occur with greater depth of embedding, presumably becausethis imposes greater demandson memory. This leads to the expectation that inserted complementizersor medial-wh elementswould be more likely to occur in sentenceswith two embedded CP domains. However, exactly the reverseis true in the child data. In the following examples, TI and SR fill the upper CP domain, but there is no complementizerin the lower clause, becauseit is an infinitival

(24)

clause .

a.

Who do you really think who Grover wants to hug? (TI, 4;9)

b.

What do you think that Ninja Turtles like to eat? (SR, 3;11)

We observed on a number of occasions that children asked fewer questions with

a medial-wh or complementizerin their object extraction questions. This contradicts another trademark of performanceerrors: that they are more often associatedwith object extraction becausethere is greater distance between the wh-phrase and its associatedgap. For example, TI filled the CP 100% of the time in subject extraction questions but only 60% of the time in object extraction questions; MA also used a medial-wh 100% of the time in subject extraction questionsbut only 60% of the time in object extraction questions; 5R produced obligatory that complementizers100% of the time for both subject and object extraction questions in the first two sessions, but by the fourth session the CP was filled in 73% (11/ 15) of her subject extraction questions but null in 100% (4/ 4) of her object extraction questions . This result is to be

expected if our Universal Grammar- compatible proposal is on the right track. We expect children to fill the Comp in subject extraction questions as evidencethat the ECP is satisfied .

Another criterion of performance errors can now be considered: that errors occur more frequently in the productions of younger children. Although questions with illegitimate complementizersand medial-wh questions occurred sporadically in younger children's productions, there was a subset of the older children who consistently asked these questions. For example, TI consistently used the medial-wh at 4;9; another child 55, used the medial -wh in subject position at 5;9; and IE consistently

askedquestionswith complementizersat 5;1. It would appear, then, that age is not a good indicator of whether or not children use obligatory complementizersor medialwh forms. This puts an explanation of the facts in the linguistic arena and makes a performanceexplanation unlikely. It is worth pointing out that the varying age of children who use an obligatory medial-wh or a that complementizerexcludes one kind of linguistic explanation of children's errors: maturation. It is highly unlikely that maturation of the grammar could account for these children's nonadult questions given that only a subset of children made the error in the first place and that their agesvaried widely . In sum, the children who consistently filled the CP give little evidenceto suggest that their productions are performanceerrors. The sameconclusion cannot be drawn for those children who produced the occasionalmedial-wh or that-trace violation . Theseungrammaticalsentencesmeet two criteria of performanceerrors: they occurred only sporadically and usually only at the beginning of a session.

Chapter23 Why ChildrenMake Good Subjects

The experimentsreported in this chaptertake advantageof the fact that many Englishspeaking children produce nonadult constructions. One such construction is the medial-wh question discussedin chapter 22. As shown there, many 3- and 4-year-olds regularly insert a wh-word between clausesin their long-distancequestions; for example, they produce questionslike (2a- b) instead of the adult (la - b) (Thornton 1990). (1) (2)

a.

What do you think pigs eat?

b.

Who

a.

What do you think what pigs eat?

b.

Who do you think who eats trash ?

do you

think

eats trash ?

Although structures like those in (2) are not grammatically well formed in English, they are well formed in other languages. Interestingly, even after the medial whphrase is expunged from children's object extraction questions, it persists in their subject extraction questions. This developmentalsequence , too, is consistentwith the theory of Universal Grammar. As suggestedby Thornton (1990), the extra wh-phrase in children 's subject extraction questions may be an overt manifestation of spec-head agreement, which is needed to ensure proper head government of the trace of the

moved wh-phrase; such an overt "agreeing" complementizerappearsin French, for example, in the familiar que/qui alternation (Rizzi 1990). English-speaking children's nonadult productions therefore support the Continuity Hypothesis, since both the structuresthemselvesand the courseof their development fall within the boundaries establishedby the theory of Universal Grammar. The Input Matching Model cannot explain this. A second construction that will playa role in the experiments reported here is negative questions(seechapter 5). Thornton (1993) discoveredthat even children as old as 5 often form nonadult negative wh-questions and nonadult negative yes/ no

questions. For example, many children ask wh-questionslike (3b) instead of the adult (3a), and many children ask yes/no questions like (4b) instead of the adult (4a) {also see Guasti , Thornton

, and Wexler

1995 ).1

(3)

a. b.

What doesn't Big Bird like ? What does Big Bird doesn't like ?

(4 )

a. b.

Doesn 't Grover like the ice cream ? Does Grover doesn 't like the ice cream ?

The (b) structures in (3) and (4) are consistent with the theory of Universal Gram-

mar and provide compelling evidence that children are guided by it in grammar formation

.

200

Chapter 23

23.1 DiscourseLinking: Syntacticor Pragmatic ? As pointed out earlier, children's nonadult structures ate also a source of data for evaluating competing linguistic proposals. The two nonadult constructionsillustrated in (1)- (4) were used to examine a theoretical controversy in the analysis of whquestions. One direction that many researchershave taken emphasizesa pragmatic distinction between types of wh-phrases, namely, between wh-phraseswith lexical content, suchas which boys(/'full" wh-phrases), and ones without lexical content, such as who or what ("bare" wh-phrases). According to one viewpoint, there is no inherent (structural) distinction between bare wh-phrasesand full wh-phrases. It was suggestedby Pesetsky(1987) and Comorovski (1989) that the distinction is pragmatic: full wh-phrases are anaphorically linked to referents that have been previously establishedin the discourse, whereas bare wh-phrasesare not related to suchdiscourseentities (also seeCinque 1990; Rizzi 1990; Chung 1994; Dobrovie-Sorin 1990). Pesetskyrefers to full wh-phrasesas discourselinkedor D-linked. D-linked wh-phrasescarry the presupposition that a specific referenceset has been preestablishedin the discourse; bare wh-phrasesdo not. The alternative viewpoint contendsthat full wh-phrasesand bare wh-phraseshave different structural properties. One structurally basedaccountis presentedby Thornton (1995). There are two parts to Thornton's account. First, Thornton assumesthat the featuresof full and bare wh-phrasesare checkedin different projections. The features of full wh-phrasesare checkedin a projection above CP, which she calls KefP, short for ReferentialPhrase(Stowell and Beghelli 1995); the features of bare whphrasesare checkedwithin the CP at S-structure. Second, Thornton assumesthat full wh-phras~s can reach their landing site in two ways, via successive -cyclic movement or via long movement. By contrast, bare wh-phrasesonly move successive -cyclically (seeCinque 1990). It has proven difficult to adjudicate between syntactic accounts and pragmatic accounts of the distinction between full and bare wh-phrases. On the one hand, there are differencesin the linguistic environmentsin which full and bare wh-phrases can appear (e.g., full wh-phrasesare more easily extracted from wh-islands; bare whphrasespermit modification by epithets such as the hell).2 On the other hand, adult judgments waver depending on the discoursecontext (bare wh-phrasesbehavemuch like full wh-phraseswhen they are embeddedin contexts that establish a reference set, becoming more extractableand lesstolerant of epithets). From the perspectiveof the structural account, the methodological desideratum is clear: with context held constant, structural differencesshould continue to be discernible. Here is where child languagecomesinto the picture. Having discoveredthat some children produce nonadult wh-questions and nonadult negative yes/no questions, Thornton (1995) investigated thesechildren's bare and full wh-questions, anticipating that potential grammatical distinctions in landing sites and in movement options might be directly revealed in the structures children produce. Devising an elicited production paradigm in which questionswith bare or full wh-phrasescould be elicited in identical contexts, Thornton uncovered evidencein favor of a structural accountof the distinction between bare and full wh-phrases. As noted, questions with bare and full wh-phrases were elicited in the same experimentalcontexts. The only differencein the experimental setup was the lead-in sentencesthat followed the story. In the lead-in designed to elicit a bare wh-phrase, the experimenterused a bare wh-phrase:

WhyChildren MakeGoodSubjects 201 Experimenter We know that Miss Piggy kissed someone. Ask Ratty who he thinks. In the lead-in designed to elicit a full wh-phrase, the experimenter used a full whphrase: Experimenter We know that Miss Piggy kissed one of the boys. Ask Ratty which boy he thinks. The findings are from two groups of children: children (N = 3; 3;11 to 5;4) who produced long-distancemedial wh-questions, as in (2) (group 1), and children (N = 7; 4;1 to 5;4) who consistently left a copy of the auxiliary verb in their matrix negative questions, as in (3b) and (4b) (group 2). Long-distancequestionscontaining both bare and full wh-phrase,s were elicited from children in group 1. The finding was that the medial-wh was obligatory in both subject and object extraction questions with bare wh-phrases, as shown in (Sa- b). In questions with full wh-phrases, however, it was optional for; object extraction questions (see (6a- b)), but obligatory for subject extraction questions, as illustrated by the contrast in (6c) and (6d), where the symbol # indicatesthe absenceof this construction in children's productions. (5)

a. b.

(6)

a. b. c. d.

Who do you think who Miss Piggy kissed? Who do you think who kissedMiss Piggy? Which boy Which boy Which boy # Which boy

do you think who Miss Piggy kissed? do you think Miss Piggy kissed? do you think who kissedMiss Piggy? do you think kissedMiss Piggy?

In (Sa- b) and (6a,c), Thornton argued, both the bare wh-phraseand the full wh-phrase have moved successive -cyclically, leaving a medial wh-phrase (agreeing complezer) behind in the intermediate Comp. The medial-wh is obligatory for bare wh-phrasesbecausethey must move successive -cyclically (Cinque 1990). To satisfy the ECP, a medial-wh is also obligatory for subject extraction questions with full wh-phraseslike (6c); the wh..:.phrase moves through SpecCPto ensure, by spec-head agreement, that there is a proper head governor in Comp for the embeddedsubject trace (Rizzi 1990). When extraction takes place from object position, as in (6b), there is ano,ther movement option for full wh-phrases: long movement, with the wh-phrase -cyclic movement is not necessaryin order linked to its trace via binding. Successive for the trace in object position to be properly head-governed. The object extraction questions (with negation) produced by children in group 2 usually had a doubled auxiliary verb with bare wh-phrases, as illustrated in (7). By contrast, other structures (in addition to doubling of the auxiliary verb) arise in the correspondingquestionswith full wh-phrases, as in (8). (7) What did the Spacemandidn't like? (8) a. What food did the Spacemandidn't like? b. What food the Spacemandidn't like? c. What food that the Spacemandidn't like? That there are more grammatical options with full wh-phrasesin object extraction matrix questions is explained by the same theoretical assumptionsused to explain

202

Chapter 23

the pattern of children's long-distancequestions. First, full wh-phrasesmove through SpecCPor they proceed directly to their landing site, SpecRefP , by long movement. Bare wh-phrases, however, move only as far as SpecCP. Movement through SpecCP is also an option with full wh-phrases, as in (8a), although their final landing site is SpecRefP , not SpecCP.The differencebetween children and adults boils down to this: children move full wh-phrasesto SpecRefPat S-structure; adults move them at LF. The Refp projection is detectablewhen the long-movement option is pursued (see (8b) and (8c)), a possibility that obtains only for full wh-phrases. The clearestexample is (8c), where the head of CP is filled by the complementizer, that, which usually appearswhen no spec-head agreement has taken place in its projection.3 The full wh-phrasehasmoved directly to SpecRefP , skipping SpecCP. Group 2 children produced adultlike subject extraction questions with bare whphrases, as in (9), but exhibited additional nonadult structureswith full wh-phrases, as in (10). (9)

Who didn't get ants in his pants?

(10)

a. Which clown didn't get ants in his pants? b. Which clown did he didn't get ants in his pants? c. Which clown it didn't get ants in his pants?

In (9), the bare wh-phrasehasmoved to SpecCP. By spec-head agreement, it provides a proper head governor for the trace of the moved wh-phrase, thereby satisfying the ECP. By contrast, the full wh-phrasesin (10) terminate in SpecRefP ; either they proceed there directly, or they proceed there in steps, first passing through SpecCP. In (lOa) and (lOb), they have moved stepwise, in order to satisfy the ECP.4 In (lOc), however, the full wh-phrasehas moved directly from subject position to its landing site; the ECP is rendered inoperative becausethe child has filled the extraction site with a resumptive pronoun. In sum, the findings from Thornton's studies of English-speaking children are interpreted as support for a structurally based distinction between full and bare wh-phrases. The distinctions in referentiality are borne by wh-phrasesthemselvesand are not conferred on them by the surrounding context. 23.2 Conclusion

In this chapter, we have illustrated that the grammars of English-speaking children can manifest distinctions that are not manifested in the adult grammar. Children's productions, but not those of adults, make clear distinctions between bare and full wh-phrases. Moreover, becausethese productions were elicited in similar contexts, they suggestthat the relevant distinction is borne by the wh-phrasesthemselvesand is not bestowed on them by the context of utterance, at least for English. As a final remark, we would emphasizethat children's and adults' questions have the same underlying structure. The differencebetween child and adult grammarsis simply that children make overt certain syntactic processesthat are covert in the target adult grammar. This suggests that children's nonadult productions are compatible with Universal Grammar, as expectedon the Modularity Matching Model.

Chapter 24 Summary of Designs

An elicited production experiment is designedto evoke particular sentencestructures from children (or adults). A particular structure is elicited by devising situations or contexts that are uniquely felicitous for the production of that structure. The context that is presented to the child is associatedwith a specific meaning; the experiment investigatesthe form of the utterancethat correspondsto that meaning in the child's grammar

.

(1) Input Output

(meaning! ) (sentence?)

One use for this simple design would be to test whether a particular syntactic structure is .part of a child's current grammar. For example, it would be appropriate if we were investigating whether 3-year-old children can produce well-formed relative clauses in felicitous

circumstances . To do this , we would

first need to find a context

that is felicitous for production of relative clausesin the adult grammar. The task would then investigate whether children produce relative clausesin the samecontext. Other elicitation experimentsmight contrast the sentencetypes children produce, either in different situations or with the situation held constant. For example, suppose that the experimental,hypothesis was that the referentiality of the wh-phrasehas an effect on the form of the question in children's grammars. This hypothesis was tested in the study reported in Thornton 1995, described in chapter 23. That study contrasted wh-questions with referential wh-phrases(like which boy or whosehat) with nonreferential wh-phrases (like who or what). Since the experimental hypothesis is that it is the kind of wh-phrase that has an effect on the form of the question, the questionswere elicited in the samesituation, expressedas (meaning2). (2) Input

referential wh-phrase, (meaning2) (sentence?) nonreferentialwh-phrase, (meaningz) (sentence?)

The elicited production technique is also used to test constraints on form. In this case, we want to know what sentencesare not produced by the 'grammar. Let us supposethe constraint we want to test is representedas in. (3) (the pairing of sentence and meaning can be representedin either order). That is, in the adult grammar, meaning3cannot be paired with sentences . (3)

(meaning3, *sentences )

To examine children's knowledge of the constraint, we could present children with a situation associatedwith meaning3 and record what utterances they produce. If children's grammars had the constraint, they would produce the same grammatical

204

Chapter 24

sentences as adults . If their grammars lacked the constraint , we would expect them to

produce at least someinstancesof the sentenceform it rules out. (4)

Input Output

(meaning3) (*sentences )

or Output (sentence6 ) This is the design used to test children's knowledge of the structure-dependenceconstraint (or the head movement

constraint ).

To test some constraintson form, we need to take the experimentaldesign in (4) a step further. This is becausethe design does not reveal whether a particular production is excludedfrom the child's grammar. It might reveal that children do not produce the ungrammatical sentences in a situation associated with meaning3; but it does not

show that the child cannotproduce sentenceslike sentences . Once again, we face the problem of principles versus preferences . Which design we use dependson the particular constraint we are testing .

To test the constraint on wannacontraction, the above design is not s~fficient. This is becausethe sentenceform ruled out by the constraint in one situation is grammatical in another situation. More specifically, contraction of want to to wannais grammatical in object extraction questions , but not in subject extraction questions . To test

the constraint, the experimental hypothesis will be that children will not contract want to to wannain subject extraction questions. Supposewe test this with the design in (4). We provide children with the appropriate situation for a target subject extraction question and observe whe,ther they use wanna contraction or find some other

meansof expression. The problem ariseswith object extraction questions, where contraction is optional. Showing that children do not contract in subject extraction questions is not convincing evidence that it is the constraint that ruled out the contraction .

It could have been that children simply prefer to use the no-contraction option. This complication necessitatesa different experimental design, which elicits both subject and object extraction questions .

To show that children do not produce ungrammaticalcasesof wannacontraction in subjectextraction questionsbecausethe contraction is excludedby the constraint, we would first want to establish that these children use wanna contraction in the object

extraction questions, the syntactic environment where it is possible. These questions serve as a control . That is, as a first step, we would want to establish that sentences

like sentencesexist in children's grammars. For example, it could be shown that meaning1 (object extraction ) is associated with sentences (wanna contraction ). The

next step would be to show that when the children are subsequentlypresentedwith meaning3(subjectextraction), they never produce sentences(wannacontraction). The design of this experiment is depicted in (5). (5) Step 1 (meaningl) - + (sentences ) Step 2 (meaning3) - + (sentenceS ?) Recent advancesin linguistic theory have led to an increasein the amount of linguistic knowledge that is hypothesizedto be innately specified. A linguistic property is considered likely to be innate if it appears universally- and if it is mastered in the -absenceof evidencein the linguistic input. It is also reasonable , though not logically necessary 1 to expect innate properties to emerge early in linguistic developm'ent. These considerations make it important to understand why language acquisition

Summaryof Designs

205

appearsto take so long. Much of the evidencefor gradualacquisitioncomesfrom longitudinalstudiesof children's spontaneous productions . Transcriptsof children's speechseemto suggestthat complexsyntacticstructures(suchaslong-distancewhquestions ) aremasteredquitelate. This is surprisinggiventhe strongassistance they receivefrom UniversalGrammar . In our viewI theseobservationssimplyunderscore the needfor additionalsources of datain assessing young children's syntacticknowledge . Focusingon the technique of elicitedproduction, we havepresentedstriking evidenceof children's early mastery of complexsyntax, includingknowledgeof purportedlyinnateconstraints . These findingspainta pictureof the courseof languagedevelopment that is in keepingwith currentlinguistictheory. A similarconclusionaboutchildren's languagecomprehen sionwill be forthcoming , aswe turn to anothermeansof assessing children's linguistic competence : the truth valuejudgmenttask.

PARTIII

The Truth ValueJudgmentTask

Chapter25 Truth Value Judgments

This chapter begins our discussionof how child language experimentsare designed and conductedusing the truth value judgment task. The truth value judgment task is a researchtechniquethat measurescomprehensionof sentencesand discourses. As discussedin chapter 1, it is often of theoretical interest to know whether children assign more than one interpretation to certain sentences . The truth value judgment task was designedto investigate which meaningschildren can and cannot assign to sentences (Crain and McKee

1985 ). That is, it can be used to examine

whether

sentences are

ambiguousfor children. There are three main circumstances in which it is worth ascertaining whether or

not a linguistic construction is ambiguousfor children. First, it is sometimesof theoretical interest to know whether children assign the same interpretations as adults

do. It is sometimesclaimed that children analyzecertain constructionsdifferentlythan adults do; the conjoined-clauseanalysisof relative clausesis a casein point (seechapters 8 and 18). Second, it is sometimes of interest to know whether children assign fewer interpretations to certain constructions than adults do . Because children are

language learners, it is conceivablethat children sometimesbegin by hypothesizing only a subsetof the adult meaningsand later extend their interpretive options to include ones that were previously absent. We looked at one such casein chapter 14, where we discusseda study of children's understandingof onesubstitution sentences . The findings of that study were interpreted as evidence in favor of the Modularity Matching Model . Third , it is often of interest to know whether children assign interpretations in addition to those that adults assign. In some such cases, the theoretical issue is whether

sentencesthat are unambiguous for adults are ambiguous for children. In testing children's knowledge, the null hypothesis is that children lack the constraint under investigation. The experimental hypothesis is that children know the constraint. According to the null hypothesis; children should be able to assignthe kind of interpretation that the constraint prohibits. Therefore, sentencesgoverned by the constraint should be ambiguous for children, though not for adults. In the syntactic component of Universal Grammar, constraints of this kind include the principles of the binding theory. In the semanticcomponent, they include the principles that govern discourseanaphora. In the chaptersthat follow , we will look at experimentsexamining both kinds of constraint. As a matter of historical note, the truth value judgment task was specifically designed to test children's knowledge of a syntactic constraint. This task has added considerably to researchers ' understanding of children's adherenceto putatively universal, innately specified constraints on sentence interpretation .

210

Chapter 25

25.1 Advantages The truth value judgment task has several advantagesover other methodological techniquesfor assessingchildren's comprehension. Most important is the degree of experimental control that it achieves. As stated in chapter I , we follow ~ ristotle in viewing language as a mapping between sentencesand their associatedmeanings, that is, (sentence , meaning) pairs. From that vantage point, the truth value judgment task can be seen as one that enablesresearchersto control both elements of this mapping. The test sentenceis presentedto the child by a puppet, so the experimenter controls the "sentence " part of .the (sentence , meaning) equation. In addition, on each trial the experimenteracts out a story in front of the child, so the experimentercontrols the context against which the puppet's sentencemay be judged. There is an intimate connection between the meaning of a sentenceand the set of contexts in which it is true (Davidson 1967). If a child can accurately distinguish those kinds of contexts that make a sentencetrue from those that make it false, one can infer with confidencethat the child is interpreting the sentencein the sameway as adults do. The truth value judgment task enablesan experi~ enter to present various types of sentencesto children, to see, for each, the range of circumstancesin which children judge it to be true or false. Other advantagesof the truth value judgment task also deservemention. One is that children do not feel as though they are being tested. Instead, the task is designed to make it appear that the judgments of a puppet are under investigation. This is a valuable asset. It makesthe task far more enjoyable for children than many other testing procedures, especiallyones in which children think they - may - have answered incorrectly. Since the frustration that often accompaniestesting situations is eliminated, children are more relaxed and enjoy themselves, at least if they find the puppet engaging and the stories interesting. As a result, the experimenter is able to spend much more time with children, often a half an hour or even longer, in a single session. To further enhancethe entertainment value of the task, children are askedto reward the puppet for correct responsesand to let it know (by giving it a lesser reward) when it wasn't "paying close attention." Without the puppet, children generally give "Yes" responses , presumably becausethey are reluctant to say that an adult said the wrong thing. Yet these negative judgments (about what sentencescannot mean) are often the critical data. 25.2 Testingfor AlternativeMeanings The truth value judgment task can be used to determine whether children have each of the alternative meanings of a sentencethat is ambiguous in the target language. Other tasks are not able to achievesimilar results. For example, one traditional technique that hasbeen used to study children's interpretations of sentencesis the act-out task, in which the child is askedto make toy figures perform the actions mentioned in the sentencespresentedby the experimenter. In this task, therefore, the experimenter controls the "sentence" part of the (sentence,meaning) equation, but not the "meaning" part. The experimentermust infer what meaning(s) children allow by watching and coding the actions they perform with the toys. The actions children perform on a single trial are associatedwith only one interpretation of a test sentence , however. Therefore, on any given trial, the act-out task

Truth Value Judgments

211

can only confirm the hypothesis that the child allows a particular interpretation. What cannot be inferred is that the child does not allow other interpretations. Moreover, it seemsreasonableto supposethat even if a sentenceis ambiguous for a child, in the sensethat her grammar can give it more than one representation, she may have a preferred analysis. Therefore, if we use an act-out task to study ambiguoussentences , we might find that a child always, or almost always, acts out just one reading. This could be true even if the ~ preferencefor one reading is slight, becausethis small preferencemight prevail on the vast majority of trials. Therefore, the act-out task might fail to give evidence for the alternative reading. Nevertheless, it would clearly be wrong to infer that the child doesn't know the alternative analysisof the sentence. To make matters worse, it is likely that the vast majority of children have the same preferencesfor resolving ambiguities, so that few responsescorresponding to the dispreferredinterpretation of an ambiguoussentenceare ever displayed in an experiment. The Dreference for one reading .. - of a sentenceamong others may be based on a large number of different factors, for example, processing considerations, parsing strategies, situational preferences , and/or the syntactic or semanticcomplexity of particular readings. It is well known that adults have such pref~rences, at least in contexts that are consistentwith more than a single interpretation. It seemsreasonableto assumethat the sameis true of children. In situations correspondingto a single interpretation, of course, a child or adult will assign the appropriate interpretation. But in situations that are compatible with more than a single reading of a sentence, one interpretation might consistently win out. Becausethe "meaning" is not controlled in the act-out task, children's responsesare likely to be influencedby the kinds of factors that make one reading preferred over others. The truth value judgment task is an aid in this situation.1 25.3 TestingAmbiguousSentences In this task, the child is presentedwith an ambiguous sentencein circumstancesthat make one of its interpretations true, and one false; the child's task is to indicate, by rewarding a puppet (who utters the test sentence) whether the sentenceis true or false. Contexts correspondingto more than one interpretation can be presented. Suppose that the child has a preferencefor one interpretation over another and that the preferred interpretation corresponds to the "false" reading in the context that has been presented. It has been found that many children neverthelessgive a high proportion of positive responses.The assumptionis that children want the puppet to say things that are true. That is, the child prefers to say "Yes" if possible. This bias to say "Yes" is apparently enough to boost the dispreferredinterpretation of an ambiguous sentencein the child's mind, thereby making it easier for the child to generate that interpretation. Especiallyif the preferencefor interpretation A over interpretation B is slight, we expect that presenting a context corresponding to interpretation B boosts its 'availability to the point that the child will easily be able to generateit . Thus, use of the truth value judgment task should allow us to obtain evidence for both readings, if the child's grammarmakesboth readings available. The truth value judgment task involves two experimenters. One acts out stories with toy charactersand props; the other plays the role of a puppet (Kermit the Frog). Following each story, the puppet says to the child what he thinks happenedin the story. This is how test sentencesare presented. The child's role in the experiment is

212

Chapter 25

to indicate whether the puppet saidthe right thing or not. If a child informs the puppet that it said the wrong thing, the experimenter agrees, and asks the child to explain A . why the puppet was wrong: "What really happened?" This is the elicitation component of the task; it enablesthe experimenter to tell whether or not the child understandsthe puppet's description of the story and is rejecting it for the right reasonor for someother reason. Other probe questionsare also useful. The probe questions that follow eachnegative responseby a child can also serve as a way of eliciting certain constructionsfrom children. Becausethe puppet describes the context incorrectly, specificdescriptionsthat explain "what really happened" are usually forthcoming. For example, the negative polarity item any cannot appear in the VP associatedwith the quantifier every. To test children's knowledge of this linguistic phenomenon, the puppet might make the statement, "Only two of the boys had any food," in a context that supported the following kind of correction: 'INo, every boy had somej"'any." 25.4 TestingConstraints The truth value judgment task can be used to investigate the possibility that children's grammarspermit them to assigna sentencemeaning that is ruled out by a constraint. A child who lacksthe constraint should allow the meaning that the constraint prohibits.2 It follows from the null hypothesis, according to which children lack the constraint, that children will assign a specificinterpretation that adults do not assign to the test sentences . The experimental procedure is to test whether both interpretations of a test sentenceare availableto children. It is also useful to test a number of adults, who serve as a control group, to make sure that adults, at least, perform as expected on the experimental hypothesis. By assumption, adults have all of the linguistic constraints. Therefore, any interpretation ruled out by a constraint will not be available to adults; the test sentenceswill not be ambiguous for them. For children, by contrast, the null hypothesis is that the constraint is absent from their grammars and that for them, the test sentencesare therefore ambiguous. Taking the null hypothesis as the starting point, the truth value judgment task makestwo alternative meaningsavailable for each sentenceon eachtri-al. This is accomplished in the context of a story that is acted out in front of the child. On one meaning, the sentenceis an accuratedescription of something that happenedin the story that was acted out on that trial. On the other meaning, it is an inaccuratedescription of something that happenedin the story. The experimenterconstructs contexts,that correspondto the meaningsin question, to seeif the child acceptsor rejects the test sentencein such contexts. If the child rejects a sentencein contexts that correspond to the meaning that is ruled out by the constraint, but acceptsit in contexts that correspondto meaningsthat are not ruled out by the constraint, theseresponses are taken as evidencethat the constraint is part of the child's grammaticalknowledge. On the other hand, if a child acceptsthe test sentencein inappropriate contexts, this responseis taken as evidencethat the child's grammar lacks the constraint. The subjects' task is simply to respond "Yes" or "No/ depending on whether they judge the sentenceuttered by the puppet to be true or false. Earlier we indicated that the meaning that is admitted by the adult grammar should correspond to the "No" response. If children's grammarmatchesthat of adults, aspredicted by the Modularity Matching Model, they will reject the target sentenceas an incorrect description of

Truth Value Judgments

213

the context. Children's responsesto the follow -up question 'What really happened?" ensurethat they are rejecting the sentencefor the right reasons. The meaning that is prohibited by the constraint should correspondto the "Yes" response. This ~tep ensuresthat the test of the experimental hypothesis is conservative; that is, it reduces the risk of making a type I error (concluding that the experimentalhypothesis is correct when in fact the null hypothesis is correct). It is well establishedthat subjects have a tendency to say "Yes" if they do not understanda test sentence.If the correct answercorrespondsto a "Yes" response, then children's responsesmay be counted as evidence for the experimental hypothesis when in fact they do not understand the test sentencesat all. To avoid the possibility that children will give the right answers for the wrong reasons, this bias must be circumvented. Therefore, the experimental hypothesis (that subjects know the principle under investigation) should be associated with the "No/ response . To summarize, the truth value judgment task is particularly useful in assessing children's knowledge of constraintson the meaningsthat can and cannot be assigned to sentences . The reasonthat this task is useful in investigating children's knowledge is that. it enablesthe experimenterto control both parts of the equation, sentenceand meanIng. 25.5 A Disadvantage One frustration with the truth value judgment task is the time needed to complete a study . A typical session takes about a half an hour . During this time period , an experimenter can administer 8 to 12 trials . If 1 trial is used as a warm -up, and there are 4 test trials and 4 control trials , then the allotted time is exhausted . It would be ideal to test each child on eight target sentences, or even more , and to test many subjects. However , given the practical constraints in administering the task, we sometimes limit an experiment to 4 test trials , and we test about 20 subjects. Subjects are quite consistent in their responses. Because the results are superior to those obtained with other tasks, and because the test can easily be extended to include more subjects and trials , the limitation of the truth value judgment task is only a practical one, and not an inherent problem with the task itself .

Chapter 26 Backward Anaphora

Consider the following sentence: He thinks the Troll is the bestjumper. Every adult speakerof English knows that the pronoun he in this sentencedoes not refer to the Troll ; its referent must be someoneelse. Linguists explain this using one of the principles of the binding theory, that component of Universal Grammar concernedwith

anaphoricrelationsamongNPs. The principlethat determinesthat a name(e.g., the Troll) cannot be the referent of a pronoun in certain structural configurations is Principle C. SincePrinciple C is a negative statementabout illicit anaphoricrelations, it is a constraint on meaning: (sentence, *meaning). The sentenceitself, He thinks the Troll is the bestjumper, is perfectly acceptable ; it simply cannot receive a particular interpretation. BecausePrinciple C is a constraint, it is a viable candidateto be viewed as an innately specifiedaspectof linguistic knowledge. As a result, it is anticipatedto be a linguistic universal, not masteredby children on the basis of their linguistic experience. Granting these assumptions, we are led to ask whether Principle C shows up early in the course of language development. The truth value judgment task has proven .to be a useful technique to addressthis question. In this chapter, we review the findings of previous experimentalinvestigations of the acquisition of Principle C. First, sometheoretical background. 26.1 TheoreticalBackground Principle C governs the anaphoricrelations between pronouns and referring expressions (e.g., namesand definite descriptions). We observedthat in some constructions, a referring expression (r-expression ) cannot corefer with a pronoun , as in (1). (1)

He thinks the Troll is the best jumper .

Here , the pronoun

he cannot refer to the Troll ; it must refer to someone else -

Cookie

Monster, perhaps. However, if the order of the pronoun and the nameis switched, as in (2), then coreferencebetween them is permitted. In (2), the pronoun can be interpreted as referring to the Troll . There is also another reading in which it refers to someoneelse- again, perhapsCookie Monster. (2)

The Troll thinks he is the best jumper .

Principle C explains why (2) tolerates an interpretation that (1) does not. Without considering further evidence, instead of Principle C, we might propose the following hypothesis to explain the facts discussedso far: (3) If a pronoun precedesa referential NP (an r-expression), then they cannot refer to the samething.

216

Chapter 26

In symbols , we can represent this as the following , linear prohibition :

(4) *prOi.. . NPi In

this

representation

script

that

independent characteristic

ing

to

the

evidence

fact

that

have

the

like

this

theory

that

the

,

NPs

( pro

reference

which

of

this

two

same

and

. As

treat

in

sentences

Universal

as

Grammar

hypothesis

NP

noted

is

.

It

incorrect

.

The

Trolli

ate

a

)

have

strings

is

the

chapter

same

20 of

,

not

-

words

therefore following

sub

structure ,

are

surpris

-

example

will

:

(5 )

this

that

tural

While

hei

sentence in

The

some

,

. The

(6 )

reading

pronoun a

structural

Principle

can ( 1 ),

relation invokes

of

he

pronoun

between

that

definition

was

the

cases

difference

constraint ing

of

find

suffice

the

they

hypotheses

not

In

,

indicates

paper

can

refer

precede

on

that

the

the

C

will

to a

one

the

is

hand

.

and ,

and

(3 )

be and is

Principle

bagel

.

Example

still

coreference

called

suffice

Troll

name

determines

c - command

Principle

, the

C . For

(5 )

therefore

coreferential (5 ), called our

on

shows with

the

other

c - command purposes

it ,

is ,

,

the

.

struc

and follow

the -

:

C

An r-expression R cannot be coreferential with a pronoun P that c-commands it .1

As (6) makesclear, our concern is with r-expressionsthat are c-commandedby NPs that are pronouns. At a coarse level of description, a pronoun c-commands an rexpression if the pronoun is positioned higher than the r-expression in the phrase structure analysis (tree diagram) of the sentence. A more precise definition is that a pronoun Pc -commandsan r-expressionR in a phrasemarker if and only if there is a path beginning at P, proceedingupward to the first branching node above P and then extending downward to R. Thus, Pc -commandsR in (7) but not in (8).

(7)

he.,

/ / / "-" " " "

IP

/ / / / ~ ""' "

R

VP

the Troll *i/j ........../ / / " '--' " " , is the best jumper

BackwardAnaphora

217

(8)

thinks

CP

rhe ' p VP ifjis /"the "'/best """ " jumper

/ / /~""' " //

IP

""' """

(In the indexing notation used here, which replacesthe underlining notation used in previous chapters, two NPs that have different referents are given different indices (i,j , k, . . .), and two NPs that have the samereferent are given the sameindex. The Troll*i/j and hei in (7) meansthat the Troll and he cannot refer to the sameindividual (*i, i) but can refer to different individuals (i,j ).) When a pronoun (e.g., he in (7)) c-commandsan r-expression(e.g., the Troll), coreference is not permitted; on the noncoreferential interpretation, which is permitted, these expressionshave different indices (i,j ), the result of applying Principle C. By contrast, Principle C does not apply to the structure in (8) becausethe pronoun does not c-command the r-expression; therefore, coreferenceis possible. Principle C is a negative statement; it stateswhen coreferenceis not permitted. As another example, notice that in (5), V'\1hilehewasreadingthepaper, the Troll atea bagel, the pronoun does not c-commandthe r-expression, becauseit is deeply embeddedinside a subordinate clause, as shown in (9). Therefore, Principle C does not apply, and coreferenceis permitted between the pronoun and the r-expression. (9)

pp

/

/

/ / "

'

~

p while reading

he the

i / j was paper

SincePrinciple C is a constraint within the theory of Universal Grammar, it is expected to exhibit all of the hallmarks of innate specification. It should be universal; it should emerge in the absenceof linguistic experience; all children should abide by it, regardless of differencesin primary linguistic data and despite its apparent

218

Chapter 26

complexity (pretheoretically speaking); it may appearearly in the course of language development. The experiments reviewed in the rest of the chapter were concerned with the hallmark of early emergence. As we will show , several early studies of Prin -

ciple C indicated that it did not bear this hallmark. Fortunately, a subsequentstudy using the truth value judgment task did confirm the predictions of the theory of Universal Grammar, namely, that children's grammarsdo include Principle C. 26 .2 Previous Research

There is a large body of researchon children's knowledge of coreferencerelations between pronouns and r-expressions. The findings of the initial studies of children's knowledge of Principle C were somewhat disconcerting. Researcherswere led to conclude that, at least initially , children apply a purely linear (structure-independent) hypothesis in interpreting pronouns and r-expressions. Summarizing one such experiment, by Lust (1981), Solan (1983) concludes that children's initial hypothesis about anaphorais to prohibit coreferencebetween a pronoun and an NP whenever the pronoun appearsfirst. This conclusionwas basedon an act-out study by Tavakolian (1978) in which children were instructed to act out the meanings of sentences

using toy figures and other props placed in the experimentalworkspace. Tavakolian found that two -thirds of the 3- to 5-year-old subjects responded in a manner that, according to Solan, indicated that children's grammars include a linear prohibition againstbackward coreference(cf. the directionality factor of Lust, Eisele, and Mazuka 1992 ).

In the figure manipulation experiment by Tavakolian, the majority of the time children selected an animal that was not mentioned in the sentence, but was present

in the workspace, as the referent of the pronoun in sentenceslike (lOa) and (lob ). (10)

a. For him to kiss the lion would make the duck happy. b. That he kissedthe lion made the duck happy.

Of the 24 subjects, 14 consistently acted out these sentencesin this fashion. Solan concludesfrom this that "children use direction rather than structural principles in restricting anaphora, .. . never allowing backward anaphora" (pp. 83- 84). This conclusion

is unwarranted

, for several reasons . Firs 't of all , since one - third

of

the subjects' responsesindicated acceptanceof backward anaphora, the statementthat children never allow backward anaphorais too strong. But supposethat every child had chosen. an unnamed referent on every opportunity . Even this would not be evi -

denceof a prohibition against backward anaphora. The possibility exists that there is a strong preferencefor the deictic interpretation of the pronoun (according to which the pronoun

refers to someone who is not mentioned

in the sentence ) over the back -

ward anaphorainterpretation (according to which the pronoun and an r-expression within the sentenceare anaphoricallylinked).2 It seemsmore likely that the apparent lack of availability of backward anaphora was simply the result of children's preferenceto assigna discoursereferent to a pronoun as quickly as possible, without waiting for linguistic information about the intended referent. This kind of preference, which conservesmemory resources, should also show up in adult sentenceprocessing, according to the Modularity Matching Model. The pressureto interpret constituents. as rapidly as possible will be encountered by children (and adults) on every trial ; therefore , we would expect to see a

Backward Anaphora

219

strong bias againstbackwardanaphorain an act-out task. By contrast, aswe will show, the truth value judgment task can be used to boost the availability of the backward anaphorainterpretation. This is one of its virtues; as compared to other tasks. It is important to note that if children actually have a linear prohibition against.backward anaphora, then we cannot test their knowledge of Principle C using sentenceslike (1). To provide evidence of children's knowledge of Principle C, we must show two things: (a) that children allow backward anaphora wherever adults allow it and (b) that children disallow backward anaphorawhen it is excludedby Principle C. 26.3 A Positive Finding

The question of children's knowledge of the backward anaphora interpretation of sentenceswas pursued in an experiment by Crain and McKee (1985). In this experiment , children encountered sentences like (11) in circumstances appropriate to the

backward anaphora (or intersentential) interpretation and the deictic (or extrasentential ) interpretation .

(11) While he was dancing, the Ninja Turtle ate pizza. The experimental procedure was the truth value judgment task. Ambiguous sentencessuchas (11) or its equivalent were presentedon two separateoccasions , in two contexts. Following both situations, the puppet, Kermit the Frog; uttered the sentence in (11). In one context for (11), the Ninja Turtle was dancing and eating pizza. This was the situation appropriate for the backward anaphorareading. Becausethe backwardanaphorainterpretation was dispreferredin previous experiments, children's knowledge of this interpretation was tested in contexts in which it was true. In the other context presentedto children, someoneelsewas dancing while the Ninja Turtle was eating pizza. The context also made (11) true . Presented in this context , the sentence was a control to check whether children would accept the .deictic interpretation

of the pronoun more often than the backward anaphora interpretation. Since both interpretations were presentedin contexts that madethe test sentencestrue, children's preferencefor one or the other interpretation could readily be assessed . The results were that the 62 children tested (mean age 4;2) acceptedthe backward anaphorareading 73% of the time, in appropriate contexts. The extrasententialreading was acceptedonly slightly more often, 81% of the time. Only 1 of the 62 children interviewed in this way consistently rejected backward anaphora. On the basis of these findings, we can conclude that children's grammarsallow backward anaphora, just as adults' grammars do. By boosting the availability of the backward anaphora reading, the truth value judgment task allowed Crain and McKee to demonstratethe existenceof a reading that did not surfacereadily in the act-out task where children were forced to choose between interpretations .

In this study, knowledge of Principle C was tested using both one-clausesentences like :'He washed Luke Skywalker " and two -clause sentences like " He ate the ham-

burger when the Smurf was inside the fence." It was found that children rejected coreference between the r-expression and the pronoun 90% of the time for the

one-clause sentencesand 84% of the time for the more complex two-clause sentences.3 This finding underscoresour contention, stated in chapter 3, that neither children nor adults find it more difficult to comprehend sentenceswith two clauses than sentenceswith a single clause.

220

Chapter 26

The findings support the view that children do not rely solely on their linguistic experience in making judgments about the appropriate mappings of sentences with their meanings . Presumably , there is nothing in children 's experience to tell them that certain sentence/meaning pairs are not allowed ; therefore , they have no way to learn the structural constraint prohibiting coreference, Principle C. Nevertheless , they appear to know it at an early age. By the logic underlying the poverty -of-the-stimulus argument , we are led to conclude that children 's knowledge of Principle C is innately specified.

Chapter 27 Fundamentals of Design : Principle C

Having shown that children can accessbackward anaphora, we are now in a position to describe how the truth value judgment task can be used to probe for children's knowledge of Principle C. In subsequentchapters, we will discussother experiments, both real and manufactured (to make points that could not be made otherwise ). In each case, experiments have been selected because they illustrate ways in which the

truth value judgment task can be modified to addressspecifictheoretical issues. To illustrate how the truth value judgment task can test children's knowledge of Principle C, we will concentrateon sentence(1). (1)

He thinks the Troll is the best jumper .

a. b.

)f.~ thinks tbe Troll is the best jumper. ~

thinks the Troll is the best jumper .

By Principle C, (1) cannot be paired with the meaning in (a), according to which the T roll thinks that he, himself , is the best jumper . Let us call this meaning of the sen-

tence meaningl. The meaning in (b) is permitted by the adult grammar. On this meaning , some salient male, other than the Troll , thinks that the Troll is the best jumper . We will refer to this as meaning2 .

The experimentalhypothesis is that children know Principle C, in which casethey disallow meaning1 and can assignthe' sentenceonly meaning2. The null hypothesis is that children lack Principle C. If so, they should allow meaningl in addition to meaning2i that is, (1) should be ambiguous for them. This chapter explains how to test children's adherenceto Principle C, by probing for both meaning1 and meaningz. The null and experimentalhypotheses, and the predictions associatedwith eachof them, can be summarized

as follows :

. Ho: Children lack Principle CExpected results: Children permit both meaning! and meaning2. HI : Children know Principle C. Expected results: Children permit meaning2, but not meaning! 2 7.1 Design Featuresof the Task \

The general strategy IIA void type I errors" dictates much of the experimentaldesign of the study and many of the specificdetails of the stories. To avoid such errors, researchersshould be conservative in designing an experiment, stacking the cards against the experimentalhypothesis and in favor of the null hypothesis. Part of the story acted out by the experimenter correspondsto meaningl of the test sentences , and part correspondsto meaningz. On the null hypothesis, children

222

Chapter 27

have accessto meaningl. To be conservative, the "Yes" response should be associated with meaningl; by making-the test sentencetrue on meaningl. The reasonfor associatingmeaningl with the "Yes" responseis to avoid type I errors. It was observed in chapter 6 that both children and adults manifest a bias to Qive a "Yes" reu - -- - sponsewhen they are confusedor fail to comprehenda sentence. Therefore, the part of the story corresponding to meaning1 should yield the "Yes'f response. If the null hypothesis is correct, children will say "Yes'f in responseto the puppet's assertion, either becausethey determine that the test sentenceis true on meaning! or for some extraneousreason(e.g., they do not understandthe sentence). The part of the story corresponding to meaningz should be designed to evoke a "No" response; the test sentenceshould be false in the context, on meaningz. If the experimental hypothesis is correct, children know Principle .. C. BecausePrinciple . L C rules out meaning!, children should not accessthat meaning. Instead, they should accessmeaning2, which is false in the story. Children should therefore say "No" in responseto the puppet's assertion. The discussionso far is summarizedin (2), where the asteriskin front of Meaning! indicatesthat this meaningis not allowed by the adult grammar. (2) *MeaningI' true: The Troll thinks he is the best jumper. Meaningz, false: He (not the Troll ) thinks the Troll is the best jumper. Having sketched these aspects of the design, we will describe how we might go about testing children's knowledge of Principle C, using sentence(1).

27 . 1 . 2 On

Truth

meaningI

curate

, the

readings

,

corefer

.

The

of

meaning2

Child

Following

child

B

choice

On

that

the is

C ,

is

story

out

the

by

Principle

meaning2

mentioned

as the

ruled

. On

by

winner

referent

name

). of

, the

On

(B

C , the

test is

the

meaningI

the

pronoun

the

Frog

test

sentences

, in

sentences are

are

false

contestant

. On

who

the

pronoun

the

target

loses

and

the

sentences

ac both and

name is

the

or

re -

, J.

27 . 1 . 3

signs

of

protagonist J' s

judge

minder

reading

descriptions

protests

the

Values

story

the

story

( this

is

, if

, the

's }

assigns

Task

the

Kermit

meaning who

's

to

puppet

test description

the

test

meaning2

, Kermit

sentence

) . The of

sentence should

the

child

story

is

should say

" No

, says

' s task correct

say "

what is

and

" Yes give

or "

he

to

indicate incorrect

and Kermit

give

thought , by . A

happened reward child

Kermit

a

a reminder

.

who reward

in

as . A

Fundamentalsof Design

223

27.1.4 The Conditionof Falsification To avoid type I errors, the experimenter should make the test sentencefalse on the meaning that is consistent with the adult grammar- the meaning that is not ruled out by the linguistic constraint under investigation. Take (1), for example: He thinks the Troll is the bestjumper. To make this sentencefalse in the story on the adult interpretation (meaningz), the judge, J, must not think that contestant B, the Troll , is the best jumper. The researchstrategy of making the test sentencefalse on a particular reading will be called the conditionof falsification. To meet the condition of falsification, the context should make the negation of the test sentencea true description of the story: in the presentcase, He doesn 't think the Troll is the bestjumper. To implement the condition of falsification, the context must provide a salient male figure, besides the Troll , to be the referent of the pronoun, he. In the present story, this is Robocop, the judge. By the end of the story, it must be made clear to the child subjectsthat Robocop doesn't think that the Troll was the best jumper. This is done by having Robocop considerwhether the Troll or another character, Grover, is the best jumper. After watching both characters , Robocop concludesthat Grover, not the Troll , is the best jumper. Now we can return to meaning! . The story must also make the test sentencetrue on the interpretation ruled out by Principle C, meaning! . It must be true that the Troll thinks that he, himself, is the best jumper. This is straightforward. In the story acted out by the experimenter, the Troll disagreeswith Robocop's decision. He says, "You're wrong, Robocop, I am the best jumper." 27.1.5 A Recordof Events Another feature of experimentaldesign has proven valuable in obtaining reliable results from children. At the end of the story, the charactersshould be situated in a way that reminds the child subjects of the events that have taken place. In the present story, for example, when Robocop judges that Grover, not the Troll , is the best jumper, Robocop ends up standing besideGrover. The final arrangementof the props should also provide a record of the events that took place. For the story correspondingto (I ), the relevant props are prizes- colored pasta. Robocop awards a prize to Grover when he judges him to be the best jumper. The Troll protests, however, and awards himself a prize. Therefore, at the end of the story both Grover and the Troll have colored pastabesidethem, reminding the child that Grover was the winner, but that the Troll thought that he was the winner.

27.1.6 TheOrderof Events If childrenaccess theillicit readingof thetestsentence , thentheyshouldjudgethe testsentence to be true. Therefore , a "Yes" response is evidence that a childlacks PrincipleC, whichrulesagainstassigning suchaninterpretation . Theorderin which thetwo partsof thestoryareactedout is important . Specifically , wehavechosen to maketheeventcorresponding to meaning ! comelastin thecontext . Themotivation is thesameasthatusedin selecting meaning ! to be themeaning thatis truein the context(corresponding to the"Yes" response ): meaninglis actedoutlastin orderto avoida typeI error. Presumably , thelast-mentioned eventis themostsalient , sothe interpretation associated with this aspectof the contextwill be favored , all other thingsbeingequal .

224

Chapter 2 7

To

recap

children sponse

} is available

the

story

present

, because

with

the

said

, if meaning

in

the

action right

Putting

the

it

sentence

that thing

is

way

null

hypothesis

, which

uous

, allowing

meaning

of

their

they

way

do

dence

in

judge

to

not favor

, Robocop

27 . 1 . 7

The

Linguistic

experimental

that

appear

potential ing would and three

say

test

something

other

Troll

are should

was people in

was

the the

meaning

it should

hypothesis

that

on

meaning

} . If

subjects

against

the

} . This

are

. On

puppet of

, we

, that

the

that

is the

jumper

, which

best

can

would

hypothesis

Troll

test

hand

experimental the

that

the

puppet

feel

the ambig

children

be

go

confident

means

is false

-

out that

compelling

sentence

.

rejecting

sentences , if

ends

hypothesis

against

the

to

" Yes " re story

experimental

other

responses

the

.d , the

indicate

find the

accessible

with

evidence

' s sentence

pattern

readily

} . Secon

the

children

as meaningz

be

, it is associated

evi that

the

.

Antecedent

the

children

Robocop

happened the

in

story

as evidence

reject

protocol

referents

that

the

, thinks

The

by

} as well

meaning

of

the

. First

' s " Yes " responses

is the

consistently

permit

in

counts

, children

, then

reasons

is true

this

children

two

expressed

, then

another

to

, for

should sentence

make

introduced assign

judge

: Grover story best

if they was

know a good he

, Cookie

Monster

would

won

then

available

avoid

not

, because

." Kermit jumper

, to

should

like , " That the

referents

. Again

be

type biased

Principle story the

. It

the

was

Troll

the

test

for

the

present

about

pronouns

order

meaningz

before

the

child

, the

toward C . In

contest

, and present

to I errors

in

which

, the

mean

story

a jumping

contest

. Let ' s see . There

. I know sentence

-

, Kermit ,

were

one

thing

that

: " He

thinks

that

."

27.1.8 What ReallyHappened Whenever children correctly reject a test sentence,they are askedto explain to Kermit why his description of the story was inaccurate. This is done by asking them, 'What really happened?" Children are told that Kermit needs to pay closer attention and that they can help him by explaining to him what part of the story he misconstrued. In the present experiment, this is accomplishedby designing the story to make it clear exactly why the test sentenceis false. Put the other way around, it must be clear why the negation of the test sentence(i.e., He QQ~~!!l think the Troll was the best jumper) is true. In the present story, this is accomplishedby having Robocop consider whether the Troll or Grover was the best jumper, finally settling on Grover. Children's explanations of what really happened are extremely useful. They provide the experimenterwith a'verbal report about the events that took place, and they may identify the referents of pronouns in the test sentences . For example, in the present story children would be expected to point out that Grover was the best jumper. If children do not provide revealing accountsof the story, the experimenter can follow up their responsesto 'What really happened?" by further questions, such as 'Why does the Troll have the pasta?" and, perhaps, (a crossover question probe) "Who did he say was the best jumper?" Children's answersto these questionsprovide a more detailed account of how they understood the story. Children's responses to the crossoverquestionsusually identify the characterthat they think is the winner of the contest, and sometimes indicate who they take to be the referent of the pronoun he in the test sentence. Children's answers to the question about the Troll sometimeselicit a sentencecorrespondingto meaningI, for example, "The Troll thinks he is the best jumper."

Fundamentalsof Design

225

27.1.9 TheConditionof Plausible Dissent(Russell 's Maxim) Considernow what would happenif we madethe following changein the experimentalcontext. Supposethat in the story actedout in front of the child, Robocop doesnot judgewho is the beStjumper. Supposehe just standsthere, for e~ample.At first glance , one might speculatethat this situationwould sufficeto makethe test sentence false, applyingthe following logic: if Robocopdoesn ' t sayanything, thenit ~h()u1dnor be inferredthat he thinksthe Troll is the bestjumper. Thereis an importantdistinction,however, betweenfalsity andinfelicity. If Robocop doesn't say anything, then it is infelicitous , not simplyfalse, to askwhetherhe thinks the Troll is the bestjumper. We havefound that childrenare confused , and makeIIerrors," in situationslike this. In our view, sucherrorsare the result of the experimenter 's putting childrenin the position of having to violate a grammatical constraint(e.g., PrincipleC) or a pragmaticone. Pragmaticconstraints~tate the conditions underwhich the sentences areusedin ordinarydiscourses . It turns out that childrensystematicallyassignthe correctinterpretationto sentences (and produce them) only in contextsthat arepragmaticallyappropriate . Thus, whenthe goal is to assesschildren's comprehension of sentences , it is importantto considerthe pragmaticcontextsin which the sentences areordinarilyused. If a child fails to perform accuratelyin responseto someexperimentaltask, it is conceivablethat the fault lies somewhere in the experimentalsetup, andnot in the mindjbrainof the child. Crain, Thornton, et al. (1996) put it this way: Childrenmakeerrorsin experimentalsituationsthat forcethemto violate one kind of linguisticprincipleor another.In suchcircumstances , childrenareforced to choosewhichkind of principleto violate. However, if childrenchooseto violate a syntacticprinciple, say, in orderto providea pragmaticallyfelicitousresponse , they shouldnot be saidto lackknowledgeof the syntacticprinciple.... Of course , the proof is in the pudding. To substantiate the claimthat somesystematicbehaviorby childrenis dueto a flaw in experimental design, the claimant needsto rectify the flaw andshowthat the error vanishes . (p. 109) A basicprincipleof pragmaticsis that sentences mustbe relevantto the discourse in which they appear . In the context under consideration , where Robocopis just standingthere, whetheror not "He thinks that the Troll is the bestj~mper" is not pertinentto the discourse . Therefore , this contextforceschildrento choosebetween violating a pragmaticconstraint , "Be Relevant ," and a syntacticconstraint , Principle C. It is hardlysurprisingthat childrensometimes chooseto violatethe syntacticconstraint. They may conjecturethat Robocopwas thinking that the Troll wasthe best jumper, evenif he did not say so. This conjectureis consistentwith the biasnoted earlier, to attemptto assignan interpretationthat makesthe puppet's statementtrue. The upshotis that it must be clearto childrenwhy a statementis true or false, if they are being askedto confirmor deny-it. If it is not clearwhy a statementis false, childrenmay end up acceptingit, despiteknowing the grammaticalprinciple under investigation(cf. Wason 1980). Therefore , clarifying the reasonfor denying the test sentencehelpsguardagainsttype II errors, rejectionsof valid experimental hypotheses . We referto this featureof the experimentasthe conditionof plausible dissent . In the presentexperiment , this conditionis satisfiedif the story makesit clearthat the Troll

226

Chapter 27

is under considerationas the best jumper. A related point was made some years ago, by Bertrand Russell(1948), who stated that "perception only gives rise to a negative judgment when the correlative positive judgment has already been made or considered" (p. 138). In the truth value judgment task, children are askedto say whether sentencesare true or false. Following Russell's observation, it is appropriate to ask children for a (possibly) negative judgment of a sentenceonly if the corresponding positive judgment has been under consideration. This is essentially the condition of plausible dissent (which we may call Russell's maxim). It is a pragmatic "felicity condition" on the use of the truth value judgment task. Consider how the condition of plausible dissent is satisfiedin the story we are developing. The condition requires meaningz to be under consideration at some point in the story. This happenswhen Robocop tells the Troll that he might be the winner. Subsequently, some event transpires that makes the sentencefalse on meaningz; Robocop decides that Grover is the winner. This eventuality also makes it clear to the child subjectwhy the test sentenceis false in the context. 27 . 1 . 10 To

see

Background how

ationalize parts by

the

the

, which

we

the

more

as

the

and

the

background specific

elliptical

, could

on . The

More

often

test

sentence

said

that

than . In

(3 )

not

the

assertion

Background Assertion

as its

be

sible for

in

to

of

this

the

have

as the

been

are

other

is

, could that

be

possibilities

ex -

replaced

and

we

.

assertions

. The

, but

by , such

such - and - such ;

assertion

Troll

-

two

established

is mentioned

background

is the

oper

elliptical

expression is the

, the

assertion

to into

a more

backgrounds

discussing the

Troll

replaced

possible

, and

by

elliptical

been

helpful

background

sentence

property

more

has

several

to

for

a

is He

(Robocop

actual

outcome

will

not

)

discuss

out

story

that

point story

that

the

, another the

considers

best

jumper

.

Troll

it quite

the

clear

, the

in

Troll event

the

test the

story , it

is

. This

is to

be

best

to

the satisfy

a possible

jumper

transpire

and

is

in

to

. To prevent

a different Grover

the

puppet

, B , with

sentence

made

applies Troll

why

sentence

should

the

must

child

test

assertion

the

, then

the

as a function

place

the

background both

to

background

to

present

think

so - and - so is the

: B (A ) . To

at one

the

thinks

perspective

background

. In

, Robocop

think

the

, however

the

are

: the

assertion

outcome

Robocop

in

by that

. The

the

be

partitioned

: Grover

the

falsification

story

replaced

( 3 ) makes

. From to

of

dissent

assertion

a specific

. There

outcome

useful

argument

a plausible

be

so - and - so ; or

we

outcome

outcome

background application

will

can

, such

: Robocop : the Troll

Possible

may

, it

r - expression

best jumper

, Grover

the

satisfied ~entence

in

there

story

is test

Outcome

.!

Actual

The actual is false .

and

information

so - and - so is the

is another them here

It

be

specific

, the

expressions

, a specific

expression

, Actual

dissent

. First

the

example

Outcome

plausible

of

best jumper

so

of

procedures

call

one

. For

, Possible

condition

requisite

replacing

pression

, Assertion

, then

an

derived

assertion

by

under test

' s statement

, A ,

applying

the

consideration sentence

the

satisfy this assertion decides

should

condition

outcome

, the

of

at some the

plau

. It

: Grover

-

point

condition

outcome

that

be

of turns

. In

Grover

the is

Fundamentals of Design

227

the best jumper. This last event, the actual outcome, indicates why the test sentence is false; Grover is judged to be the best jumper, not the Troll. It is anticipated that children will describe the actual outcome when they are asked, What really happened? There is still one loose end to tie up. Do children ever assign Grover, instead of Robocop, as the referent of the pronoun in the test sentence? We have already indicated a reason for thinking that children will not make Grover the referent of the pronoun: Grover does not say anything in the story. Robocop, on the other hand, offers relevant information about who is the best jumper. In any event, if children were to assign Grover as the referent of the pronoun, they would presumably respond Yes, in virtue of having conjured up a mental picture in which Grover thinks that the Troll is the best jumper (in order to assign an interpretation that makes the puppets statement true). This strategy would therefore result in responses that would count against the experimental hypothesis, not for it. Hence, this assignment of reference could lead only to type II errors, not type I errors. A simple way to avoid the issue entirely is to make Robocop decide that a female character is best jumper. Then this character could not be taken as the referent of the pronoun in the test

sentence.

27.2 A Summary in Pictures

This section summarizes all of the design features introduced in this chapter, in the order in which they should occur in the experiment. The summary is in the form of pictures of the story as it unfolds in real time. 27.2.1 TheJumping Competition The characters and the setup are introduced.

228

Chapter 27

Storyline

ExperimenterThis is a story about a jumping competition.The judge is Robocop. Last year he won the jumping competition,so this year he gets to be the judge. This year, these guys, Cookie Monster, the Troll, and Grover are in the jumping competition. They have to try and jump over this log, the barrels, and the benches

over here.

Comment

The first event in the story establishesthe background.Becauseof the particular backgroundwe have chosen (He (Robocop) thinksso-and-so is the bestjumper),we have decided to make the story about a jumping contest. To make the story more interesting,Robocopis describedas the judge of this years Best Jumper contest,because he was the winner of last years contest.

27.2.2 The Prizefor the BestJumper

The judge, Robocop,introducesthe prize:coloredpasta.

Storyline

Robocop The winnerof the competitiongets a great prizecolored pasta!See, its in this barrel right here. 2 7.2.3 The Competitors The competitors prepare for the contest.

Fundamentals of Design

229

Storyline

Robocop Line up, everyone. Get ready to try and jump over all these things. 2 7.2.4 The First Competitor:CookieMonster Cookie Monster unsuccessfully attempts the course.

Storyline

Robocop You go first, Cookie Monster. CookieMonster OK. Here I go. I made it over the log. Now Ill try and jump over the barrels. Oh no! I crashed into them. Oh well, Ill try and jump the benches. Phew, they werent so hard.

230

Chapter 27

2 7.2.5 The SecondCompetitor:The Troll The Troll clears the course successfully.

Storyline Robocop Your turn next, Troll.

Troll OK. Im a good jumper.This shouldbe easy for me. Over the log I go. Yeah!Now Ill try the barrels. Good. I jumped over them easily.Now for the benches. Good, I didnt knock anything over! 27.2.6 The Final Competitor:Grover Grover clears the obstacles cleanly in record time.

Fundamentals of Design

231

Storyline

Robocop OK, Grover. Your turn. Grover Im a good jumper, too. Watch me! See how easily I could jump over the log? Now Ill jump over the barrels and the benches. Great, I didnt smash into anything, and I was really fast. 27.2.7 Judging the Competition: The PossibleOutcome Robocop considers the performance of the first two competitors.

Storyline

Robocop All right. Line up, guys. Im ready to judge the competition. Lets see who wins this great colored pasta.

Robocop Cookie Monster, Im afraid you arent the winner. You crashed into the barrels. I think youve been eating too many cookies. You better eat fewer cookies and lose some weight. Then you will be a better jumper.

232

Chapter 27

RobocopTroll,youjumpedverywell.Youdidnt crashintoanythingat all.

You couldbe the winner.Butlet me judge GroverbeforeI decide. Comment

Robocopsstatements aboutthe Trollestablish meaning 2 as a possibleoutcome: He thinksthe Trollis thebestjumper.Thissatisfiesthe conditionof plausibledissent. Robocopindicatesthat the Trolldoesindeedjumpverywellandmightturnout to be the winner,but that he willhaveto thinkaboutit afterconsidering the other contestant, Grover. At this point in the story, it could turn out that the Troll is the

winnerof thecontest,in whichcasethetestsentencewouldbe trueon meaning . 2 2 7.2.8 TheAwards Ceremony:TheActual Outcome Grover takes the prize.

Fundamentals of Design

233

Storyline

Robocop Grover, your jumps were very good, too. You didnt knock anything down, and you were also very fast. So, I think you were the best jumper. You win the prize, this colored pasta. Well done, Grover. Great job! Comment

To meet the condition of falsification,Robocop picks someone other than the Troll as the winner of the contest. This is the actual outcome: Robocop decides that Grover is the winner. As a record of this event, Robocop remains standing by Grover at the completionof the story. The colored pasta also serves as a record of the event. 2 7.2.9 The Troll Rejects the Decision

The Troll is unhappy with Robocops choice of Grover as winner. He takes some colored pasta for himself.

Storyline

Troll No, Robocop, youre wrong! I am the best jumper. I think I should get the prize. I m going to take some colored pasta for myself. (Troll helps himself to some of the prize) Comment

The story concludes by making the test sentence true on the meaning precluded by PrincipleC, 1.meaning This is accomplishedby having the Troll protest to Robocop, saying, No, Robocop, youre wrong! I am the best jumper. Once he has said this, the test sentence is clearly true on meaning . 1 This event comes last to boost the availability of this reading if it is consistent with childrens grammars (i.e., to avoid type I errors).

234

Chapter 27

27.2.10 Kermit the Frog Describesthe Story

The characters are placed alongside props to remind the child of the events of the story.

Storyline

Kermit Let me try to say what happened.That was a story about Robocop, who was the judge, and Cookie Monster, and Grover, and there was the Troll. I

knowone thing that happened.He saidthat the Trollis the best jumper. Child

No!

Comment

The potentiallinguisticantecedentsfor the pronoun are introduced.The order of introductionis designedto avoidtype I errors;the potentialreferentof the pronounon meaning 2 (the adult interpretation)is introducedfirst,and the potentialreferentof the pronoun on meaning 1 (the illicit interpretation) is introduced last. If children use re-

cencyof mentionto guide their selectionof the referentof the pronoun,this order favors meaning 1.

Controlsentencesshouldbe includedin the experimentto ensurethat childrenaccept 1meaning when the grammarpermits it. One type of control would be sentences in whichthe pronoun and the nameappearin reverseorder (e.g.,TheTrollsaid that he is the bestjumper).

Fundamentals of Design

235

27.2.11 The Childs Interpretation

Kermit the Frog checks the childs understanding of the story.

Storyline

Kermit No? Then what really happened? 27.2.12 The Childs Explanation of the Events The child explains to Kermit the Frog that something is wrong with his version of the story.

236

Chapter 27

Storyline

Child Grover was the best jumper, not the Troll! Kermit Then why does the Troll have some pasta? Child Becausehe thought he was the best jumper, but the judge didnt think so.

Experimenter OK. Lets help Kermit. That was a story about Robocop, who was the judge, Cookie Monster, Grover, and the Troll. Who did he say was the best jumper? Child

Grover.

Comment

The experimenterelicitsthe childs version of what really happenedin the story. The potentiallinguisticantecedentsare reintroducedbefore the crossover-questionvariant of the test sentence is presented. 27.2.13 Kermit the Frog Sees the Light

Kermitunderstands the childs explanation.He pays penance for saying the wrong thing.

Storyline

Kermit Oh, I see. So, I guess I dont get to eat the melon. So I have to do push-ups. Comment

The melon is the reward, administered by the child whenever Kermit correctly de-

scribeswhat happenedin the story. If Kermitis incorrect,he volunteersto do pushups to wake himself up, so he can pay closer attention to the next story; this is the reminder.

Fundamentalsof Design

237

27.3 WrittenOutlineof theStory Before constructing any real stories, it is helpful to make an outline, or even to draw pictures, of a prototypical story. Here is one outline of the featuresof the story under discussion: T est sentence:He thinks'the Troll is the best jumper. . Background Context, part 1: There is a contest. Robocop judges who is the best jumper. (Robocop thinks someone is the best jumper.) . Condition of plausibledissent: Meaningz is under consideration Context, part 2: Robocop could end up thinking the Troll is the best jumper. (possible outcome) , . Condition of falsification: meaning2, false or negation of meaning21true Context, part 3: Robocop doesn't think the Troll is the best jumper. Robocop thinks Grover is the best jumper. (actual outcome) . Final event: Meaning!, true Context, part 4: The Troll protests. He saysthat he is the best jumper. 27.4 Conclusion Three essentialfeatures of the truth value judgment task are worth repeating. The first concerns design. In this task, the target sentence, 5, is true on one reading, meaningl, and false on another reading, meaning2. If children are to give negative judgments in responseto sentenceS on meaning2, then the corresponding sentence, NOT S, must be true in the context on this reading. In a nutshell, this is the condition of falsification. Second, it is important to make the denial of a test sentencefelicitous. To accomplish this, it is essentialthat the sentence,on meaning2, be under considerationduring the trial. It is felicitous to ask whether a sentence, Sf is false on a reading only if the discoursecontext is such that S has been under considerationon that reading. This is the condition of plausibledissent. Third, it should be clear to the child exactly why NOT S is true, rather than S. Therefore, the experimenter must make the denial of S on meaning2 as plausible as possible. This is accomplishedby choosing a good background/assertion pair, namely, a pair that enablesthe child to explain "what really happened'! with least effort. Placing the charactersand props at the end of the story in such a way that they record the events is helpful in making it clear what really happened. In some cases , the exact proceduresrequired to maximize plausibledissentdepend on the particular linguistic construction(s) under investigation in the experiment. We will introduce some guiding principles for selectingbackground/assertionpairs in chapter 33. As a final note, we wish to emphasizethat this chapter has describedthe design features of test sentencesonly; control sentencesare another matter, which we take up in later chapters, especiallychapter34.

Chapter28 What's Wrong with This Picture?

This chapter presentsan alternative to the experimentaldesign presentedin chapters 26 and 27. There are severalpoints to this exercise. First, introducing an alternative to the truth value judgment task gives us a chanceto review some of the methodological prescriptions for the design of experimentsusing the task. Second, it enables us to reinforce the importance of satisfying the condition of plausible dissent; it also make it clear that this condition is not easily satisfied in other experimental tasks. Third, discussing a hypothetical experiment using another technique allows us to raise, yet again, the important issueof when the null hypothesis should be rejected, in favor of the experimentalhypothesis. When a reading is absentfrom a child's grammar, the Modularity Matching Model nredicts that the child will not accept a sentence with that reading , even if the sentence ~

-

is true if that reading is imposed on it . More specifically, the Modularity Matching Model predicts that when an unambiguoussentenceis presentedto children, they will accept the sentence (on the correct interpretation ) if the correct interpretation makes the sentence true in the context , and they will reject the sentence if the correct inter :: pretation makes the sentence false in the context . Of course, there is always some amount of noise in an experiment with children , arising from factors like inattention .

In our experience, however, children succumb to these factors only slightly more than adults do. Therefore, as a rule of thumb, we expect to find performancecoming close to 100 % accuracy in most cases-

in any event , above 90 % accuracy . The fact

that such levels have been achieved in many experiments (some of which are re-

ported in this book) provides strong support for the Modularity Matching Model and leadsus to apply it in other cases. In part I, we discussed several circumstances in which children may give " incor -

rect" responses , despite grammaticalcompetence. In general, this should not happen, according to the Modularity Matching Model, if the experiment that reveals such "errors" is conducted properly. The possibility exists, however, that children will sometimesrespond incorrectly even to unambiguoussentences , and even when they have the grammatical competenceto analyze the sentencecorrectly. For example, it can happen that children accept a sentenceas true on a reading that is not made available by their grammars ("false positive" responses ). The possibility also exists that children will incorrectly reject a sentencebecausethey are misled into thinking that it should be interpreted in a way that is not consistentwith their grammars. This will be discussedin chapter 35, where the topic is children's interpretation of sentences containing the universal quantifier every. For simplicity's sake, we will not consider such casesin this chapter, but it should be understood that many of the points raised here carryover

to children 's " false negative " responses.

240

Chapter 28

Perhaps

the

struct

main

contexts

matter

of

that

the

the

negation

of

that

test

of

,

it

itously

"

The

to

it

on

principle

choice

:

There one

a

of

responses

,

the

some

( perhaps

tence

and

As

A

usual

, we

ature

.

( 1992

For

For

( 1 )

th not

,

it

C

mentioned ruled as

meaningz

meaningz

.

),

under

dissent

untrue

,

.

is

then

one

not

sat

Sentences

-

Felic

of

-

is

are

child

the a

always

crucial

,

forced

's

,

,

conclusions

a

prag

-

to

a

prag

-

violate has

to

to

choose

not

about

adopt the

sen

sentences

children

two

determined

presumably

present

.

of

accepting

to

a

violate

one

is will

make

principle

choose

sometimes

therefore

children

pragmatic

to

child

then

violating

child

response

the

responding

firm

the

will

child ,

the

violate

is

case

,

,

assigned ,

forced

situation

or

such

, semantics

correctly

inappropriately

children

,

In

syntax

are

's

-

under

syntactic

and

points

an

based

on

the

chapter

11

. Here

plausible

studies

resembles

,

example of

that

the

example

dissent

have

been

experiment

by

will

. Although reported

Lust

,

in

Eisele

be

an

this

,

the

and

ex

-

liter

-

Mazuka

.

we

will

use

the

target

sentence

( 1 ),

which

that

the

Ninja

Turtle

can

play

the

trumpet

is

.

is

showing

Bert

that

the

Ninja

Turtle

can

play

the

: &

is

showing

Bert

that

the

Ninja

Turtle

can

play

the

the

Ninja

.

.

that

the he

sentence

Meaning

! be

r - expression in

as

,

sentence

C

to

with condition

} : &

Meaning2

it

is

Bert

Principle

takes

errors

condition

.

showing

the

by

a

is

condition

without

this

a

for

experiment

pronoun

in out

it

in

C

dictates the

.

the

closely

trumpet

as

In

When

draw

- ese

discussed

"' Meaning

(b )

Principle

as

plausible

" Use

sentence

children

all

principle

It

a

used

principles

strategy

satisfy ,

trumpet

erence

.)

that

a

to

that

sentence

.

.

is

out

believe

it

illustrate does

is

(a )

that

I

the

this

knowledge of

representation

conditions

order

Principle

He

a

Model

hypothetical by

:

-

Experiment

was

the

as

principle

sentence

turn

)

in

example

governed

the

violates

hypothetical

) , which

If

con

.

will

is

if

might

rejecting

that

periment

of

sentence

requires

/ semantic

which

Hypothetical

experiment

is

to

assuming

sentence

condition .

,

type

This

test

met

First

avoid .

the

not

pragmatic

semantics

grammatic

to of

sometimes conditions

,

syntactic

nonlinguistic

knowledge

the is a

context

a

Matching

semantic

a

and

accept

reason

of

felicitous

. 1

it

to

kinds

each

is

rejecting

of

in

it ,

priori

Modularity

. This

of

shown .

to

context

assertion

failure

have

required

order

the

dissent

for

form

order

these

reading

syntax

. assign

ignore

no

the

in

is

. We ~ re

( in

true

plausible

sentence

the

phonetic

in

adult

violation

a if

to

response

context

be

children

" ) .

. ( Similarly

either is

just

its able

the

that

conditions

of

be

matic

,

the

negative

in

crucial

by

maneuvers

false

of

Relevant

a

should

is

correct

to

jiBe

, based

principle

be

it

responses

experimental

condition

. Even

matic

28

the

interpretation

not

by

,

amounts ( or

for

two

sentence

felicity

pragmatics

may

,

test

the

pragmatic

isfying

and

on

positive

appropriate

Second

that

false

should

the .

Suppose

of

design

sentence

consideration

the

are

research

falsification

of

source

say

,

( 1 ) . Big

Bird

meaningl

,

the

pronoun

takes interpreted

and

deictically

The

. As we to

Turtle

,

pronoun before refer be . The

cannot

must ,

we

to

refer

the

coreferential null

have refer to

meaning

the

experimental

same

the

ref

someone

meaning that

with and

the to

is

-

not that

is

permitted

r - expression

;

hypotheses

,

What's Wrong with This Picture?

243

The essential features of the experiment are summarized in (4).

(4) Test sentence: He is showing Bert that the Ninja Turtle can play the trumpet . a. Meaning11 true

Context, part 1: The Ninja Turtle is showing Bert that the Ninja Turtle can play the trumpet. b.

Meaning21 false

Context, part 2: Big Bird isn't showing Bert that the Ninja Turtle can play the trumpet. We now turn to the condition of plausible dissent, which requires the truth of the test sentence on meaning2 to be under consideration at some point in the trial . Sub-

sequently, some event should occur that makes the sentencefalse on meaning2. In the present experiment, the condition of plausible dissent dictates that sentence(1) should be under consideration in figure 28.2. That is, the proposition expressedby sentence (5 ) must be under consideration

.

(5) He (Big Bird) is showing Bert that the Ninja Turtle can play the trumpet. It is here that the hypothetical experiment breaks down. There is nothing in figure 28 .2 that puts the content of the test sentence under consideration on meaning2. Be-

causethe condition of plausible dissent is not met, the remaining desideratafor the truth value judgment task are also flouted. Most important, becausethe condition of plausible dissent is not satisfied, it may be unclear to the child why the test sentence is a falsedescription of figure 28.2. (Recallthat the actual outcome of the story should make it clear to the child subject why the test sentence is false on meaning2 .)I

The failure (of the experimenter) to satisfy the condition of plausible dissent raises a problem for the child. For the child, the test sentenceis a true description of the context -only on an ungrammaticalinterpretation, meaningl. The meaning on which the test sentenceis grammatical, meaning2, is not supported by the context, because there is nothing that can plausibly be denied. The experimental context therefore violates a pragmatic principle, namely, Russell's maxim that "perception only gives rise to a negative judgment when the correl,ative positive judgment has already been made or considered." According to the Modularity Matching Model, children should avoid violating syntactic and pragmatic principles, if possible. In the present circumstance, however, children have beenplacedin a predicament: they must chooseeither to violate the principles of grammar or to violate Russell's maxim. One option is to disregard the principles of grammar and selecta meaning that, although ungrammatical, does not violate Russell's maxim . On this option , children must assign meaningl ,

in violation of Principle C. Once this option is selected, though, the pragmatic condition of plausibledissentno longer applies, becausethe sentenceis a true description of the context on this interpretation. A secondoption is to assign the interpretation provided by the principles of grammar, meaning2, but to ignore 'the pragmatic infelicity of interpreting a sentenceas a false description of the discoursecontext when the content of the sentence, on that interpretation , was not under consideration in the

discourse. If children occasionallytake the first route, choosing to violate their grammatical principles so as to make the sentencetrue in the context, this should not be held against them , in our view .

244

Chapter 28 8

.,..,...

I

"

7

\

I

6

J 5

I

4

I

2 1

28.1

Figure28.2 \

I

3

\ .i

0

% of f'Yes" Responses Figure28.3 Predicteddistributionsof responses to picturesin figures28.1 and28.2 on the CompetingFactorsModel

.

.

.

.

.

.

a

.

.

SUOIJBAJ8SqO ~O #

.

.

On the basis of results in the literature, we feel confident in reporting the findings of the hypothetical study. In responseto figure 28.2, children will give false positive responsesroughly 25% to 35% of the time (cf. the 32% error rate in the study by Lust, Eisele, and Mazuka). That is, this is how often children adopt a strategy that leads them to accept ungrammatical sentencesas true descriptions of contexts like figure 28.2, in violation of grammatical-principles. Children will correctly accept(1) in responseto figure 28.1 much more frequently, however. The level of acceptancein that case should approach 100%. As a consequence , if enough children are tested on enough trials, the result will be a significant differencebetween the rates of acceptanceof (1) in responseto figure 28.1 and in responseto figure 28.2. The distribution of responseswill therefore be as shown in figure 28.3. Now , we find ourselvesin a predicament. On the null hypothesis, children should find test sentenceslike (1) ambiguol.1s , permitting both meaning1 and meaning2' On the experimental.hypothesis, children should permit only meaning2' Therefore, to reject the null hypothesis, we must show that the pattern of responsesin figure 28.3 is not also characteristicof the acceptanceof both meaning1 and meaningzfor test sentenceslike (1). That is, we must show that the pattern of responsesin the figure is not characteristicof ambiguous sentences . This is not easy to show, as we have pointed out. The problem is that it is sometimes difficult to distinguish preferencesfrom principles- that is, whether a reading is present in the grammar but inaccessible , or absentaltogether. The confound between principles and preferencesdoes not always get the recognition it deserves. It seemsto be commonly assumedthat where children can say I'Yes/" they will (e.g., Grimshaw and Rosen 1990). If a sentenceis ambiguous, then both interpretations will be acceptedin contexts that are appropriate for either read-

What'sWrongwithThisPicture ? 245 ing

.

Therefore

,

many

grammatically

ical

in

knowledge

We

.

for

either

.

sentence

,

is

even

if

ciently

strong

,

to

. 2

in

How

Now

let

meet

the

of

consider

27

,

Recall

task

the

by

sentence

( 6

the

an

( 1 )

)

the

( 7

)

)

We

it

relative

to

Suppose

the

child

.

for

that

meaning

than

to

accessed

figure

some

"

1

meaning2

meaning2

" Yes

a

for

interpretations

response

the

) -

( Big

designed

of

answer

.

28

. 1

is

not

,

is

The

where

suffi

the

.

properly

the

test

into

by

Several

.

-

time

This

in

pattern

The

methodological

sentence

established

.

( 9

be

Following

the

is

( 6

He

He

the

( Big

Ninja

- such

a

background

replacing

possible

key

is

to

prescriptions

a

and

specific

an

as

expression

backgrounds

and

-

in

assertions

) .

Bird

)

is

showing

so

- and

- so

that

Bird

)

is

showing

Bert

that

so

)

is

showing

Bert

that

the

)

is

doing

the

Ninja

Turtle

can

- and

- so

can

play

the

He

Turtle

( Big

Bird

Ninja

Turtle

can

do

.

:

play

:

showing

:

with

as

an

chosen

vey

the

will

present

a

the

He

trumpet

( Big

Bird

Bert

that

such

the

task under

features

- and

Ninja

of

to consideration

test

to

pair

of

the

complications

sketch

of

reader

features

in

/ assertion

a

the

the

/ assertion

design

further

judgment

of

for

background

background

is

list

exercise

the

main

Here

tence

is

two

for

yields

exhibit

appropriate

.

Background

conclude

ticular

-

- such

Turtle

.

~ an

play

the

trumpet

.

Summary

leave

have

.

may

are

accessible

be

could

.

partition

in

: - and

Assertion

. 3

gram

Bert

:

Assertion

28

-

their

Dissent

expression

Background

)

other

less

also

1

experiment

to

yen

:

such

( 9

its

.

Assertion

( 8

is

Background

trumpet

of

children

has

preference

will

Plausible

background

:

on

ungrammat

that

"

child

in

the

,

ambiguity

is

meaning

dissent

trumpet

Assertion

1

of

the

:

play

the

heightened

there

plausible

Background

"

if

it

If

its

evidence

contexts

the

that

.

sentence

on

.

Condition

gi

of

than

but

meaning

elliptical

are

answer

then

how

that

further

"

because

.3

of

sentence

for

,

the

condition

.

the

.2

28

Satisfy

us

chapter

sertion

28

be

given

in

sense

,

a

than

.

others

accessible

" Yes

,

figure

to

less

have

notion

grammars

accept

often

sentences

the

relevant

will

the

they

ambiguous

relevant

the

is

' s

however

figure

depicted

28

them

children

investigation

to

the

in

with

response

is

,

meaning2

associated

then

over

children

of

is

response

course

of

accessibility

it

in

if

more

under

interpretation

Of

with

,

principle

ambiguous

one

consistent

Even

one

reading

A

contexts

the

that

significantly

these

of

disagree

preference

maintain

interpretation

interpretation

matical

it

researchers

correct

pair

the

in

value

design

,

experimental

features

( 7 ) .

truth

features

' s .

corrected

these

For

into

now

,

judgment

including

the

story

goal

task

the

a

.

is

In

reasons

.

format

.

simply

later

for

design

to

We

We

con

chapters

,

selecting

a

-

we

par

-

.

design

children

the

put

knowledge

of

an

of

experiment

Principle

using

C

,

as

it

the

applies

truth

to

value

the

sen

-

246

Chapter 28 Test sentence: He is showing Bert that the Ninja Turtle can play the trumpet. . Background Context, part 1: Big Bird is showing Bert that someone can play the trumpet. . Condition of plausible dissent: Meaning2 is under consideration Context, part 2: Big Bird could end up showing Bert that the Ninja Turtle can play the trumpet. (possible outcome) . Condition of falsification: Meaningz, false, or negation of meaning2, true Context, part 3: Big Bird ends up showing Bert that someonebesides the Ninja Turtle can play the trumpet. (actual outcome) . Final event: MeaningI, true, but violates Principle C Context, part 4: The Ninja Turtle shows Bert that he (the Ninja Turtle) can play the trumpet.

Notice that thesefeaturesof the experimentaldesign cannot be satisfiedwithin the confinesof a static picture. At the very least, a seriesof pictures would be necessary , showing the transition through a number of events. In our view, telling the story in a dynamic fashion with toys and props is the easiestway to satisfy the experimental conditions.

Chapter 29 Strong Crossover

Principle C governs the interpretation of certain complex wh-questions called crossover questions . The kind of crossover questions that have been investigated with children include indirect questions like (1), for reasons that will become clear later . (1)

I know who he thinks has the best smile.

In stating (I ), the speaker presupposes that the referent of the pronoun he is not among the individual (s) who he thinks has the best smile. This is reminiscent of the disjoint reference between the pronoun he and the r-expression , the Troll , in the corresponding declarative . (2)

I know he thinks the Troll has the best smile.

Assuming the trace theory of movement , we can use the same constraint , Principle C, to explain the anaphoric prohibitions in both constructions - that is, why the pro noun in (1) cannot be anaphorically related to the wh-phrase, and why the pronoun in (2) cannot be coreferential .with the r~expression . According to the trace theory of movement , the wh-phrase in (1) originates in the same position occupied by the r-expression in (2). (3)

I know hej thinks whoi has the best smile (cE. I know hej thinks the Trolli has the best smile.)

In deriving the indirect question (1) from its underlying representation (3), the whphrase is moved , IIcrossing over " the pronoun , to its surface position (hence the name " crossover " question ). As the wh-phrase moves , however , it leaves a " trace" behind at the site of origin . Given the obvious similarities between the indirect crossover question in (1) and its declarative counterpart in (2), and in light of the fact that anaphoric relations are restricted in both constructions , an attempt to unify them is war ranted . Chomsky (1981) advanced the proposal that Principle C governs anaphoric relations in both constructions . To make this proposal work , only one further assumption was needed, namely , that the trace of wh-movement .is an r-expression . If the trace in (4) is an r-expression, then it has the same status as the r-expression , the Troll , in (2); the trace would be subject to Principle C, which would require it to be disjoint in reference from any pronoun that c-commands it . The c-command relations can be seen in the simplified tree diagram in (4). P identifies the position of the pronoun , and R the position of the r-expression , in this case the wh-trace.

248

Chapter 29 (4)

/ / / "'-' " " " whoi

IP

/ / / / ""'""""

VP

p

he*i/j

/ / / / ""' """, thinks

CP

/ / / / "'-'" " "

IP

/ / / / ""' " " " R VP ti

........../

/ / " " " "' -" " "

has the best smile

When the wh-trace is c-commandedby the pronoun, as it is in crossover sentences , the pronoun cannot be coindexedwith who, as shown in (Sa), becausethis configuration violates Principle C. The sentenceis perfectly grammaticalif the pronoun has a different index, as shown in (5b). (5) a. *1know whoi hei thinks ti has the best smile. b. I know whoi hej thinks ti has the best smile. The prohibition on anaphoric relations in sentenceslike (1), where the pronoun ccommandsthe trace, is stronger than it is in a related crossoverconfiguration like (6) (where the pronoun does not c-commandthe wh-trace). Therefore, questionslike (Sa) are called strong crossoverquestions , whereas ones like (6) are called weak crossover questions . (6) ?I know whoi [hisi trainer] thinks ti has the best smile. The focus of this chapter is children's knowledge of the constraint on strong crossover questions.

29.1 Children 's Knowledge of theConstraint In asking whether children know the constraint on strong crossover, we are dealing with knowledge of the form (sentence, "'meaning). A certain well-formed sentence cannot receive one of the interpretations that would be permitted in the absenceof the constraint. As with the noncoreferencefacts concerning declarative sentences , it has been maintained that Principle C also constrains anaphoric relations in indirect questionslike (2). More specifically, it is proposed that the trace of a wh-phraseis an

Strong Crossover

249

r-expression . Granting this , we can invoke Principle C to handle crossover sentences, just as it handles declaratives . In crossover sentences, the pronoun c-commands the trace of the wh-phrase. If the trace of a wh-phrase is an r-expression , then the pro noun and the trace must be disjoint in reference, just as the pronoun and the phoneti cally realized r-expression in declaratives like (2) must be disjoint in reference. By transitivity , the wh -phrase and the pronoun are also disjoint "in reference in crossover sentences. Because Principle C requires the pronoun and the wh-phrase (via its association with the wh-trace) to refer to distinct individuals , this prohibition on anaphoric relations is known as the strong crossovereffect.

29.2 BoundVariableQuestions The constraint responsiblefor strong crossovereffects, Principle C, does not pertain to sentenceswhere the wh-phrasedoes not cross over the pronoun. Coreferencebetween the pronoun and the wh-phraseis permitted, for example, in questionslike (7). It should be noted that the pronoun in (7) can also be interpreted deictically. Nevertheless, we will refer to structureslike (7) as boundvariablestructures, becausethey allow the other kind of interpretation as well. (7) I know who thinks he has the best smile. (Comparethe unambiguousquestion in (1), I know who he thinks hasthe bestsmile.) As the simplified tree diagram in (8) shows, the pronoun does not c-command the trace in the structure underlying (7). In the bound variable structure, the wh-phrase does not cross over the pronoun. The wh-phrase moves from the subject position of the embeddedclauseto the position for wh-phrasesin the CP projection, leaving behind a wh-trace in subjectposition, the position labeled R in (8).1 (8)

/ / '~ whoi

IP

,/ R ti

/ / ~' """,

VP

~/ / thinks

/ / '"""'" " " "~ CP

/ / // """"

IP

/ / / """" " " "

rp VP hei/j . / / / "---" ' ",,hasthebestsmile

250

Chapter 29

Because the pronoun

does not c-command

the trace , it can be coreferential

with it . In

sentenceslike this, the wh-phrase, who, and the pronoun he may pick out the same person, meaning I 'know who said that he, himself , has the best smile'. Another dif ference between the two constructions will prove important in the experiments we

discuss. The crucial observation is that despite its singular form, he in questionslike (7) can refer to more than one person. The speakercan say, without contradiction, "I know who thinks he has the best smile. The Troll , Grover , and Yogi Bear." That is,

the pronoun can be interpreted as a variable, bound by the wh-phrase. On this multiple referentinterpretation , (7) can be paraphrasedas in (9). This is why such questions are frequently termed boundvariablequestions . (9)

I know which x is such that x thinks x has the best smile .

The multiple referent interpretation is not permitted in crossoversentences . In (I ), for example, the pronoun cannot be interpreted as referring to severalpeople; this question cannot be interpreted to meanthat severalpeople think they have the best smile. The wh-phrase, who, can pick out severalpeople, but not the pronoun. The pronoun, being singular, can refer only to one person. This differencebetween the possible interpretations of questions like (1) and (7) has been exploited in several experiments with children. The null hypothesis in these studies is that children will fail to exhibit strong crossover effects. If so, they should interpret questions (1) and (7) alike. On the

experimentalhypothesis, by contrast, children should display strong crossovereffects in responseto questions like (1), rejecting the multiple referent interpretation. However, they should permit this interpretation of questions like (7), to some degree. Children's level of acceptanceof questionslike (7), on the multiple referent interpretation , serves as a "baseline" against which to compare their responses to questions like

(1) on the sameinterpretation. If the Modularity Matching Model is correct, children should prohibit the multiple referent interpretation of questions like (I ), as should adults .

On the Competing Factors Model, the results of a (parametric) statistical test would suffice to establish whether or not children know the constraint on strong crossover . This conclusion

would

rest on the demonstration

that children

reject the

multiple referent interpretation of strong crossoverquestionsat a significantly higher rate as compared to the baseline (i.e., the rate of rejection for the bound variable questions). On the Modularity Matching Model, a significant difference in rejection rates would not suffice. This model has an additional requirement : that all children

correctly reject the multiple referent interpretation of strong crossoverquestions90% to 100 % of the time .

29 .3 Previous

Research

Two previous studieshave investigated children's knowledge of the constraint on interpreting crossover questions . These studies focused on the contrast between cross-

over questions, where a singular pronoun cannot refer to more than one individual, as in (10), and bound variable questions, where the pronoun can have multiple referents, as in (11 ). (10 )

Who does he think has a hat ?

(11 )

Who thinks he has a hat ?

Strong Crossover

251

On the basisof findings from a comprehensionstudy of the constraint responsible for strong crossover effects, Roeper et al. (1984) concluded that children interpret crossover questions and bound variable questions in the same way. According to these authors (also see de Villiers, Roeper, and Vainikka 1990), children interpret the pronoun as bound by the wh-phrase. This conclusion is based on the claim that children acceptmultiple referent answersto crossover questions like (10). However, as McDaniel and McKee (1993) observe (also see Thornton 1990; Thornton and Crain 1994; Crain 1991), careful scrutiny of the experimental results shows that few of the children tested by Roeper et al. gave multiple referent responsesto crossover questions(only about 12 of the more than 100 child subjects). A subsequentstudy was conducted by McDaniel and McKee (1993). They investigated children's comprehensionof crossover and bound variable questionsusing a different experimentaltechnique. Instead of having children evaluatethe truth of the puppet's statement, they had them judge the "appropriateness " of a particular answer to both kinds of questions. On a typical trial, one experimenteracted out a scenario with toy figures and props while two puppets (played by the secondexperimenter) and the child looked on. After each story was acted out, the first puppet askedthe seconda question about what had happened. That puppet' s reply, although true, was either appropriate to the question or not. (12)

a.

Boundvariable Who said he was under the blanket? Answer: He did and he did.

b.

Crossover Who did he say was under the blanket? Answer : *He did and he did .

The critical question / answer pairs were the crossover trials , as in (12b); the bound variable trials served as controls . As McDaniel and McKee observe, the multiple referent answer (" He did and he did " ) is appropriate for bound variable questions, but not for crossover questions . This is because bound variable questions can contain a bound pronoun , whereas the pronoun in crossover questions cannot be bound ; that is, the crossover questions display strong crossover effects for adults . The goal of McDaniel and McKee 's study was to determine whether children show strong crossover effects. Lacking the relevant constraint , they would be expected to permit a bound pronoun , hence a multiple referent .answer, to both question types . Children had the opportunity to demonstrate knowledge of the constraint by rejecting the multiple referent answer (" He did and he did " ) as inappropriate for crossover questions , but accepting it as an appropriate answer for bound variable questions . To summarize the main finding , the mean level of performance for children in this study was 46% correct rejections of the answer "He did and he did " to crossover questions . A control group of adults performed slightly worse than this (38% correct rejections ). Despite the high rate of overacceptanceof the multiple referent answer by both groups , McDaniel and McKee claim that the findings demonstrate that both children and adults know the constraint on strong crossover . The reason is that both children and adults accepted the same answer significantly more often for the bound variable questions (90% acceptance by children , and 96% by adults).

252 Chapter 29 McDaniel and McKee argue as follows: As long as children and adults perform with a similar degree of successin distinguishing crossoverand bound variable questions, it can be inferred that both groups have similar grammaticalcompetence,whatever their level of correct performance. Adults are assumedto know the constraint on crossover questions ; therefore , if children pattern like adults, it is reasonable to infer that children , too , know

the constraint . Failures

for both

groups

are attributed

to

performance factors and should not be taken to indicate a lack of grammatical competence. The Competing Factors Model clearly lies at the heart of McDaniel and McKee 's

argument. On this model, subjectsare not expectedto perform perfectly. As long as the level of performanceby both children and adults is similar, and both groups distinguish crossoverand bound variable questionsto a significant degree, it is valid to conclude

that both groups know the constraint .

In assessinglinguistic competence, however, performance should be compared with the model that is being assumed. On the Modularity Matching Model, children (and adults) who know a grammaticalprinciple will perform almost perfectly, at leastif the experiment is constructedproperly. According to the Competing FactorsModel, too, a certain pattern of linguistic responsesis anticipated. As noted in chapter6, every subject should be influenced by the same set of competing factors, to a greater or lesser degree. Therefore , the pattern of responses across subjects in an experiment

should approximate a normal distribution, with a single IIgroup" behavior centering around a mean level of performance. Individuals will deviate from the mean only probabilistically (becauseof extraneousfactors such as fatigue or lack of attention), such that a proportion of observationswill taper off gradually from the mean in both directions . The result is a unimodal

distribution

.

Figure 29.1 illustrates the expectedpattern of responsesfor the two constructions, by either group, according to the Competing FactorsModel. It is this pattern of behavior

that serves as the basis for McDaniel

and McKee ' s conclusion

that children

know the constraint on strong crossover . The argument rests on the assumption that

both children and adults display the samepattern of behavior, in which they respond differently to crossoverand bound variable questions, acceptingthe multiple referent answer (i.e., saying IIYes") significantly more often for the bound variable trials. ' On the Competing FactorsModel, the anticipated pattern of responsesfor crossover questions alone is as illustrated in figure 29.2 (figure 6.3 repeated; see the discussion of the model in chapter 6). The mean rate of correct "No " responses for both groups should be above chance, with the proportion of responses due to error taper-

ing -off in both directions away from the mean. As the figure depicts, the Competing Factors Model allows that children could have a different J'J'balance point " from adults, because children

have less resistance to extraneous

factors ; if so, the contribution

of

grammaticalknowledge to their behavior will be less. The findings of McDaniel and McKee's study were not, in fact, as predicted by the Competing Factors Model. In figures 29.3 and 29.4, respectively, we report the responsesto crossover questions by children and adults. As the figures indicate, the pattern of responses by both groups more closely approximates a bimodal distribu tion than a unimodal distribution . There are clearly two patterns of responses within each group in response to crossover questions . Some members of each -group were

successfulin rejecting the multiple referent interpretation of crossoverquestions(and therefore in distinguishing them from bound variable questions). However, other

254

Chapter 29

# of Trials with a "Yes" Response Figure29.3

m

~

~

Distribution of children's responsesto crossoverquestions

2

# of Trials with a "Yes" Response

~

~

Figure29.4 Distributionof adults'responses to crossover questions

StrongCrossover 255 membersof eachgroup did not, in the main, reject the multiple referent interpretation of crossover questions .

Both children and adults responded more accurately, and more consistently, to bound variable questions. However figure 29.3 indicates that 7 children accepted the multiple referent answer to crossover questions at least 75% of the time; and figure 29.4 indicates that 8 of the 12 adults acceptedsuch an answer at least 75% of the time. Clearly, there was no substantialdifferencein performanceby thesechildren and adults in responding to bound variable questions and crossover questions. Al though it may be reasonableto infer that the adults who performed poorly in responding to crossoverquestionsneverthelessknow Principle C, the observation that a group of children also performed poorly does not license the inference that they, too, know Principle C. Obviously, something about the task used in this study prevented even the majority of adult subjectsfrom displaying grammaticalcompetence. Given the poor performance by half of the children and three-quarters of the adults, one might ask how both groups could show a significant difference in responding to crossover questions and bound variable questions. The answer is that McDaniel and McKee applied a statistic (the sign test) that discardsany subject who performs at the samelevel of successon both constructions. Consequently, subjects were excluded if they answered "Yes" incorrectly to crossover questions just as often

as they answered"Yes" correctlyto bound variable questions. This left only subjects who treated the two constructions differently, and the majority of these subjectsdid answer "Yes" at least once more in responding to the bound variable questionsthan to the crossover questions . The .research issue, however , is whether or not children interpret crossover questions and bound variable questions in th ~ same way . The null

hypothesis is that they do, assigning multiple referent answersto both. The experimental hypothesis , as advanced by McDaniel and McKee , is that children (and adults ) ..

~

-

analyze the two constructions differentlyI rejecting the multiple referent answer significantly more often for crossover questionsthan for bound variable questions. It is clearly inappropriate to use a statistic that discards subjects who disconfirm the experimentalhypothesis. Despite the overall similarity in the linguistic behavior of children and adults, the fact that the distribution

of scores for both groups forms a bimodal

distribution

nulli -

fies the validity of McDaniel and McKee 's argument . The validity of the argument rests on the assumption that responses by both children and adults form unimodal

distributions, as predicted by the Competing Factors Model . If this had been the pattern of results, with both children and adults successfullydistinguishing between bound variable and crossoverquestions, then the logic used by McDaniel and McKee might have been warranted on the Competing Factors Model (but see Crain and W exler, forthcoming). Another feature of McDaniel and McKee 's study also demonstrates their adher-

enceto the Competing FactorsModel . Acknowledging that both children and adults made a high number of errors, the authors remark: We think that such subjects may not be retaining the form of the question , i.e.,

they unconsciouslychangethe word order of the question to match the answer. (p . 288 )

The idea is that children and adults unconsciously change the word order of the crossover question , "Who did he say was under the blanket ?" such that it becomes the

256

Chapter 29

bound variable question "Who said he was under the blanket ?" Although this way of explaining data is consistent with the Competing Factors Model , the Modularity Matching Model is incapable of making such a cl.aim. On the Modularity Matching Model , the processing of unambiguous sentences, such as the crossover questions in the present study , is not open to extragrammatical influences such as the apparent lack of fit between a question and its associated answer. This concludes our discussion of previous research on children 's knowledge of the constraint on strong crossover . The findings , which purport to support the Compet ing Factors Model , do not in fact support it . Neither do they seem to support the Modularity Matching Model . For this, we need to adopt a different experimental design, as described in the next chapter . As we will show , changing the experimental design leads to findings that are more in line with the predictions of Universal Grammar , and with the Modularity Matching Model .

Chapter30 Strongest Crossover

In this chapter, we present a new experiment to investigate children's knowledge of the constraint on strong crossover . We describe the details of the entire experiment :

the hypotheses, experimentaldesign, results, and data analysis. As we layout the experiment , we will check to see that we have satisfied all the items on our methodo -

logical checklist, in particular, the need to establisha baselinecontrol and to fulfill the condition of falsification and the condition of plausible dissent. The experiment presentedhere takes the lead from McDaniel and McKee's (1993) insightful observation that the constraint must be tested by homing in on the one placewhere a differencein the interpretationspermitted for crossoverand bound variable Questions is perceivable : the possibility of a multiple referent response. McDaniel .&

...

-

and McKee achievedthis by having one puppet ask a question and a secondpuppet give the multiple referent answer. Children were to judge the appropriatenessof the secondpuppet's answer. As it turned out, however, the situations were complicated, and both children and adults had a great deal of difficulty judging the appropriateness of the second puppet 's responses. An important innovation

that we made in the

experiment describedhere was to simplify the experiment by using just one puppet, Kermit the Frog. By embedding the crossover question under the carrier phrase "I know who. ~.," we were able to have the experimenterplaying Kermit the Frog control both parts of the question / answer pair . That is, Kermit the Frog would say something like , " I know who he said has the best food . Grover and Yogi Bear." In

addition, we 'introduced a truth value judgment task; children were judging whether or not the character referred to by he had, in fact, said that Grover and Yogi Bear had the best food .

Our new experimental design examinesafresh the experimental and null hypotheses that Roeper et al. (1984) and McDaniel and McKee (1993) had tested .l The ex-

perimental hypothesis is that children have knowledge of the constraint on strong crossover(as statedin Principle C of the binding theory). In this case, children will not treat crossover and bound variable question / answer pairs alike. Crossover sentences

will not be treated as ambiguous sentencesbecausethe constraint on strong crossover will exclude the multiple referent interpretation of the pronoun. On the other hand, bound variable questions will be ambiguous. They allow the pronoun to be treated as a bound variable, which gives rise to the multiple referent response; in addition , these questions allow a deictic interpretation of the pronoun . (1)

Crossover I know who he said has the best food . a. *1 know who he said -t has the best food . b. I know who he said -t has the best food .

258

Chapter 30 (2) Boundvariable I know who said he has the best food. a. I know who said -he has the best food. b. I know who said he has the best food.

According to the null hypothesis, children should not distinguish the interpretations permitted for crossover and bound variable question/answer pairs. On this hypothesis, children lack the constraint on strong crossover, so nothing prevents them from assigningthe multiple referent answerto a crossoverquestion. The pronoun can be treated both as a bound variable and as a deictic pronoun. It follows from this hypothesis that both the crossoverand bound variable questionswill be ambiguousfor children. The null and experimentalhypothesesare summarizedas follows: . Ho: Children treat crossoverand bound variable questionsalike. Expectedresults: Children should allow the multiple referent response for both crossoverand bound variable questions. . HI : Children distinguish crossoverquestionsand bound variable questions. Expected results: Children should not allow the multiple referent responsefor crossoverquestions. 30.1 Design

As is usual in experimental studies using the truth value judgment task, the experiment describedhere consists of a series of trials in which a story is acted out with toys and props. The story contexts make availableboth the meaning ruled out by the constraint and the meaning permitted by the constraint. In the experiment, the meanings that children do and do not allow are studied by examining the answers that they allow for crossover and bound variable questions. Following the requirement that we be conservative in testing the experimental hypothesis (in order to avoid type I errors), the question/answer pair that is ruled out by the constraint on strong crossover is designed to evoke a "Yes" responsefrom children, and the question/ answerpair allowed by the adult grammar is designedto evoke a "No" response. At issue is- whether or not children allow the same interpretations for crossover and bound variable questions. Sincethis is the issue, it is important to make sure that the context is kept constant in our experimentaltest of both question types. Our test of the multiple referent response for bound variable questions should incorporate exactly the same "stories." In designing the crossover part of the experiment, we designatedthe multiple referent responseas the "Yes" responsesince this is the responseassociatedwith lack of the constraint on strong crossover. Sincewe must keep the story contexts constant, the multiple referent responsemust also be associated with the "Yes" responsein our test of bound variable questions. Both meaning! and meaning2are permitted by the grammar in this case. The design for investigating the constraint on crossoverquestionsis as follows: JiOMeaningl , true: Multiple referent answersto crossoverquestions Meaning2, false: Deictic answersto crossover questions There is another important reason for having the multiple referent response of bound variable questions set up as the "Yes" response. This allows us to establish a baselinefor children's acceptanceof the bound variable interpretation. Recall that

Strongest Crossover

in

order

to

children

in

effectively

order

to

dency

assign that

-

hypothesis

is

the

the

over

If

the

constraint

The

design

,

response

to

both

do

not

we

the

possibilities

they rated

is

not

the

are

stories

can

question

30

We

argued

to

The

now

turn

' s

should

the

,

or

tested

not

"

%

of

cross

easily

-

access

interpret

the

answer

,

knowledge

of

for

in the

design

.

In

toys

that

test

is

used

,

,

depending

.

this

stories

,

/

them

,

bound

and

vari

what

way

"

question

a

on

In

-

theme

the

to

or

nar "

any

interesting

battery

that

the

,

-

the

being

for

pair

Natu

test

unlikely

story

respects

be

.

to

change

/ answer

pair

the

but

deictic

story

it

the

essential

looks

crossover

the

times

and

if ,

can

question

complete

response

all

story

crossover

deiclic

to

,

pair

crossover

and

same

three

children

same

any

of

story

/ answer

the

for

,

the

same

boring

,

response

questions

following

the

use

bag

a

a

needed

specifics

kind

of

of

-

kind

no

of

story

can

.

on

decision

by

judge

typical

a

food

)

"

out

that

Four

of

.

the

him

design

.

"

,

is

for

the

in

,

decides

Joker

"

a

judge

two

winners

of

dispute

and the

the

story

:

mistaken on

he

each

rejecting

- be

focuses

referent

judge

openly

would

decision

story

,

altogether

,

The

his

developed

required

contestants

the

the

we

are

the

that

Because

,

the

contest

discourse

follows

In

characters

all

the

to

instead

as

.

checks

" winner

salient

protocol

best

judge

chosen

made

experiment

.

The

pointing

been

is

the

competition

.

settling

have

( 3

100

the

referent

referent

characters

by

particular

contestants

and

judge

the

to

some

three

them

" No

Experiment

depicts

and

a

to

will

of

a

for

but

both

do

evaluation

question

principle

the

accompanied

favor

is

' s to

in

pick

pair

be

This

want

that

is

,

%

to

multiple

variable

with

Kermit

we

can

pair

to

different

means

/ answer

. 2

,

/ answer

question

.

to

substituting

be

the

bound

children

permit

Children

who

multiple

for

all

present

Instead

This

.

story

able

be

by

.

pair

that

.

90

"

,

answer

time

questions

our

the

questions

attention

novel

identical

answer

to

close

" Yes

-

experimental

the

pairs

say

Children

reject

from

allows

grammars

pay

to

response

of

want

.

The

referent

/ answer

ten

questions

.

of

variable

data

experiment

types

pairs

bound

some

variable

multiple

will

-that

,

have

bound

the

show

.2

referent

their

will

of

the

multiple

to

children

them

their

crossover

of

questions

cause

"

to

interpretation

questions

question

,

need

children

proportion

crossover

of

eliminate

of

we

particular

that

crossover

" Yes

/ answer

will

a

show

significant

correct

,

assign

to

say

a

for

question

strong

need

to

pairs

is

This

constraint

to

interpretation

interpretation

.

we

children

answer

must

on

,

.- if

variable

we

,

a

have

responses -

hypothesis

deictic

case

their

/ answer

referent

as

Here

referent in

of

they

.

supported

bound

pronoun

rally

multiple

null

multiple

which

the

referent

.

and

the

constraint

question

multiple

time

the

be

variable

to

tendency

absent

will

bound

knowledge

whatever

obey

to

tendencv

the

demonstrate

override

259

that

judge

' s

crossover

the

they

decision

,

question

which

.

contestant

In

has

.

Protocol

Experimenter

food

the

Last

contest

three

Turtles

Grover

for

your

year

That

people

,

To

.

in

and

Yogi

,

teeth

"

So

Cookies

Joker

him

the

Bear

you

.

the

makes

.

was

this

contest

:

The

Joker

have

a

aren

Grover

,

.

a

' s

walks

cookie

' t

chosen

year

good

one

of

the

winner

of

the

best

the

over

Cookies

to

taste

food

as

judge

."

To

of

food

Teenage

each

Mu

,

Bear

best

Here

are

.tant

contestant

yummy

Yogi

the

.

Ninja

in

but

,

they

" A

hot

turn

' re

dog

.

bad

.

I

260

Chapter 30 love

hot

the

dogs

Ninja

good

for

you

the

best

definitely

Yogi

. And

and

it

the

of

."

But

are

' re

-

kind

They

best of

them

like

,

Then

,

you

, Joker

looks

on

at

it . That

and

onions

you

' re

lot

of

a

. Hot

vegetables

, he

cheese

" Joker

give

food

."

has pepper

says

.

the

in . It

green

Grover

great

, I have

. They

stuff

' s good

vegetables

food

, " No

bad

. That

dogs

's

. You

wrong

. I

energy come

." with

."

trial I know NO .

who

The food

Deictic

said

really

Joker

said

has

the

best

happened

that

he

food

. Grover

and

Yogi

Bear

.

?

( pointing

to

the

Ninja

Turtle

)

has

the

.

crossover

Kermit

trial

I know

Child

YES

Bound

variable

Kermit Child

he

What

Child

c.

has

best

Experimenter

b .

even

mustard

Kermit Child

best

a lot

, " Pizza

. Cookies

says

Crossover

have says

food

Bear

ketchup a.

they and

have

have And

, but

Turtle

who

he

said

has

the

best

food

. The

Ninja

Turtle

.

. trial

I know YES .

: Multiple

who

referent

said

he

has

response the

best

food

. Grover

and

Yogi

Bear

.

the

best

food

. Grover

and

Yogi

Bear

.

or , d .

Bound

variable

Kermit Child

trial

I know NO .

who

Experimenter Child

N ow

I

does

chapters

? At

ground

and

food

the

experiment in

(4 )

Background

In

order

false true

to

on . So

best

it

food

ha ving

the

the

to

be is

best

(5 )

true

food

Next

, let is ,

the

food

the

Ninja

Turtle

has

the

Joker

said

has

in

(4 ) :

has

falsification way

Joker

the

and

,

around

didn ;

: I know

said

the

outlined the

best

best

in food

food

previous

. The

. So - and

back

-

- so .

Bear

of

story

conditions

Joker

the

question

, the

negation

' t say

that

Joker

Yogi who

he

of

Grover

explicitly

Bear

/ answer

and

pair

meaningz Yogi

named

the

Bear

must

be

must

be

have

Ninja

Turtle

the as

.3

said

has

the

best

food

. Grover

and

1 : Grover

and

Yogi

Bear

each

say

that

they

have

the

. false

the

2 : The

best

to of

various

that

.

, part

check

truth

the

said

true

Meaning2

us

the

other the

, part

have

)

summarized

Yogi

Grover

Bear

Context

That

who

the

sentence

Context best

Joker

the

are

and

that in

Meaningl

b .

story

the

, not

Test

?

the

who

condition

true

Yogi a.

this

; or , put

. This

is

: I know

satisfy

has

satisfy

: Gt ; over

meaningz

happened

to

story

for

Assertion

has

.

the

assertion

he

really

( pointing

best

response

said

What He

issue

: Deictic

the

food

make test

Joker

doesn

't

say

that

Grover

of

plausible

and

Yogi

Bear

.

sure

that

question

the

condition

/ answer

pair

should

be

dissent under

is

fulfilled

consideration

.

Strongest Crossover

261

at some point in the trial. This occurs. The Joker considersGrover's and Yogi Bear's food and says that he likes cookies and hot dogs. So the Joker might have chosen Grover and Yogi Bear; but in the end, he rejects them. Severalother featuresof the task deserve comment. First, some event takes place that makes the sentencefalse on meaning2' This happens when the Joker rejects Grover's and Yogi Bear's food as not being the best. Second, there is a reasonfor rej ecting cookies and hot dogs as the best food. The Joker goes through his reasoning in the story: cookies are bad for people's teeth, and hot dogs are full of "bad stuff'! (preservatives). Third, after the Joker choosesthe Ninja Turtle as the winner of the contest, the experimenter ensures that he remains beside the Ninja Turtle in the workspace. This is a record of the events that took place. Finally, the last-mentioned event in each scenariois the complaint by the two "losers" of the contest. These are the two charactersnamed by Kermit in his explanation of the story. This event is placed last to put the pragmatic focus on these characters , so that the incorrect interpretation of the sentenceis promoted, if it is consistent with a child's grammar. The question is whether children's knowledge of Principle C will causethem to override this pragmatic bias and reject the multiple referent interpretation. 30.3 Subjectsand Procedures The subjects who participated in the experiment were 12 children attending a preschool .4 These children ranged in age from 3;7 to 4;8 (average 4;2). The children were tested individually , in a quiet room at their school . The experiment was divided -into two sessions. In the first session, children were presented with four target crossover questions (to which they would respond " No " if their grammar included the constraint on strong crossover ) and two deictic crossover controls (to which the correct response was "Yes" ). The bound variable controls were tested on the same children in a separate session, so this was a within -subjects design .s There were two bound variable controls .

30.4 Results The children were steadfast in rejecting the multiple referent answer to crossover questions, children accepting it on only 8% of the trials. In addition, the children had no difficulty accessingthe deictic interpretation of the pronoun in crossover questions. This was acc~pted by all children on all trials, 100% of the time. The multiple referent interpretation of bound variable questions was acceptedless often, 50% of the time when the bound variable interpretation was true in the context. These results are summarizedin table 30.1. 30.5 Analyzing the Data At first glance, it appearsthat the experimentalhypothesis is confirmed. According to the experimental'hypothesis, children should not treat crossoverquestionsand bound variable questionsalike. This was so. Crossover question/answerpairs were accepted only 8% of the time, and bound variable question/answerpairs were accepted50% of the time. We should be wary, however, of how we interpret the 50% acceptancerate for ,the bound variable question/answerpairs.

262

Chapter 30

Table30.1 Groupresultsfor 12children Type of response Multiple referent responsefor crossoverquestions Multiple referent responsefor bound variable questions Deictic responsefor crossoverquestions

% acceptance 8 SO 100

In order to show that children have the constraint on strong crossover, we need to demonstratethat they allow the multiple referent responsefor bound variable questions, but not for crossoverquestions. To do this, it is sufficient to show that children can give the multiple referent responseto bound variable questionson at least some proportion of trials.6 Recall that the bound variable trials were ambiguous. The multiple referent responsewas set up as the "Yes" response, but there was another grammatical interpretation of the bound variable question. If children interpreted the pronoun as deictic, the sentencewas false, and children would have said 'No ." Both interpretations were made available by the story context, and both constitute legitimate responses.Sincethe multiple referent responsewas set up as the "Yes" response, however, its availability should have been heightened. An examinationof the data from individual subjectsshowed that not all 12 children accessedthe multiple referent answer of the bound variable questions. Five of the children acceptedthe multiple referent responseon both bound variable trials, and the other 7 children rejected it on both. For the 5 children who acceptedthe multiple referent answer to the bound variable questions, we have good eviden~e that they know the constraint on strong crossover. Narrowing our attention to these5 children, we find that they acceptedthe multiple referent answer to crossover questions only S% of the time, as comparedwith 100% acceptanceof this answerfor bound variable questions. These 5 children clearly dispute the null hypothesis, that children accept the multiple referent answerfor both question types. However, closer consideration of the data from the remaining seven children is essential, to assesswhat bearing, if any, these data have on the experimental and null hypotheses. These children did not acceptthe multiple referent interpretation of either the bound variable questionsor the crossoverquestions.7 The data from these children cannot either confirm or disconfirm the null hypothesis, that children will accept the multiple referent interpretation of both crossover and bound variable questions. However, there are data that can, in principle, show that even thesechildren are not treating the two sentencetypes in the sameway. This lends indirect support to the view that thesechildren, too, know the constraint on strong crossover. The data come from the elicited production component of the experiment that follows children's 'INo" judgments. Following a "No" response, children are always prompted to explain "what really happened" in the story. In rejecting crossover question/answer pairs like "I know who he said has the best food. Grover and Yogi - Bear,'! the children correctly explained that Kermit the Frog was wrong becausethe Ninja Turtle had the best food (not Grover and Yogi Bear). Children's rejection of the bound variable question/answerpairs implied that thesechildren were accessingthe alternative deictic interpretation of the pronoun, as shown in (3d). That is, these children were inter-

Strongest Crossover

263

preting the pronoun "he" in "I know who said he has the best food" as referring to the winner of the contest, the Ninja Turtle. Sinceit was the Jokerwho said the Ninja Turtle had the best food, the answerKermit the Frog provides (i.e., "Grover and Yogi Bear" ) was wrong. By way of explaining why they rejected the que~tion/answerpair, children pointed out that Grover and Yogi Bear didn't say the Ninja Turtle had the best food, the Joker did. If children were treating crossoverand bound variable questions alike, presumably they would have rejected both sentencetypes for the same reason. This did not happen. Children's responsesclearly indicated that they interpreted the two sentencetypes differently.

Chapter31 PrincipIe B

The interpretation of ordinary pronouns such as him and her is syntactically constrained by the binding theory. One restriction on the interpretation of pronouns is illustrated (1)

in (1), where him cannot refer to Geraldo . Geraldo

a. b.

admires him .

Geraldoi admires himk . (i .e., Geraldo admires Regis.) *Geraldoi admires himi . (i .e., * Geraldo

admires himself .)

It should be noted that the form of sentence(1) is not at issue. It is a grammatical sentenceof English. A certain meaning cannot be assignedto (1), however: it cannot be used with the meaning that one and the same individual is the referent of the pro noun and the referent of the name. Sentence (1) does permit another meaning , how ever : one that takes the name and the pronoun to refer to different individuals . In short , there is a constraint on sentences like (1) to the effect that the 'pronoun must refer to some male individual who is not mentioned in the sentence . Therefore , (1) is

consideredto be ungrammaticalon a particular interpretation . The constraint in question is Principle B of the binding theory. Principle B can be stated roughly as follows: (2) PrincipleB If an NP c-commands a pronoun within the same clause, they cannot be

coindexed; hence, they cannot be anaphoricallylinked.1 A simplified phrase structure tree for (1) is given in (3); the subscriptsshow that the pronoun (3)

cannot refer to Geraldo .

266

Chapter 31

Notice that Principle B does not always prohibit pronouns from referring to individuals who are mentioned in the samesentence. In (4), for example, the pronoun him can be used to refer to Geraldo , as in (a) (cf . (b)).

(4) Geraldo believesthat Oprah admireshim. a. Geraldoi believes[that Oprah admireshimi/j]' b. .v.Oprah believes[that Geraldoi admireshimiJ. Coreferenceis possiblein (4) becausethe pronoun and the name Ceraldoare in different clauses.This is illustrated in the phrasestructuretree in (5).. Of course, the pronoun in (4 ) can also be taken to refer to some other male individual

not mentioned

in the

sentence, as shown by the j subscript on the pronoun. (5)

. maIn cIause Geraldoi embedded clause

Oprah

These examples show that Principle B also imposes a restriction on the interpretation of pronouns that co-occur with nonreferential NPs like no talk-show host. Rather than referring to an individual or a set of individuals, (6) meansthat there is no individual with the joint properties in question: being a talk-show host and admiring some salient male person in the domain of discourse. NPs without inherent reference, such as no talk-showhost, are called operators(or quantificationalNPs). Principle B

Principle B

267

pertains to the anaphoricrelations between pronouns and operators, as well as to the anaphoric relations between pronouns and referential NPs. When a pronoun is anaphorically linked to an operator, it is called a boundpronoun. One interesting property of bound pronouns, noted earlier, is that they are interpreted as variables. As a variable, the bound pronoun can range over several individuals. For example, even the singular pronoun he in (7) can be interpreted as a placeholder for every individual in the domain of discoursewith the relevant property (possibly including females, if one uses he as a gender-neutral pronoun). In . (7), the wh-phrase who is the operator that binds the pronoun (hence the term boundvariablequestion ). This bound variable question is contrasted with a crossover question like (8), in which the singular pronoun he can refer only to a single male individual. (7) I know who thinks he is intelligent. (i.e., Geraldo thinks so; Regis thinks so; . ..) (8) I know who he thinks is intelligent. (i.e., Geraldo thinks Regis is.) To help clarify matters furtherI we briefly sketch the semanticrepresentationthat is assignedto sentenceswith operatorsand bound pronouns. In the following examples, we use the variable x to range over individuals in the domain of discourse. (9) a.

I know who thinks he is intelligent. I know which person: x [x thinks x is intelligent] b. Every talk-show host thinks he is intelligent. Every: x [If x is a talk-show host, then x thinks x is intelligent] c. No talk-show host thinks he makesenough money. No : x [x is a talk-show host and x thinks x makesenough money]

31.1 An AlternativeFormof PrincipleB On the version proposed by Chomsky (1981, 1986), stated in (2), Principle B applies to pronouns that are c-commandedby either a referential NP or an operator. By contrast, Reinhart (1983b, 1986) has proposed a theory in which Principle B appliesonly to pronouns that are bound by an operator. On this account, as far as syntax is concerned, a referentialNP and a pronoun in the samelocal context may corefer. (10)

Geraldo admireshim. a. Geraldoi admireshimk. (i.e., Geraldo admiresRegis.) b. Geraldoi admireshimi. (i.e., Geraldo admireshimself.)

On Chomsky's account, by contrast, a pronoun cannot be coindexedwith a local referential NP, so only the (a) reading is possible. Nevertheless, something in Reinhart's theory has to rule out coreferencebetween a referentialNP and a pronoun when they appearin the samelocal context. After all, the sentence.Geraldoadmireshim has only one interpretation, at least for adult speakers . According to Reinhart, this is handled by a pragmatic rule, which she calls Rule I. Rule I allows speaker/hearerswho are actively engagedin a conversation to figure

268

Chapter 3I

out

what

coreference

syntax

more

ascertain

a

in

of

the

a

The

lows

:

was

Ceraldo

,

the

would

speaker

of

(

This

use

at

our

the

,

same

,

.

we

in

a

Cairns

,

tences

,

is

,

one

alternative

In

the

the

to

grammatical

A

void

Ambiguity

. "

reasons

a

as

reflexive

because

fol

-

pronoun

it

and

this

intended

admires

,

him

conveys

who

in

-

corefer

him

.

interpretation

test

.

-

Since

of

'

two

sentence

the

him

with

with

a

is

the

)

it

been

(

and

i

. e

Maxfield

1992b

;

the

research

and

of

the

chapter

the

assignment

have

A

,

Rule

I

,

capacity

to

this

point

understanding

of

who

and

C

studies

Chien

it

been

know

children

many

;

-

enough

might

' s

even

and

prag

in

long

return

children

Lee

a

processing

will

1992

;

s

Principles

in

1984

remainder

. ,

Wexler

findings

We

on

replicated

Jakubowicz

'

that

is

in

children

children

found

I

pronoun

,

that

.

Rule

memory

pronoun

As

.

ordinary

in

contend

.

accounted

I

representations

an

reflexive

1993

sentences

Rule

structural

studies

has

s

utterance

research

It

of

Reinhart

actual

been

These

are

)

and

and

adhere

appear

to

is

attested

Wexler

Wexler

1990

1987

;

Sigurjonsd6ttir

,

designs

,

Hyams

that

;

McDaniel

,

were

and

used

.

,

,

concerned

*

meaning

)

same

of

the

,

.

to

see

With

the

sentences

whether

Similarly

,

the

and

,

not

constant

shed

observation

different

the

sentences

,

light

the

that

are

;

one

favors

sentences

an

are

can

be

respond

pre

-

associated

that

in

subjects

-

two

contexts

context

differences

-

sen

assess

There

observation

on

knowl

to

.

different

other

two

s

to

attempt

sentences

the

'

meanings

designs

in

,

held

can

certain

test

presented

or

context

these

to

procedure

differently

.

,

assign

sentence

second

children

of

is

the

investigating

Therefore

can

sentence

interpretation

context

.

subjects

the

In

with

prohibit

that

one

knowledge

semantics

apply

Avrutin

;

sentence

meaning

the

. ,

discuss

.

same

same

respond

ii

B

interpretation

in

.

and

)

meanings

.

favors

sented

. g

that

(

of

procedures

context

:

between

deictic

operative

at

McDaniel

Principle

with

a

the

becomes

look

1986

we

range

basic

e

principles

that

to

ordinary

interpretation

listener

Geraldo

the

the

.

speaker

not

interpretation

the

designs

exact

to

(

has

1990

linguistic

I

B

,

consume

research

of

see

;

of

I

to

Rule

Koster

1990

Acquisition

edge

a

maintain

,

finding

Thornton

them

. 2

when

,

,

and

Reinhart

sentence

This

and

Hsu

;

obtain

is

about

of

to

(

,

and

,

selected

,

to

Rule

Principle

on

Koster

1988

the

the

hearers

as

coreference

That

syntax

capacity

languages

,

.

) ,

him

where

intended

)

10

intended

remark

and

by

.

the

apply

position

B

several

the

deictic

rule

,

,

representation

constraints

Chien

To

of

Principle

Deutsch

himself

capacity

processing

are

other

violate

admires

one

Grodzinsky

governed

to

. ,

(

been

)

representation

knowledge

sentences

then

Hearing

had

have

add

another

.

their

Now

.

the

sufficient

grows

.

pronoun

the

time

instead

below

)

ambiguity

on

to

requires

whether

lack

,

cases

instructs

position

pragmatic

10

I

same

. e

pronoun

reflexive

remarks

except

maintain

decide

a

the

Rule

.

that

must

but

lying

not

B

produced

with

did

principle

one

the

)

in

,

in

another

speaker

Geraldo

interlocutors

Essentially

( i

ordinary

the

.

available

(

would

said

Principle

mind

If

( without

completes

by

matic

The

.

interpretation

for

31

10

on

pronoun

have

intended

to

possible

)

reference

sentence

an

their

10

appear

direct

hinges

chosen

explicitly

ence

I

(

are

of

for

reflexive

terpretation

in

options

one

works

has

also

as

can

both

Rule

I

speaker

,

if

is

of

The

himself

in

;

Rule

by

,

pronoun

pronoun

how

intended

option

reflexive

application

is

one

sentence

ordinary

Here

are

than

whether

pronoun

to

relations

allows

their

subjects

under

in

-

the

Principle B

269

sameway to different sentences , with context held constant, can also shed light on their

grammars

.

In the studies reviewed earlier, a sentence was matched against two different con-

texts: one corresponding to the sentencemeaning that is prohibited by the linguistic constraint Principle H, the other corresponding to the sentencemeaning that is permitted by the adult grammar. The two examplesof studiesusing this researchstrategy were by Hirsh-Paseket al. (1995) and Grimshaw and Rosen (1990) (see chapters 7 and 10). Another study of Principle B, by McKee, Nicol, and McDaniel (1993), used reaction time as the dependentvariable (seechapter 7). Probably the best-known result in the acquisition literature comes from a study by Chien and Wexler (1990) (see also Wexler and Chien 1985). In this study , children were shown pictures and were asked questions about the characters and events de-

picted there. In a picture designedto check children's adherenceto Principle B, there were two characters: Papa Bear and a monkey . Papa Bear was shown covering him -

self, and the monkey was shown standing nearby, watching Papa Bear. The experimenter introduced the characters (" This is Papa Bear; this is a monkey " ). Then children were asked the test question in (11). (11)

Is Papa Bear covering him ?

In response to questions like (11), many children between the ages of 3 and 6 answered IIYes," giving this answer 50% of the time or less. The correct answer is IINo "

for adult speakers , becausethe only possiblereferent for the pronoun in the picture is the monkey, and PapaBear is not covering him. In other words, children are apparently able interpret the pronoun him as if it were a reflexive pronoun , linked to a referential

NP

(its antecedent

) . It should

be noted , however

, that

these

children

are also

able to assign the correct meaning to pronouns . So if Papa Bear had covered some-

one else, but not himself, these children would be willing to accept sentence(11) as a correct description of what happened. BecauseEnglish-speaking children assign an IIextra" meaning, beyond that of adult speakers , the error is one of semantic overgeneration .

This example of children's semanticovergenerationillustrates one horn of the language leamability dilemma: how children recover when they have overshot the target language. The example of children's interpretation of pronouns raisesa different but corollary question: how children avoid undershooting the target langu'age. One way to resolve this horn of the dilemma

would

be to claim that children

are conservative ,

thereby avoiding the problem associatedwith syntactic and semanticovergeneration. As we have shown , this is' not how children

behave .

In sentencesin which the pronoun is c-commandedby an operator, Principle B appears to be respected, however.2 That is, the samechildren who accept coreference between the pronoun and the referential NP in (II ), reject (12) as a description of a story in which every bear covers himself , and not the monkey . (12)

Every bear covered him .

Weare not out of the woods yet. There is still something to explain, namelyI why children mistakenly permit coreference between a pronoun and a referential NP in sentences like (13) (but not between a pronoun and an overt operator in sentences like (14 )).

270

Chapter 3 I

(13)

Is PapaBearcovering him?

(14)

Is every bear covering him?

A seriousleamability problem arisesif one kind of explanationis pursued. For example, supposethat some children apply Principle B to bound pronouns only (see Wexler and Chien 1985). If so, thesechildren not only assignthe adult interpretation to sentences like (13), they assign a nonadult interpretation as well. This means that the evidence these children encounter will always be consistent with one of the interpretations they assign. In the absenceof evidencefrom the environment that instructs children about what meaningscannot be assignedto sentences , these children would be hard pressedto jettison the incorrect meaning from their grammars. An alternative explanation of children's responsepattern avoids the problem of learnability. The proposal by Grodzinsky and Reinhart (1993) salvages children's grammatical knowledge by placing the blame for the problem elsewhere, in the domain of linguistic performance. Although children who respond incorrectly to sentenceslike (13) know pragmatic Rule I, according to this account, they lack the processingcapacity to executeit . Recall that to apply Rule I to a sentencecontaining an ordinary pronoun, the listener must maintain the actual utterance in memory long enough to see if an alternative linguistic representation can be constructed for the sentence, one with a reflexive pronoun. According to Grodzinsky and Reinhart, some children cannot maintain the two representationsin memory at the same time, so they cannot decide whether or not Rule I applies; consequently, they simply guess about the coreferencerelations between the pronoun and the local referential antecedent in sentenceslike (13). In short, on this scenariochildren know Rule I but lack the processing capacity to apply it . As the processing capacity of these children increases , however, Rule I becomesavailable for execution. Hence, the leamability problem is circumvented. 31.3 Conclusion

Even if children's failures are due to limitations in processing capacity, it should be clear by now that such differencesbetween children and adults are not anticipatedby the Modularity Matching Model. Therefore, the model would have to be amended if Grodzinsky and Reinhart's proposal turns out to be correct. Amending the model would be preferableto abandoningit altogether, of course. As we have stated, maximizing the similarities between children's cognitive systemsand those of adults has the virtue of being responsiveto the general problem of languagelearnability- the observation that all children successfullyconverge on an adult linguistic system despite the considerablelatitude in their linguistic experience. To the extent that the cognitive mechanismsof children and adults are similar, the learnability problem is nullified. This is why we assumeequivalenceas the Null Hypothesis. By contrast, other models abandonthe Null Hypothesis from the start. On thesemodels, children differ from adults in language processing. Therefore, these models add unwanted degreesof freedom, by tolerating a wide range of processing explanations for differencesbetween children and adults. In addition, they raise a new question to be addressedfor each processing account: how do children modify or expunge their current processingsystem so as to converge on the adult system?

Chapter32 FollowingUp on PrincipleB

The previous chapter reviewed a striking finding from the literature on language acquisition: namely, that some young children appear to violate Principle B in comprehending sentenceswith referential pronouns (e.g., A vrutin and Wexler 1992; Jakubowicz 1984; Wexler and Chien 1985; Chien and Wexler Cairns , and Hsu 1990 ; Thornton

1990; McDaniel ,

1990 ). The result does not appear to be a methodo -

logical artifact. Even when tested in a truth value judgment task that pays attention to the condition of plausible dissent, the condition of falsification, and so forth, some children allow the pronoun him to refer to Papa Bear in response to sentence (1) as a

description of a story in which PapaBear covers himself~ (1) As the

Papa Bear covered him .

noted

pronoun

same

, children

appear

is c - commanded

children

( 1 ) reject

who

by

accept

I know

typical

story

(3 )

who that

Protocol

of

evokes for

test

sleep

one

outside , and

this

after

cold

need

ket . I don cover

."

with

this

whole

own

as follows

:

thing

the

Bear

to

in

which

) . That

is , the

referential covers

blanket

Kermit

I know

what

Kermit

I know

who

Bear that

NP

himself

Grover

in .

)

me

Papa

/ ll have can

warm

Bear

blanket

?" for

. " You

."

to

a very

shiver

cold

. Grover

But

Papa

you

too

to

Bear

. I ' m so

get

have

. Lie

said ,

, then

decided

was to

blanket

. You

keep

my

began

Bird

Bird

. It

enough

warm Big

feathers

under

big

Big

stars

with

keep said

, and

the

Papa

is not

,"

my

down

see

me

blanket

covers

lie

Papa

and

, Grover

Bird

and

which

could

cover

it , because

. I ' ll his

you

" Here

( Big

? Good

himself a.

' t need

you

Grover

."

element

pronoun

Bear , Grover

they

, Grover

, but

the

, Grover

sentences

.

be

, Papa

, so

a while

, Grover

in

B

Bear , could

, I will

might

story

night

said , " Sorry

blanket

result

, in

Bear

B

quantificational the

story

. Papa

of Principle In

said , " Papa

same

Principle

( or

between

him

this

respect

operator

the

covered

Experimenter

night

an

to

coreference

( 2 ) as a description

(2 ) A

, however

another my

down

blan

, and

-

I ' ll

" Are

you

all

( Papa

Bear

covers

set

) happened

in

that

story

. Papa

Bear

covered

him

.

or b .

covered

him

. Papa

Bear

. \

If children them

. In

noun pretation

lack

Principle

particular

refers

to

, they Papa

, according

B entirely should

Bear ; this to

which

be

, the

test

able

to

interpretation the

pronoun

sentences assign

the

would refers

.in

interpretation be

to

( 3 ) should

in

some

addition unmentioned

be

ambiguous

in

which

to

the

the adult

individual

for pro

-

inter

-

. In

272

Chapter 32

the story in (3), it is true that PapaBearc~overed himself with a blanket. On the basis of the story, then, children who lack knowledge of Principle B should acceptKermit's assertionsin (3a) and (3b). That is, it should be possible for them to interpret both "Papa Bear covered him" and "I know who covered him. Papa Bear" to mean that Pap~ Bear covered himself. On the other hand, if children have knowledge of Principle B, they should reject Kermit's description of the story in eachcase. When askedto explain what really happened, they should say something to the effect that PapaBear didn't cover him (Grover), Big Bird did. The finding reported in the literature is that children acceptthe illicit interpretation of "PapaBearcoveredhim," but reject the sameinterpretation of "I know who covered him. PapaBear." Apparently, the differencehinges on the presenceof the operator who in the latter. In sentencesin which an operator binds the pronoun, even children who allow an illicit interpretation of certain sentenceswith referential NPs, in apparent disregard for Principle B, evidently obey the constraint. A serious leamability problem arises for children who apply the constraint on the interpretation of -pronouns to bound -pronouns only. - For these children,. sentences ... with referential pronouns have a meaning that adults cannot assign. In other words, these children are committing semantic overgeneration. In the absenceof evidence from the environment that instructs children about what meanings ..... cannot be assigned to sentences , these children would be hard pressedto jettison the incorrect meaning from their grammars. Several attempts have been made to explain away the leamability problem associated with children's apparent violation of Principle B in sentenceslike PapaBear coveredhim. We will focus on only one: the acquisition scenarioadvancedby Grimshaw and Rosen (1990), responding to experimental data reported by Chien and Wexler (1990). Chien and Wexler tested children between 5 and 6 years old and found that they acceptedsentenceslike (4a) about 50% of the time on the coreference interpretation. This contrastedwith much lower acceptances , about 15% for the same interpretation, for sentenceslike (4b) in which the antecedentof the pronoun was an NP like everybear. Like the NP no talk-showhost, the NP everybeardoes not have inherent referenceand behaveslike an operator. (4)

a. b.

Papa Bear covered him . Every bear covered him .

Grimshaw and Rosen's account of Chien and Wexler's findings begins with sentenceslike (4a), with referential pronouns. They contend that factors such as intonation and pragmatic context may be responsible for the reduction in correct responsesfor some children to sentenceswith referential pronouns. For example, if stressis assignedto the pronoun in (4a), as in (5), then the pronoun and the namecan be coreferential. (5) PapaBear covered HIM . If children assignedstress to pronouns in experimental studies of Principle B, then this could have led them to make "errors" on such sentences . If so, one cannot conclude from their performancethat they lack the grammaticalconstraint on the interpretation of referential pronouns.1 Still to be explained, howeverI is children's high rate of correct responsesto sentenceswith pronouns that are bound by an operator, like (2) or (4b). Grimshaw and

Following Up on Principle B

273

Rosen's explanation has two parts . First , they argue that the linguistic factors (intonational and pragmatic ) that may have contributed to children 's poor performance in response to sentences with referential pronouns do not , by and large, apply to sentences with operators . Notice , for example, that Every bear coveredHIM cannot mean that every bear covered himself , no matter how much stress is given to the pronoun . Second, Grimshow and Rosen suggest that some children may have difficulty accessing the bound interpretation of a pronoun . In their view , this leads children to compute only the interpretation in which the pronoun him to refers to some male individual not mentioned in the sentence. In effect, sentences with operators are unambiguous for children , whereas sentences with referential NPs are ambiguous . This is why children reject sentences with operators more often than sentences with referential NPs; knowledge of Principle B is not at issue. Let us consider what Grimshaw and Rosen's proposal means for an experiment testing children 's knowledge of Principle B. In a truth value judgment task, the experimental design would be as in (6). The meaning ruled out by the constraint would be set up as true and would evoke a "Yes" response from children . This is the bound variable interpretation . The meaning that is allowed by the adult grammar would be set up as false and would evoke a "No " response. (6)

Test sentence: Every bear covered him . a. *Meaningl1 true Context : Every bear covered himself . b. Meaning21 false Context : Every bear covered some other male.

In a truth value judgment task, we assume that children who lack Principle B will say " Yes" because they interpret the sentence to mean that every bear covered himself . A 'IN 0" response, on the other hand, is taken to mean that children do not permit the bound variable interpretation . Consequently , they take the only interpretation made available by the grammar . If Grimshaw and Rosen are correct , however , children 's IINo " responses cannot be interpreted as rejection of the bound variable interpreta tion (and as demonstration that the children know Principle B). Rather, children 's I'No " responses merely demonstrate that the only interpretation they can generate is false. In effect, the fiN 0" responses reveal nothing about children 's knowledge of Principle B. It is at least in part because these children 's grammars lack the ability to form the binding relation between operators and pronouns that the children perform as well as they do . This explanation of children 's performance circumvents the leam~bility problem raised earlier . On this account, children 's grammars do not overgenerate semantic interpretations of sentences. In fact, they undergenerate in some cases. Therefore , they will encounter the positive evidence that they need to extend their grammatical options ; the evidence will consist of sentences with operators and pronouns where only the bound pronoun reading is true in the context . Children who lack the capacity to form operator /pronoun relations will not have the full inventory of semantic interpretations for a host of constructions . For example, consider sentence (7). (7)

Every boy likes his dog . a. Every boy likes his own dog . b. Every boy likes a certain guy 's dog .

274

Chapter 32

For adults, (7) is ambiguous, with the interpretations paraphrasedin (a) and (b). If Grimshaw

and Rosen ' s account is correct , then some children

at least will not have

the bound pronoun interpretation, (7a), available to them at some stage of language development. In chapter 33, we will describe how to design a study that tests this prediction. (As far as we know, the study has not been conducted.) The design of the study proves to be quite complex. It is worth describing in detail, becauseit offers severalvaluable lessonson the appropriateuse of the truth value judgment task.

Chapter33 Setsand Circumstances

The truth value judgment task is used in this chapter to make the alternative meanings of sentences like (1) available .

(1) Every boy likes his dog. a. Meaning! : Every boy likes his own dog. b. Meaningz: Every boy likes a certain guy's dog. On one meaning, the sentence(uttered by a puppet) is an accuratedescription of something that happenedin the story that was acted out on that trial; on the other meaning, the sentenceis an inaccuratedescription. The subjects' task is simply to respond "Yes" or "No" depending on whether they judge the sentenceto be true or false. The null hypothesis is that children have both meaningsavailableto them, just as adults do. The experimentalhypothesis is that children have accessonly to meaning2. The hypothesestake this form becausewe are adopting the viewpoint of Grimshaw and Rosen (1990 ), who maintain

that children

lack access to the bound variable

interpretation of pronouns . In previous studies we have reviewed , children were also

predicted to lack something, namely, a meaning that was ruled out by a linguistic constraint

.

These considerations influence the decision about which meaning to associate with the " Yes " response

(true in the context ) and which

to associate with

the " No " re -

sponse (false in the context ). We continue to adhere to the research strategy of stack-

ing the cards against the experimental hypothesis (to avoid type I errors); if it turns out that the findings favor the experimentalhypothesis, the evidencethey provide will be more compelling. Therefore, in the present experiment, the meaning that children are proposed to lack, meaning1, correspondsto the "Yes" response. If the experimental hypothesis is correct, children will only assign meaning2, interpreting the pronoun as referring to some salient male person not mentioned in the sentence. Therefore, the test sentenceshould be false in the story, evoking a "No" responseon this interpretation. This part of the design is summarizedin (2). (2) Test sentence:Every boy likes his dog. a.

:f.Meaningl , true

Context: Every boy likes his own dog. b.

Meaning2 , false

Context: Every boy doesn't like a certain guy's dog. Following these guidelines, let us construct the storyline. First, to make meaning} true in the story, we must establishthat every boy likes his own dog. A child who allows this reading of the test sentence, contrary to the experimental hypothesis,

276

Chapter 33

should judge the test sentenceto be true becauseof this part of the story. A "Yes" responsewould therefore be taken as evidence that the child is able to understand pronouns as bound pronouns. This brings us to meaning2 and the condition of falsification. 33.1 The Conditionof Falsification The condition of falsification calls upon the experimenter to make the negation of (1) true in the context, on the interpretation that children have available to them according to the experimentalhypothesis (i.e., meaning2)' The negation of (1) on this interpretation is (3). (3) Every boy doesn't like a certain guy's dog. The problem is that (3) is ambiguous. The two interpretations are indicated in (4) and (5). In (4), the universal quantifier everyis interpreted outside the scopeof negation; in (5), negation has wide scope, as shown by the paraphrases . (4) Every boy is suchthat he does NOT like a certain guy's dog. (i.e., All boys dislike the dog.) (5) NOT every boy is suchthat he likes a certain guy's dog. (i.e., At least one boy dislikes the dog.) Which of the two ways of negating the meaning of (3) should be used in the truth value judgment task? Should (4) be used as its negation, or (5)? To answer this question, imagine the thought processesof the child subjectin the experiment. In part, the child might think: Let me see. The puppet said, "Every boy likes his dog." This is either true or false. If it's false, then everyboy doesn 't like his dog. The italicized portion of the child's thought process is ambiguous, however. The child will actually be following just one of theselines of reasoning: (A )

The puppet said, "Every boy likes his dog." This is either true or false. If it's false, then noneof the boyslikesthat guy's dog.

(B)

The puppet said, "Every boy likes his dog." This is either true or false. If it's false, then at leastoneboy dislikesthat guy's dog.

It seemsto us that the secondline of reasoning is more likely, for reasonswe take up next. 33.2 PlausibleDissent: ChoosetheMore GeneralReading At this point, we are consideringhow to satisfy the condition of plausibledissentfor the sentence"Every boy likes his dog." In satisfying the condition of falsification, we are facedwith an ambiguity: there are two possibleways to form the negation of the test sentence.Which of the two interpretations of (3) offers the better way to satisfy the condition of plausible dissent? We suggest that the negation of the sentence should correspondto the (B) reading, according to which at least one boy dislikes the guy's dog.

Sets and Circumstances

A Every boy does not like his dog

277

B Not every boy likes his dog

I Boy1dislikes theguy's dog I Boy 3 dislikes the guy's dog Boy 4 dislikes the guy's dog Boy 1 dislikes the guy 's dog

Boy 3 dislikes the guy's dog Boy 4 dislikes the guy's dog

Boy 1 dislikes the guy's dog Boy 2 dislikes the guy's dog Boy 3 dislikes the guy's dog

Boy 1 dislikes the guy's dog Boy 2 dislikes the guy's dog Boy 3 dislikes the guy's dog

Boy 4 dislikes the guy 's dog

Boy 4 dislikes the guy 's dog

Figure 33 .1 Sets of circumstances corresponding to the alternative readings of sentence (3)

To see why (B) offers the more natural way to satisfy the condition of plausible dissent, let us consider an example. In the domain of discourse, there are four boys. The (A ) reading pertains to all four boys, and all of them dislike the guy's dog: boy! dislikes the dog, boyz dislikes the dog, boY3 dislikes the dog, and bOY4dislikes the dog. Therefore, there is just one circumstancethat makes the sentencetrue on this reading. On the (B) reading, however, severalcircumstancessufficeto make the sentence true. All that is required on this reading is that at least one boy dislike the guy's dog. Figure 33.1 illustrates four of the possiblecircumstancesthat could make the (B) reading true. Notice that one of them is that none of the boys like the guy's dog. That is, the (B) reading remainstrue even in the one circumstancethat makesthe (A ) reading true: where all four boys dislike the guy's dog. As noted in chapter 14, when a particular relation of entailmentsholds between the alternative readings of an ambiguousexpression, the circumstancescorresponding to the alternative readings exhibit a subset/superset relation. The relevant notion of entailment is as follows: A reading Q of a sentenceentails another reading R iff every circumstancein which Q is true is also one in which R is true. To say that (A) entails (B), then, amounts to the claim that every circumstancein which "Every boy does not like the guy's dog" is true is also one in which "Not every boy likes the guy's dog" is true. This relation among circumstancesdoes indeed hold. Consider the circumstancesin which (3) is true on the (A) reading. In every one of these circumstances , all of the boys in the domain of discourse dislike the dog belonging to some salient male. Sentence(3) is also true on the (B) reading in all of these circumstances , because(B) requires only that one (or more) of the boys dislike

278 the

Chapter 33

guy

'

dislikes

The

( A

( B

( A

.

In

short

)

.

is

true

) ,

is

not

every

.

1

boy

might

. 3

section

,

the

.

NP

,

its

Bill

'

not

B

)

3

)

it

)

does

then

,

on

B

)

the

which

entail

)

is

'

false

.

( B

s

dog

does

,

not

be

guy

' s

)

but

entail

nevertheless

the

.

reading

guy

(

)

not

( A

true

likes

B

entail

but

is

may

boy

B

the

,

boy

(

not

true

dislikes

,

one

entails

in

(

is

(

boy

dog

)

does

which

s

least

( A

circumstance

that

every

false

dog

,

some

Bill

a

.

If

a

particular

time

is

a

is

of

tension

.

.

a

a

For

.

set

at

ferently

NP

,

ent

extension

the

bottom

Bill

has

(

,

in

of

it

is

we

have

of

a

VP

example

,

Clinton

.

On

,

then

.

This

the

sen

sleeps

extension

the

of

a

-

basis

sentence

the

notion

the

function

its

VP

sleeps

different

in

( Kripke

.

separated

Figure

of

. e

. ,

and

in

the

)

) ,

,

An

argument

of

the

to

in

its

-

ex

-

circumstances

property

of

sleeping

circumstances

)

states

other

of

NPs

has

this

of

of

the

dif

a

-

differ

discussion

special

.

affairs

behave

America

summarizes

intension

The

.

that

different

but

-

affairs

more

follows

possible

States

ex

?

of

little

from

the

For

argument

a

the

Clinton

circumstances

different

. 2

.

its

-

,

Bill

state

function

1972

the

is

it

mean

short

NP

possible

,

out

United

33

out

a

,

a

( i

extension

the

what

In

extension

the

intension

picks

times

of

,

is

set

is

that

its

together

a

.

of

extension

two

from

president

is

of

is

function

a

value

value

the

extension

intension

its

the

of

or

the

intension

circumstances

figure

extension

by

For

)

property

between

reference

is

designator

the

The

the

otherwise

the

,

( reference

.

connection

its

Putting

same

rigid

NP

,

.

-

expression

case

Bill

false

circumstance

out

a

at

the

a

the

different

the

is

and

Clinton

example

that

is

taking

ex

extension

true

is

whose

a

an

intension

affairs

times

For

of

linguistic

this

-

linguistic

denoted

is

and

in

expres

a

Clinton

in

sleeps

intimate

function

.

,

of

an

where

is

the

)

with

circumstances

individuals

different

.

value

,

1893

a

the

Bill

,

VP

VP

function

flesh

expression

example

a

the

can

The

of

states

the

we

linguistic

of

,

Now

function

function

is

,

notion

linguistic

of

property

that

the

associated

of

case

the

and

circumstances

the

.

there

is

function

extension

of

has

of

of

,

this

sleeps

(

expression

the

)

example

in

by

false

,

linguistic

the

set

NP

Frege

or

which

on

meaning

have

and

expression

of

,

Clinton

true

linguistic

Clinton

tension

either

For

individuals

followed

Bill

determine

based

class

the

( meaning

.

an

follow

,

one

to

intension

extension

Clinton

if

refer

individual

NP

Bill

of

to

that

that

will

,

value

intension

an

semantics

a

the

by

only

we

of

is

is

to

reading

meanings

used

or

property

NP

and

value

the

a

the

conditional

of

,

the

,

how

superset

.

is

to

,

is

formally

the

definitions

denoted

if

intension

Now

reading

entailment

more

that

Clinton

,

,

-

few

reference

of

true

a

addition

individual

truth

or

different

)

(

one

guy

if

is

with

a

of

truth

in

In

sleeps

an

argument

and

as

observation

properties

the

at

,

establish

least

even

which

term

.

as

is

this

to

:

B

which

of

likes

dog

concerned

consisting

intension

the

at

then

every

in

,

somewhat

,

Another

consisting

a

s

and

however

are

such

If

sentence

at

,

we

such

sentence

tence

is

(

,

Therefore

that

can

definition

boy

'

indicate

have

.

ample

We

.

the

in

these

the

every

reading

intension

,

VP

sleeping

ing

,

case

circumstances

of

guy

.

dog

also

is

the

true

the

true

Sets

expressions

be

is

is

That

be

three

each

not

)

circumstance

to

will

.

is

many

In

if

we

sentences

pression

to

,

On

the

First

,

,

of

According

subset

Ultimately

the

must

)

dislikes

( B

.

it

( A

are

.

dislikes

entailment

an

.

and

is

a

)

,

however

) ,

which

there

boy

true

.

this

of

,

is

some

,

words

every

)

( A

in

( A

boy

reading

of

hold

entail

shows

Entailments

sions

if

identifying

does

other

every

of

.

to

reading

boy

In

)

by

that

( A

one

,

on

that

is

whenever

not

( B

also

33

false

In

point

,

For

therefore

figure

)

The

does

reading

but

33

dog

.

converse

)

As

( A

s

it

-

.

class

At

of

Setsand Circumstances

279

Expression Extension Intension NP Individual Circumstances . Individuals Properties of Circumstances " Properties Individuals of Individuals

VP

Expression S

Extension Truth

Intension Circumstances

.

Value

Truth Values

Figure33.2 Extensionsand intensionsfor different kinds of expressions

Proposition = Characteristic Functionof a Set Circumstance

27

. .

Circumstance

49

~

True True

Circumstance

103

~

True

Circumstance

222

, . .

False

Circumstance

368

~

False

OR :

Proposition = Set of Circumstances Circumstance

27

Circumstance

49

Circumstance

103

Figure33.3 Two waysto think aboutpropositions

linguistic expressionsthat we will be concernedwith , sentences . Sincethe extension of a sentenceis a truth value, the intension of a sentenceis a function from circumstancesto truth values. Following common usage, we will call the intension (meaning) of a sentencea proposition . Propositions are functions from circumstancesto truth values. In other words, to know the meaning of a sentence(the proposition the sentence expresses ) is to know those circumstancesin which the sentenceis true and those in which it is false. Propositions can also be viewed as sets. That is, rather than viewing a proposition as a function from circumstancesto truth values, one can view it as a set of circumstances, namely, as the set of circumstancesin which it is true. As figure 33.3 illustrates, these two ways of looking at propositions carry the same informational content.

280

Chapter 33

Weare ready at last to state the relationship between entailments and sets. The definition

of entailment

can be revised as follows :

A proposition Q entails another proposition R if and only if every circumstance in which Q is true is also one in which R is true . or

Proposition Q entails proposition R if and only if Q ~ R. Also, we can now formally characterizethe subsetjsupersetreadings of ambiguous sentences . For example, we have establishedthat the (A ) (= (4 ) reading of (3) (repeatedhere) entails the (B) (==(5 ) reading, but not vice versa. (3) Every boy doesn't like a certain guy's dog. (A ) Every boy is suchthat he does NOT like a certain guy's dog. (i.e., All boys dislike the dog.) (B) NOT every boy is suchthat he likes a certain guy's dog. (i.e., At least one boy dislikes the dog.) Now we can state the subsetjsupersetrelation among these readings, based on the earlier definition

:

If reading Q entails reading R, but R does not entail Q , then Q is the subset

reading and R is the supersetreading. By this definition, (A ) is the subsetreading of (3), and (B) is the supersetreading. On the basis of the joint observations that the (A ) reading of (3) entails the (B) reading, but that the (B) reading does not entail the (A ) reading, we can concludethat (A ) is true in a subset of the circumstancesthat make (B) true (see figure 33.4). In some sense to be made clear, it is " easier" to make (B) true than to make (A ) true . Therefore , we will call (B) the general interpretation and (A ) the specificinterpretation .

Having observed that (B) is a more general interpretation of (3) than (A ), we can now say why people find it more natural to hypothesize (B). The reason is simply that people tend to prefer the more general interpretation of an ambiguoussentence. To explain this preference, we need only invoke an already familiar principle of

A

Figure33.4 Specificandgeneralinterpretations

282

Chapter 33

Matching Model , we are assuming that children have the same parsing preferences as adults do . Therefore , children too should favor reading (B) as the meaning they assign to (3). Therefore , this is the reading to falsify in the experiment . Perhaps another example will help to underscore our claim that the more general reading of an ambiguous sentence is the more " natural " interpretation . Imagine that you hear a used-car salesman on TV say, " Every car on the lot has under 100,000 miles on it ." Later, you learn that the ad was taken off the air because it violated truth -in -advertising laws . Would you expect to find that the salesman had lied because in fact none of the cars on the lot had been driven under 100,000 miles? This would be the specific interpretation . Or would you expect to find that he had lied because at least some of the cars on the lot had been driven over 100,000 miles? This is the general interpretation . It seems to us that people favor the general inter pretation of ambiguous sentences, that is, the reading that makes a particular sentence true in the largest set of circumstances.

33.4 Conclusion This brings us back to the experiment. The test sentence(1) will be false on one interpretation. This meansthat the negation of (1) will be true. The negation of (1) is (3), which is ambiguous. Our present task is to decide which interpretation of (3) to use in the experiment, (A ) or (B). (1) Every boy likes his dog. (3) Every boy doesn't like a certain guy's dog. (A ) Every boy is suchthat he does NOT like a certain guy's dog. (i.e., All boys dislike the dog.) (B) NOT every boy is such that he likes a certain guy's dog. (i.e., At least one boy dislikes the dog.) Weare guided by the strategy of plausible dissent, which instructs us to identify the interpretation of (3) that children will find most natural. Because(B) is the generalinterpretation, it is more natural. Therefore, it will be easierfor children to seewhy (1) is false if the (B) meaning of (3) correspondsto meaning2' Since this meaning will be on children's minds, they will most easily grasp why it is incorrect, as dictated by the condition of plausible dissent. What should be true in the context, then, is that at least one boy dislikes the guy's dog (not for all of them to dislike it). We end the discussionby adding a caveat. Our reasoning in selecting the general interpretation dependson two additional assumptions: that sentence(3) is ambiguous for children, and that children employ the samestrategiesfor resolving ambiguities as adults do. Only if both of theseassumptionsare correct would we expect children to find it more plausibleto deny (1) in circumstancescorrespondingto reading (B) of (3).

Chapter 34

Discourse Binding

Within the generative framework, many aspectsof grammaticalknowledge are representedas constraints, that is, as sanctionsagainst linguistic analysesof one kind or another. Constraints are negative statements. It is safeto assumethat not all children, perhapsno children, encounter evidencepertaining to constraints. The pertinent evidence would be information about which linguistic expressionsand meanings are prohibited in the target language. It follows from the absenceof such negative evidencein children's experiencethat knowledge encodedby constraints is not learned from experience. If not, then this aspect of linguistic competencemust be innately specified, as part of Universal Grammar. This is the familiar argument from the poverty of the stimulus. The poverty-of-the-stimulus argument and its conclusions are generally accepted within the domain of syntactic knowledge and have served as impetus for much researchon the early emergenceof syntactic principles. Similar researchhas recently been extended to children's mastery of semantic principles, although considerably lessis known in this area. Still lessis known about how knowledge of discourseprinciples develops. Nevertheless, certain discourseprinciples have been characterizedas linguistic constraints and are therefore subject to the same poverty-of-the-stimulus considerationsthat apply to constraintswithin sentencegrammar. In this chapter, we extend the poverty-of-the-stimulus argument to a specificdiscourseconstraint, called the closureconstraint.The closure constraint explains why the singular pronoun, he in the second sentenceof discoursesequenceslike (1) and (2) cannot be anaphorically linked to the quantificational antecedentsno mouseand every mousein the first sentences . (1) No mousecameto Simba's party. He was upset. a. No mouse cameto ~!!!!Q~' s party. & was upset. b. No mouse cameto Simba's party. >1-& was upset. (2) Every mousecameto Simba's party. He was upset. a. Every mousecameto ~!!!!.Q ~'s party. & was upset. b. Every mousecameto Simba's party. >1-& was upset. Application of the closure constraint is restricted to particular quantificational antecedents. The examplein (3) shows that a singular pronoun canbe related to an indefinite NP that appearsin a preceding sentence. That is, the discoursesequencein (3) is ambiguous, unlike the sequencesin (1) and (2). (3) A bear sleepwalkedinto Genie's house. He ate the spaghetti. a. A bear sleepwalkedinto g ~!!~ 'Shouse. ~ ate the spaghetti. b. :6 _.Q~~!: sleepwalkedinto Genie's house. ~ ate the spaghetti.

284

In

Chapter 34

addition

,

lated

to

tence

it

,

as

(

( 4

4

)

)

No

/

.

~

tied

,

called

straints

.

.

a

being

part

be

anaphorically

re

within

the

of

same

-

sen

-

.

' s

party

said

~

was

upset

.

34

these

This

. 5

principles

are

,

.

The

that

take

not

that

is

of

-

to

grammar

,

,

in

the

they

things

course

of

of

the

closure

biological

all

the

findings

that

draw

semantic

Therefore

early

the

view

con

and

sentence

.

human

is

,

linguistic

syntactic

emerge

the

it

representations

for

by

,

invited

apparatus

confirmed

of

form

of

-

more

experience

therefqre

do

dis

as

However

discourse

part

the

,

.

the

we

of

viewed

through

on

support

,

singular

experimental

principles

gained

Weare

should

is

results

Grammar

are

language

principles

an

knowledge

that

.

are

of

report

to

semantic

as

conclusion

we

commonly

constraints

the

,

.

argument

principles

of

.

constraint

or

of

. 5

stimulus

experience

component

antecedents

34

constraints

discourse

possible

section

-

constraints

knowledge

of

In

principles

discourse

.

upset

the

through

upset

was

discourse

from

was

he

syntactic

of

he

said

this

-

that

party

these

is

learned

Universal

language

can

appear

' s

range

of

linguistic

section

of

he

that

said

~

.

-

,

substantive

in

is

Q

the

poverty

knowledge

be

party

g !

restricts

than

development

reported

,

knowledge

about

,

!

because

Although

equal

language

,

principles

inferences

constitute

Simba

interest

other

constraints

at

' s

cannot

same

mouse

the

world

Like

presumably

~

experience

-

pronoun

NPs

sequences

distinguish

real

singular

' s

at

then

children

to

to

the

ery

Simba

mouse

discourse

special

important

study

constraint

endowment

for

.

.

1

Discourse

The

Binding

discourse

tween

constraint

pronouns

within

Because

space

Chierchia

'

During

,

we

to

in

(

introduced

5

)

,

in

chapter

governs

from

to

for

later

example

,

for

,

as

pronoun

the

.

to

,

the

can

-

1995

)

sketch

.

NP

may

estab

linked

as

will

be

.

of

explication

interpreted

links

-

dis

(

rudimentary

indefinite

be

our

Chierchia

anaphorically

anaphoric

be

frame

complete

)

be

he

Such

a

more

( 5

can

pronoun

sentence

than

a

such

relations

chosen

by

more

book

NP

a

first

have

advanced

offering

his

anaphoric

We

anaphora

indefinite

which

the

.

discourse

readers

an

referent

underlining

.

us

refer

with

sentence

boy

of

prevent

discourse

discourse

second

theory

theory

this

antecedents

limitations

s

in

quantificational

the

a

a

discussed

and

cussion

.

In

-

the

referring

to

represented

by

.

(

5

)

~

_E

Chierchia

'

conditional

~

s

Y

walked

here

content

)

of

is

a

,

,

.

The

by

(

6

)

[ 51

each

5

,

a

was

6

)

p

:

the

[ 52

in

a

discourse

&

q

]

.

*

is

variable

serves

The

[ 51

&

with

&

own

q

]

into

( bound

by

placeholder

propositional

one

its

a

be

in

variable

truth

,

in

by

discourse

a

-

operator

filled

of

sentence

usual

lambda

to

principle

in

its

52

a

partitioned

p

operating

variable

,

,

as

general

propositional

sentence

]

.

propositional

sentence

(

tall

variable

subsequent

&

&

sentence

and

subsequent

in

the

.

propositional

illustrated

placed

in

theory

content

omitted

ing

E

at

of

of

closely

In

/

certain

of

is

the

Every

application

course

lish

~

/

constraint

The

the

quantified

mouse

N

investigation

34

Every

No

in

that

universally

.

.

closure

pronouns

-

noting

illustrates

b

so

worth

and

a

The

is

negatively

bind

discourse

:

q

the

is

in

( 6

)

.

-

re

-

DiscourseBinding By

this

scope

process of

the

sentence

the

A

boy

logical

formula

in

conversion

in . -He

walked

was

representations within

; the

once

(8 )

3x .,lp [boy

(9 )

3x [boy

Illicit

Discourse

the

sentence by

in

the

( 7 ) finds

its

indefinite

way

inside

the

NP , a boy , in the

first

in

( 8 ) and of

- free

variable

(x ) &

walked

(x ) &

walked

between , however

discourse

trast

, a pronoun

NP

contains

quence

can

( 10 )

cannot

course

eliminated

operators

( 11 ) Working

follows been

The

of we

applied the

dynamic ( 12 )

No

will

call

of

dynamic

of

/' paste

[boy

second outside

negation

of

an

sentence

illicit

is

which

NP . By

an earlier

a con -

, if this

discourse

se -

1 , the

"

operator

, and

in

of

negation we

of dis -

proceeds

in

negation

is

a different

i , which

in

, this

scope

dynamic

is given

first

location require

call

the

,

two " paste

"

( 11 ) .

the

eliminates

, T , which

/ ' cut "

operator

first

, i 5 = sentence

the

original

is subsequently : 1 S =

, r , introduces

definition

walked

(x ) &

propositional dropped

S ( T ) . After

a new

this

propocsitional

2p

(5 & p ) .

in

( 10 ) is illustrated

in

. This operator variable

,

( 12 ) .

(x ) &

- in (x ) & p ]

walked

walked

walked

(x ) & is scope

T ]

- in {x ) ]

walked integrated of

- in (x ) & p ] - in (x ) & p ] (T )

- in (x ) &

{x ) & walked

the

of

domain

in .

(x ) &

sentence

the

is generated

parts

scopal ' s system

negation

within

" cut " operator

of

(x ) &

[boy

the

Chierchia

, =: ! . . Dynamic

v .ariable two

renders . In

variable

a tautology

of

1 3x Ap [boy

i - - , 3x [boy

falls

in

, which

negation

. By

walked

AP [- - , 3x the

NPs in

indefinite

example

binding

" cut " operator

negation

boy

i - - , 3x

pronoun

lambda

! [5 & p ]

definition

i - - , 3x Ap [boy

When

second

by

quantificational

a construction

NP

constraint

the

it with

3x AP [boy

i --,

the

.

pronominal

. These

interpretation

~

some

a preceding

. An

down

dynamic

out , the

, the

scope

sat

propositional

saturating the

quantifier

propositional

negation

inside

from

to

a quantificational

closure " to

=: ! . [ 5 & p ] = } i J

, p , by

and

( 7 ) illustrates

linked

to

from

definition

from

variable

the

, a new

scope

. The

first

q]

pronoun

. Example

in . . -He

existing

: ! , which

operator

outside

, the

; second the

the

.

tall (x ) &

a singular

linked

" closed

results

: first

, x , in

in

- in (x ) & p ] [ tall (x ) & q ]

a universal

walked

NPs

constraint

outside

be

( 10 ) illustrates

steps

variable

quantifier

( 10 ) .

boy

quantified

bound

anaphorically

or

in

No

Example these

be

negation

is given

, the

existential

Binding

dependence

pronoun

( 9 ) indicate

the

is now

- in (x ) &

in

singular

tall .

scope

Referential

has

second

, 3 , introduced

prohibited

two

the

quantifier

is brought

34 .2

pronoun

.

(7) As

, the

existential

285

the

- in {x ) ] & with

p] the

quantificational

first

(by

by

definition

of :::!.

by

definition

of

by

A - conversion

by

law

of

identity

by

definition

of

lambda

antecedent

conversion . As

shown

!

i

) , the in

( 13 ) ,

286

Chapter 34

the pronoun (i.e., the secondoccurrenceof x) is free. This accountsfor the fact that the pronoun in (10) cannot be anaphorically linked to the negative quantificational NP, no boy. (13)

. . . He sat down. Jvp[--, 3x [boy(x) & walked-in(x)] & p] [sat-down(x) & q] --, 3x [boy(x) & walked-in{x)] & [sat-down(x) & q]

Similarly, quantificational NPs with a universal quantifier in one sentencecannot bind a pronoun that appearsin a subsequentsentencein a discourse. (14)

Every boy walked in. *& sat down.

Anaphoric relations are prohibited in discoursesequenceslike (14), becausethe definition of dynamic universal quantification derives from that of dynamic negation: -V {::;- -. 3.- . 34.3 DiscourseConstraintsin Child Language The remainderof the chapter presentsan experiment designedto examinethe possibility that the closure constraint on discoursebinding may be operative in the grammars of young children. The experiment was designedand conductedin collaboration with Laura Conway; indeed, she should be credited with many insights in the experimental design. Considerationsof languageleamability in the absenceof negative evidenceinvite us to expect that children will adhereto the closureconstraint. In order to acceptthis experimentalhypothesis, however, we must reject the null hypothesis, which is that children's linguistic behavior will include violations of the constraint. Tuming this around, if the null hypothesis is correct, children will acceptthe bound variable interpretation of the pronoun in discoursesequenceslike (15). (15)

No mouse cameto Simba's party. He wore a hat. a. Meaning! : No mousecameto Simba's party. ;v.& wore a hat. b. Meaning2: No mousecameto ~ !!!.Q ~'s party. & wore a hat.

Children who lack the closureconstraint will be able to link the pronoun in the second sentenceof (15) to the quantificationalNP in the first sentence.This will permit them to assigna reading to (15) that is not availableto adults: meaningl, which we will call the boundpronouninterpretation. Another reading will be available to children who lack the constraint. On this reading, the pronoun refers to Simba. Sincethis reading, meaning2, is given by direct reference, we will call it the directreference interpretation. The anaphoriclink between the pronoun and the r-expressionSimbais indicated by underlining. If children lack the closureconstraint, they should find discoursesequenceslike (15) ambiguous. If they know the closure constraint, meaning1 will not be accessibleto them; the only interpretation possiblefor them will be meaning2, just as for adults. 34.4 EliminatingAlternativeHypotheses It is not straightforward, however, to reject the null hypothesis. Supposethat children consistently respond in an experimental task by assigning meaningz. Although this

DiscourseBinding

287

could reflect their adherenceto the closure constraint on discoursebinding, we must be certain that they are not giving the "right" answersfor the wrong reason. Other factors besides the closure constraint could cause children to consistently assign meaning2 to discourse sequenceslike (15), despite the availability of meaning! in their grammars . There are at least two alternative

scenarios to consider . On both of them , children

produce the "correct" pattern of responses , consistently assigning meaning2to the test discourse

sequences . However , children ' s responses are not based on the closure

constraint , but on some other strategy for interpreting pronouns in discourse. There -

fore, experimentalmaneuversmust be engagedto untangle the effects of the closure constraint and these alternative sourcesof children's "correct" linguistic behavior. One reasonchildren might consistently analyzediscoursesequenceslike (15) using meaning2 is that , on this analysis, the pronoun in the second sentence refers to the

most recently mentioned discoursereferent. This kind of strategy for relating linguistic expressionsis reminiscent of the Minimum Distance Principle (Chomsky 1969). Supposing that children systematically assign meaning2to discourse sequenceslike (15) (repeatedhere), we must be able to rule out this alternative explanation of their behavior .

(15)

No mouse carne to Simba's party . He wore a hat. a. Meaningl : No mouse came to Simba's party . ,y.~ wore a hat . b . Meaning2 : No mouse came to fu!!!~~'s party . ~ wore a hat .

To checkfor this possibility in the present study, items such as (16) were included as controls ; we will call them indefinite NP controls. The closure constraint does not

apply to the indefinite NP controls; the first sentencein (16) contains two legitimate discourse

antecedents

for the pronoun

in the second sentence . Use of these controls

therefore enabled us to identify any tendency children exhibited to interpret pro -

nouns as referring to the closestprecedingpotential antecedent. (16)

A bear sleepwalkedinto Genie's house. He ate the spaghetti. a.

C loser antecedent

A bear sleepwalkedinto g ~.!);~ ' s house. ~ ate the spaghetti. b.

More distant antecedent

...6_.Q ~~r sleepwalkedinto Genie's house. ~ ate the spaghetti. In the discourse sequence in (16 ), both the closer NP , Genie, and the more distant one ,

a bear, in the first sentenceare potential antecedentsfor the pronoun in the second sentence. In designing the experiment, we decided to enhancethe accessibilityof the closer NP as the antecedent

of the pronoun

in the indefinite

NP controls . To accom -

plish this, we made the referent of the closer NP the agent of the last event in the discoursecontext. For example, the last event in the context corresponding to (16) was performed by Genie, not by the bear. (This was not an event in which Genie ate spaghetti, however, for reasonsthat we will come to.) All things being equal, the interpretation associated with the last-mentioned event should be most salient; con-

sequently, the closer potential antecedentshould be favored over the more distant one in the ambiguous discourse sequencespresented in the indefinite NP control condition . If children assign - the more distant , indefinite NP , as antecedent for the pronoun, despite this contextual bias and despite the proximity of the other NP, then we can be confident that they are not relying solely on a parsing strategy suchas the Minimum DistancePrinciple in assigningreferentsto pronouns.

288

Chapter 34

All things are not equal, however. There is another important factor that biases children (and adults) toward one analysis of an ambiguous sentencerather than another. As noted earlier, (seechapters6 and 26)/ children tend to resolve ambiguities by assigningwhichever analysismakesthe sentencetrue in the discoursecontext. In experimental investigations with children, contexts are devised to make one meaning of a test sentencetrue and the other meaning false. In the present study/ we adopted this researchstrategy/ as follows: the secondsentencein every discourselike (16) was falsein the context if the pronoun was anaphoricallylinked to the closerNP; if the pronoun was linked to the more distant NP, the second sentencein the discourse was true. To the extent that subjectsare influenced by the strategy of interpreting sentencesin a discourse in a way that makes them true, then this should boost the availability of the representation in which the pronoun is linked to the more distant potential antecedentin the discourse(e.g., the indefinite NP in (16)). There is a good reasonfor designing this part of the experiment in this way. In the test discourse sequences , the corresponding analysis that links the pronoun and the more distant NP is meaningl, which is ruled out by the closure constraint: (17)

No mousecameto Simba's party. He wore a hat. Meaning! : No rnou~~ cameto Simba's party. .~ wore a hat.

If children consistently establishanaphoriclinks between the pronoun and the more distant NP where this is permitted, as in (16), but refuse to do so where this is prohibited, as in (17), then we would have compelling evidencethat they were adhering to the closure constraint, as long as alternative explanations of their behavior are ruled out. There is a secondreasonwhy children might consistently assignmeaning2to discoursesequencessuchas (15) that are governed by the closureconstraint, despite the availability of meaning1in their grammars: their parsersmight lead them to prefer the direct referenceinterpretation of pronouns over the bound pronoun interpretation. If this parsing preferenceexists, then children will assign meaning2 on most trials; meaning1will rarely surfacedespite its availability in children's grammars. One reasonfor thinking that children may prefer the direct referenceinterpretation (meaning2) of discourse sequenceslike (15) over the bound pronoun interpretation (meaning1) is that they contain an occurrenceof the singular pronoun he rather than the plural pronoun they as the form of the bound pronoun. The plural pronoun serves this grammaticalfunction in the grammarsof some young children (Thornton 1990). Such children sometimeseven produce agreement errors in order to use the plural form of the pronoun (e.g., 'Which one thinks they are the best?"). To establishthat the children in the present study could interpret the singular pronoun he as a bound pronoun, control sentenceslike (18)- the boundpronouncontrolswere included. The bound pronoun controls consist of single-sentencediscourses; but it seemshighly unlikely that children prefer one form of bound pronoun in sentencegrammar and another in discourse. (18) No mouseat Simba's party said he wore a hat. a. Bound pronoun: No mouseat Simba's party said ~ wore a hat. b. Direct reference: No mouseat ~!!!!.Q ~'s party said ~ wore a hat. In the experiment, the control sentenceswere true on the bound pronoun interpretation and false on the direct referenceinterpretation. Therefore, children's level of

DiscourseBinding acceptance like

of

the

( 18 ) can

in

be

response

to

who

fest

for

the

indefinite

makes

this

them

distant

. By

contexts

was

Therefore

, both

course

sequences

on

task

the

resolute There

is and

"

true

one the

had

the in

was

a

flavor

talking

, and

the

they the

A

Following

( 20

)

There

third

were

story

No

are

serves

the

that

were

more

is

a legit in

-

the

discourse

se -

interpretation

makes

to

would

that

the

event

test

pronoun

analysis

keep

in

that

the

the

perform

.

test

dis

-

successfully

therefore

mind

demonstrate

, however

meaning

a

: both

the

control

corresponding

interpretation

, in

to

which

the

the

pronoun

.

that

, both

as

the

of

these

proportion a

comparison

is

ruled

out

hat

a hat

of

of

who did

one a hat

the

most

abundance

they

on

own

wear

of

do

an

wear

put

their to

typical

discussing

that

not

did

not

a

that

this

has

not

at

the

see party

, and

.

made

- wearing

a

, and

because

all

of

the

hat

the

" Yes

constraint

.

correspond

"

design

trials .

earlier

facilitates

associated

, where

the

of

bound

is

last in

heighten control

the

,

none

, associating

available

should these

readings

assigning

with is

both

( because interpretation

noted

response

to

true

reference

) . As

interpretation

responses

to

is

direct a hat

is

experimental

experimental

closure

and wore

pronoun

a hat

interpretation

the

pronoun

the

)

:

( 19 ) that

interpretation

bound of

wore

in

Simba

with

bound

sentence

pronoun

, wore say

the

the

is are

assert

discuss

he

described

, that

for by

.

did

Simba

target

said

bound

aspects of

decided

the

interpretation

if

then

he

' s ' party

context

, also

that

assert

attraction

he

is

characters

, who

mice

that

that

, the

.

Simba

did

main party

. Two

J; nice

presents

that

the

believe

how

silly

mouse

party

, saying

. The

he , himself

one

the they

and

. The at

protocol

course

abundance

mice

at , so

a , of

a party

disagrees

the

. Therefore

. The

a hat

at

like

three

a hat

summarizes

protocol

hosting are

himself

such

. Notice

grammars

is

too

of

is

pronoun

story

words

, a puppet

aspects

said

use

recounts

mouse

context

bound

of

hats

interpretation

pretation

NP

, ( 19 )

not

mouse

mouse

( because

bility

closer

actual

on

each

the

the

last

with

way

NP

, the

and

children

reference

the

on

put

spectacle

in

the

. There put

Simba

the

to such

. In

lion

hats

already

mice

experiment

direct

and

closer

a

-

.

designed

technique

do

Simba silly

false

the

an

if

pronoun the

,

mani

. As

in

bound

judgments

constraint

the

controls

( 19 )

the

overridden negative

on

of

pronoun

the

assign

sentences

. Finally

the

they

Experiment

give

but

to

, only

controls to

pronoun constraint

sentences the

constraint

pronoun

bias

be

were

always

between

bound

closure

control

analyze

sequences

bound

closure

trials

to

performance

preference

pronoun

corresponded the

to

dependent

The

bound

and

constant test

response

the event

the

whatever

have

closure

the

by

sentences

their

( 15 ) . Because

bound

discourse

control

compare

prohibited

link

the

' s consistent to

referentially

34 . 5

both

bias

. Children

test obey

; this

this

adherence

trials

the

same

the

children

ambiguous to

override

anaphoric

the

children with

the

bias

the

, in

, if

associated

quences

To

favor

contrast

antecedent

to

, any

is

must

interpretation

will

! )

constraint

the

which like

( meaning

the

controls

true

NP

imate

" No

to

of

against

sequences

discourse

adhere

NP

baseline

discourse

the

assigning

interpretation

another

test

of

children

pronoun

as

the

interpretation

is

bound

used

289

this event

children its

stories pronoun

's

accessi

-

therefore inter

-

290

Chapter 34

Another feature of the story in (19) satisfiesthe condition of plausible dissent. This condition is relevant to the interpretation that is false in the story, namely, the direct reference interpretation . On this interpretation of (20), the pronoun he refers to Simba. For the direct reference interpretation to be felicitous , it must be clear to the

child that if events had taken a slightly different turn, the responseto the sentenceon this interpretation would have been " Yes" rather than " No ." For (20) to be true on

the direct referenceinterpretation, all three mice would have had to deny that Simba wore a hat . This could easily have been the outcome , since two of the mice say that

they did not see Simba wearing a hat. However, the third mouse disagrees, saying that he saw Simba wearing a hat. This last event makes sentence (20) false on the direct reference interpretation .

Children were also presented with two -sentencetest discoursessuch as (21), to determine the range of interpretations they allowed for discoursesthat are governed by the closure constraint . (21)

No mouse came to Simba's party . He wore a hat .

A child lacking the closure constraint should allow both the bound pronoun and the direct reference interpretations ; that is, this child should treat discourse (21) as am-

biguous. If so, the samefactors would weigh in the child's decision about which analysis to assign; the bound pronoun interpretation

should be accepted to the same

extent in response to (21 ) as it is in response to (20 ).

Two modifications were required for the experimentalprotocols corresponding to (21). First , it was necessary to alter the truth value judgment task in order to present discourses, rather than sentences, as in previous research. Because the evaluation of a discourse is not a composite of the truth values of the individual sentences of the

discourse, we could not ask children for a single response (see Groenendijk and Stokhof 1991). Instead, we had the puppet report two things about the story (the two sentences of the discourse ), and children

were asked to evaluate

each sentence sepa -

rately . If both sentences were correct , the child gave the puppet two of its favorite

treats; if only one sentencewas correct, the child gave the puppet only one treat. The second design modification was introduced to satisfy the condition of plausible dissent. As with bound pronoun control sentences like (20), the truth or falsity

of test discourse sequenceslike (21) hinges on the interpretation of the pronoun. However, in a discourse, anaphoriclinks are constructedin a different way. Sincethe pronoun is in the second sentence, only the truth or falsity of this sentence turns on the interpretation of the pronoun . It was decided, therefore , to construct contexts such that the first sentence in all of the discourse

sequences was true ; the second sen-

tence was true if the bound pronoun interpretation was assigned, and false if the direct referenceinterpretation was assigned. It follows that the condition of plausibledissent is relevant only to the secondsentence on the direct reference interpretation. The story summary provided in (22) illustrates how plausibledissentwas met for the two -sentencediscoursein (21). (22)

Simba invites his three mice friends to a party . There is lots to do, and

specialhats to put on. At the last minute, the mice conjecturethat Simba 's invitation

is a trick , to lure them to his house so that he can

eat them. Frightenedby this prospect, the mice decideto go to a nearby rodeo . Meanwhile

, the innocent

Simba decides that his mice friends are

DiscourseBinding

291

not coming to his party . To cheer himself up, he considers wearing one of the party hats, but finds that they are specially sized for mice, and will not fit a big lion like Simba. He decides, instead, to wear his new lion cape. Off at the rodeo , the mice decide to wear cowboy hats, so they will look like the cowboys at the rodeo . The first sentence in the discourse is true : No mousewent to Simba's party . The second sentence, He wore a hat, is true if the bound pronoun interpretation is assigned, but is false if the pronoun is interpreted by direct reference. To satisfy plausible dissent, the interpretation of this sentence, with the pronoun referring to Simba, must be under consideration at some point in the story . The sentence ultimately turns out to be false on this interpretation . Finally , it is clear to children "What really happened?" on the direct reference interpretation : the hats were too small for Simba, so he put on his new cape. One complete set of test items is provided in (23). ( 23

There

)

.

Indefinite

.

Every

c

.

N

to

.

Every

e

.

No

two

accept

from

3

to

5

were

The

The

main

an

( 23d

pronoun

.

wore

He

hat

up

for

a

.

hat

.

wore

a

a

hat

a

wore

with

Data

from

like

( 23d

In

response

)

a

hat

hat

.

.

.

to

five

unrelated

response

15

those

and

in

( 23e

a

bias

,

( e

.g

.,

un

a

-

bias

children

( 23a

-

ranging

c ) ,

and

in

data

age

from

12

) .

to

being

only

' s

grammars

trials

where a

,

like

the similar

suggestive

but

( 23

perhaps

b

) ,

quantified extent

.

where

NP .

every

19

%

of

.

Because

that

not

all

the

quantified

contained

b

.

is

the

with

permitted

time

to

permitted

and

a

sen

-

pre

-

bound

for

there

.

Children

constraint

of

the

the

the

the

closure

,

sequences

closure

) ;

,

of

us

discourse

incidence

time

NP

none

the

lower

( 23c

the

the

Finally

in

a

)

to

appears

of

) ,

tolerance

this

%

children

where

was

( 23

children

It

,

There

mysterious

.

that

77

( 23a

their

where

"

interpretation

like

two

NP

Yes

like

indicating

) .

and

not

,

distant

"

pronoun

none

remain

a

sequences

time

evidence

accepted

by

grammars

considered

as

sequences

contributed

whose

the

pronoun

did

were

discourse

of

and

bound

They

discourse

%

responded

singular

like

?

93

children

this

the

sequences

,

to

interpret

permit

to

"

pronoun

,

elements

discourse

children

the

interpretation

responses

between

course

.

he

he

a

wore

party

said

monitor

items

" Yes

( with

also

this

) ,

wore

He

) .

like

.

Similarly

We

quantificational

children

( 23c

) .

interpretation

sponses

be

e

follows

between

.

-

children

such

true

the

party

interspersed

to

He

.

' s

said

sessions

responded

link

like

noun

as

items

as

grammar

cludes

,

.

party

Simba

party

used

two

from

children

anaphoric

with

type

said

from

were

,

discourse

Did

' s

were

in

' s

party

' s

to

Simba

Simba

' s

Simba

came

at

item

analyzed

findings

tences

mice

fillers

puppet

analyzed

NPs

by

each

Simba

to

Findings

indefinite

of

the

at

interviewed

were

. 6

of

to

came

mouse

The

the

were

of

mouse

.

came

mouse

Every

No

trials

mouse

None

:

whatever

children

:

:

A

Every

trials

filler

Children

:

:

one

d

were

ambiguous

34

a

b

second

of

responses

of

,

the

constraint

was

no

is

-

" Yes

"

these

re

part

-

two

can

of

many

difference

,

both

of

findings

systematic

every

rejected

pro

sentences

majority

present

contained

bound

and

trials

types

like

of

dis

-

292

Chapter 34

34.7 Conclusion The results of the study discussedin this chapter tend to support the experimental hypothesis. However, it would be premature to reject the null hypothesis in light of the finding that two children acceptedviolations of the closure constraint. When presented with contexts favoring the bound pronoun interpretation, though, many children responded in accordancewith the closure constraint on discourse binding, rejecting the second sentenceof the test discoursesequencesover three-quarters of the time. The findings from the control conditions help rule out alternative accountsof these children's correct rejections. We can eliminate the possibility that children's negative judgments were due to a parsing preferencefor the direct reference interpretation over the bound pronoun interpretation, and we can eliminate the possibility that children's correct responsesreflect a performance strategy that makes closer NPs more attractive antecedents. On a larger front, the successfulperformanceby many children in this study makes it clear that knowledge of pragmatic constraints may be evinced early in the course of language development, just like knowledge of syntactic and semanticconstraints. This further reinforces the poverty-of-the-stimulus argument, in yet another component of the languageapparatus, where principles of discoursereside. For further discussionand empirical findings, seeConway 1997.

Chapter35 Universal Quantification

For the past 30 years, it has been widely believed that even children as old as 4 or 5

misunderstandsentenceswith the universal quantifier, such as (1) and (2). (1) Every farmer is feeding a donkey. (2) A farmer is feeding every donkey. English-speakingchildren, it is claimed, understandboth (1) and (2) in the sameway, to entail that every farmer is feeding a donkey and, in addition, that every donkey is being fed by a farmer (see, e.g., lnhelder and Piaget 1964; Roeper and de Villiers 1991a; Takahashi 1991). In this chapter, we present theoretical arguments and empirical evidence in favor of the view that children learning English analyze the universal quantifier just as adult English speakersdo. The basic finding is this. If shown a picture like that in figure 35.1, which we call the extra object condition, many 3- to 5-year-old children who are asked the question

"Is every farmer feeding a donkey?" will respond by saying "No ." When asked to explain this answer ("Why not?"), children point to the unfed donkey as the reason (e.g., Philip 1991, 1995; Roeper and de Villiers 1991a; Takahashi 1991). Similarly, when asked the question "Is a farmer feeding every donkey?", if there are farmers that are not feeding donkeys in the context, children will say "No" and point to the " extra " farmers. It seems that children are interpreting both (1) and (2) in the same

way, as demanding symmetry between farmers and donkeys. These responsesby children have been called symmetricalresponses , becausechildren appearto reject any asymmetry between donkeys and farmers-

the mapping must be one to one.

Inhelder and Piaget (1964) attempted to explain children's symmetrical interpretation of sentenceslike (I ) and (2) in nonlinguistic terms, as the result of their inability to distinguish part-whole relationshipsamong sets. A linguistic account of children's comprehensionfailures, called the SymmetricalAccount, has been advancedwithin the generative framework (Philip 1991, 1995). We will concentrateon the most recent version of this account (Philip 1995), according to which children ignore the surface position of the universal quantifier in sentenceslike (I ) and (2). 35.1 TheSymmetricalInterpretation To explain children's nonadult interpretation of sentenceswith a universal quantifier,

the SymmetricalAccountappealsto a distinctionbetweenquantificational elements , which may function as either determiners(D-quantifiers) or adverbs(A -quantifiers). In adult English, the universal quantifier everyis aD -quantifier. As such, it is a two -place relation; its domain of quantification (first argument) is limited to the individuals

Universal Quantification

295

Although the truth conditions statedin (3) contain two occurrencesof the universal quantifier- one ranging over farmersand one ranging over donkeys - the Symmetrical Account maintains that children hypothesize a single univers.al quantifier, but assign it scope over more than one nominal

constituent , as if it were an unselective binder

in the senseof Heim (1982). Although the SymmetricalAccount differs from Heim's unselective binding approach in several respects, there are enough similarities between the two approachesto make it worth reviewing the mechanismsof unselective binding. 35.2 Unselective Binding In Heim's system, indefinite NPs carry no quantificational force of their own. When an.indefinite NP, such as a boy, appearsin discourse, it is semanticallyrepresentedby an open predicatewith a free variable: (4) a boy ~ boy (x) The open predicate inherits its quantificational force in one of two ways. One pertains to indefinite NPs within the scopeof a quantificationalelement, such as everyor always. In such cases , the domain of the quantificational element is extended beyond its usual limits , so as to encompass

all disenfranchised

indefinite

NPs . That is , the

quantificationalelementis an unselectivebinder, with scopeeven over indefinites that it does not c-command. Alternatively , indefinite NPs can inherit quantificationalforce by Existential Closure, which inserts an existential quantifier to bind indefinite NPs that stand alone (i.e., with no quantificationalantecedent). An example of unselectivebinding occurs in conditional donkeysentenceslike (5),

in whichthe adverbof quantification , always , impartsits quantificational forceto both of the indefinite NPs, a boy and a dog. The semantic representation postulated by Heim for such sentencesis an extension of an earlier analysis, by Lewis (1975), who was the first to observe that adverbs of quantification can take scopeover severalindefinites at the sametime. The truth conditions that result correspondto a reading in which (5) is true only if every boy in the discoursecontext takes every dog that he has to the park. This is the strongreading. (5) If a boy has a dog, he always takes it to the park. In addition to the strong reading, donkeysentenceslike (5) have a weak reading. For (5) to. be true on the weak reading, all that is required is that eachboy take at least one of his dogs to the park. Following Lewis, Heim's semanticrepresentationof donkeysentenceslike (5) partitions them into three elements: a quantifier, a restrictor, and a nuclear scope. As the

sample tripartite structure in (6) indicates, the domain of the quantifier is given by the restrictor and may include more than one open predicate. The quantifier binds the variables of these predicates, unselectively. The nuclear scope states conditions that must be satisfiedby the quantified variables. Notice that this representationyields the truth conditions of the strong reading of the donkeysentence. (6)

Quantifier

Restrictor

Nuclear scope

ALW A YSx,y boy(x) & dog(y) &: has(x, y) takes-to-the-park(x, y) '(Always) if a boy has a dog, he takes it to the park.'

296

Chapter 35

Relative

- clause

framework

( 7 )

Like

Every

the

an

donkey

boy

adverb

)

,

)

and

how

as

the

) ,

truth

and

the

strong

To

To

The

,

:

that

,

quantifier

event

,

( 9

)

both

,

given

a

the

' s

is

A

farmer

.

In

( x

as

.

, y

.

takes

,

the

,

,

dog

both

,

this

that

as

he

true

owns

- clause

and

both

is

such

is

relative

binding

same

short

sentence

every

,

the

In

sentences

cases

account

)

have

reading

both

boy

and

have

the

.

of

has

sentences

several

( e

.

even

.g

.,

Determiner

On

quantifier

as

individual

.

children

farmer

is

feeding

)

and

a

a

( 2

) )

like

in

unselectively

variables

that

( 1

a

an

( a

the

the

semantic

donkey

and

the

-

of

,

' s

as

well

disjunction

tripartite

assign

re

children

way

binding

Adopting

the

adverb

,

second

-

universal

into

Account

adults

' s

main

with

in

Symmetrical

of

-

Heim

Account

analyzed

that

quanti

with

sentences

donkey

the

from

universal

common

Symmetrical

semantically

.

differs

The

and

a

in

simple

farmer

is

with

features

sentences

interpret

a

ART

( Earmer

P

ART

( donkey

,

( or

,

The

Nuclear

[ P

event

the

;

of

)

framework

,

representation

sentence

A

farmer

is

) ,

,

a

or

is

feeding

which

a

a

ones

is

in

the

every

the

donkey

- Eeeding

- a - donkey

participates

feeding

( e }

truth

of

domain

This

if

there

is

false

at

.

least

one

It

any

or

farmer

both

such

is

in

The

events

test

do

adopt

feeding

explains

is

donkey

children

not

not

this

a

why

who

by

events

. l

who

farmer

also

indicated

:

children

Every

discourse

is

if

,

disjunction

,

why

sentence

of

events

a

participates

explains

the

donkey

e .'

over

form

donkey

a

in

ranges

sentence

.

donkey

restrictor

a

the

them

or

a

every

the

which

that

in

in

- is

farmer

farmer

in

states

deny

scope

Earrner

quantifier

donkey

donkey

or

( e ) ) )

mentioned

scope

unfed

in

( e } )

universal

participates

nuclear

e ,

both

events

feeding

an

- park

donkey

( 7 ) .

analyzed

restrictor

sentences

strong

reading

donkey

Every

representation

there

- the

donkey

the

is

the

.

every

farmer

semantic

- to

unselective

strong

Restrictor

farmer

by

have

anything

Heim

scope

takes

every

of

maintains

( e )

e .

a

if

quantifier

sentence

representation

which

)

binding

the

children

than

Quantifier

variable

only

nominals

rather

the

' For

deny

in

every

inside

conditional

as

instances

universal

participates

this

, y

interpretation

as

the

Every

the

and

to

not

donkey

)

quantifier

:

unselective

quantification

,

every

( 9

if

analysis

.

- clause

such

Account

Account

to

( x

both

,

if

the

universal

and

Symmetrical

feeding

park

predicates

has

for

symmetrical

the

universal

variables

the

true

on

both

analyze

In

similar

universal

open

relative

derived

are

adults

,

children

the

counterparts

approach

is

of

to

the

&

sentences

is

,

)

that

Symmetrical

putting

That

analysis

,

( y

to

' s

)

unlike

by

quantification

in

a

Account

binding

.

dog

clear

sentences

( 1991

unselective

strictor

it

up

children

' s

it

all

conditional

Symmetrical

Philip

tains

is

&

corresponding

explain

fier

it

)

donkey

sum

conditions

. 3

takes

over

are

donkey

truth

) ,

reading

.

conditional

35

receive

Nuclear

( x

their

- clause

park

dog

always

conditions

relative

the

a

scope

boy

( 8

conditions

to

( 7 )

Restrictor

x ,y

( 6

has

with

Quantifier

Comparing

on

like

quantification

binder

EVERY

truth

who

of

unselective

( 8

( 5

sentences

.

feeding

Universal Quantification

297

The Symmetrical Account predicts that children should respond differently to sentences with intransitive verb phrases, such as Every cat is waving . This sentence contains only one nominal , cat, for the universal quantifier to bind . Therefore , such sentences are not susceptible to the symmetrical interpretation . The Symmetrical Account predicts only adultlike responses to these sentences. Any residual errors must be due to experimental noise. Indeed, Philip (1995) reports a significant increase in adultlike responses to sentences with intransitive verb phrases over adultlike responses to ones with transitive verb phrases. The higher incidence of symmetrical responses to the transitive sentences is advanced as support for the Symmetrical Account .

35.4 Problemswith the Symmetrical Account At first blush the Symmetrical Account has much to recommend it . It appears to explain several intriguing linguistic phenomena, including children 's apparent misunderstanding of simple sentences with universal quantification . Despite its theo retical interest , however , the account makes the wrong predictions about children 's interpretation of certain sentences. On the Symmetrical Account , children should respond to both kinds of donkey sentences- the relative -clause and the conditional - in the same way . However , as we describe in chapter 36, the results of a study by Con way and Crain (1995a,b) show that children do not do this . As a group , children in that study tended to reject the weak reading of conditional donkey sentences signif icantly more often than they rejected this reading of relative -clause donkey sentences. This finding is difficult to reconcile with the Symmetrical Account . Moreover , the evidence in favor of the Symmetrical Account is not overwhelming . Of the 276 children who were interviewed in the studies reported by Philip (1995), only 87 (32%) were J'pure symmetry children " (children who prefer the symmetrical interpretation over other interpretations ). This subgroup of children produced symmetrical responses only 57% of the time ; an adultlike interpretation accounted for the remainder of their responses. Moreover , experimental noise accounted for roughly half of their symmetrical responses. In sum, the symmetrical interpretation accounts for only about 10% of children 's overall responses in the experiments and for only about 30% of the responses by " symmetry children ." To continue our examination of the Symmetrical Account , it is instructive to look more closely at the relative proportions of three factors that contribute to children 's responses to questions with transitive verb phrases and the universal quantifier : (a) the symmetrical interpretation , (b) the adultlike interpretation , and (c) the amount of experimental noise. We continue our critique of the Symmetrical Account by showing that one of the Symmetrical Account 's fundamental claims, that the " symmetry child " prefers the symmetrical interpretation over an adultlike interpretation , is false. Consider children 's responses to questions with transitive verb phrases and to ones with intransitive verb phrases. (10)

Is every farmer feeding a donkey ?

(11)

Is every cat waving ?

Of the 49 children who participated in an experiment comparing sentences like these, 44 children produced nonadult responses to questions with transitive verb phrases.

298

Chapter 35

Included among the. 44 children were 25 pure "symmetry childrenil according to criteria invoked by Philip 1995, which excludes a different nonadult response. The "symmetry children" produced 61% symmetrical responsesto questions with transitive verb phrases, such as (10). In response to questions with intransitive verb phrases, such as (II ), this figure dropped to 40%. The difference between 61% and 40% was significant by a t test (p < .01). On the basisof this statisticalfinding, Philip (1995) infers that children's symmetrical responsesto sentenceswith transitive verb phrasesare linguistic in nature. With thesefigures, we can measurethe relative contributions of both linguistic and nonlinguistic factors in children's responsesto questionswith transitive verb phrases, such as (10). The 25 "symmetry children" produced 39% adultlike responsesto such sentences . On this proportion of trials, then, these children accessedan adultlike linguistic representation. The total percentageof children's nonadult responsesto questions with transitive verb phraseswas 61%. These nonadult responsesare due to a combination of factors, only one of which can be consideredto be part of children's linguistic competence(i.e., the symmetrical interpretation). The other factors are not the product of children's grammars, nor are they under experimentalcontrol; they are "uncontrolled secondaryfactors" (p. 109) or "generalized'noise' " (p. 124). To calculate the proportion of nonadult responsesdue to linguistic knowledge (the symmetrical interpretation), we need to subtract out the noise. 35.5 EstimatingNoise To estimate noise, we look at children's responsesto the control sentences . In the present study, the control sentenceswere questions with intransitive verb phrases, such as (11). On the SymmetricalAccount, there should be no nonadult responsesto these sentences ; either children assign these sentencesthe samesemanticrepresentation as adults do, or they assign them an event-quantincational analysis that yields the same truth conditions as the adult analysis. Therefore, all 40% of the nonadult responsesby children to sentenceslike (11) are due to noise. Having obtained an estimateof the percentageof nonadult responsesdue to noise, we can now calculatethe contribution of the svmmetrical interoretation to chi]dren' ~ L nonadult responsesto questionswith transitive'verb phrases. This is accomplishedby subtracting the noise, estimated at 40%, from the percentageof nonadult responses to the transitive sentences , 61%. The result, 21%, is a (liberal) estimateof how often children accessedthe symmetrical interpretation for questions with transitive verb phrases. To sum up, the responsesby "symmetry children" to transitive questionscan be partitioned as follows: these children assignedan adultlike interpretation 39% of the time; they assignedthe symmetricalinterpretation 21% of the time; the remaining 40% of their responsesare unaccountedfor. Thesefigures permit us to assessthe relative contributions of linguistic representations and nonlinguistic factorsto children s linguistic behavior in responseto sentences with a transitive verb phrase. The conclusion to be drawn is that the symmetrical interpretation is not preferred to an adultlike interpretation of questions, like (10), with the universal quantifier and a transitive verb phrase: an adultlike interpretation was favored nearly 2:1 by "symmetry children" when responsesto intransitive questions were used as the basis of the estimate. This casts doubt on one tenet of the Symmetrical Account: the claim that quantification over events is easier, and resides in

Universal Quantification

299

children's grammars earlier than quantification over objects. To the extent that a preferencefor event quantification is deemedimportant for the SymmetricalAccount, the account is undermined by the finding that an adultlike responseis at least as accessibleas the symmetricalinterpretation. 35 .6 Other Concerns

There are several other reasonsfor concern with the experimental findings. First, it turns out that uncontrolled factors accountfor as much or more of children's linguistic behavior than does the symmetrical interpretation. Second, it is a mystery why different proportions of children's responsesare attributed to the symmetrical interpretation for different constructions(or, to put it differently, it is a mystery why there are different levels of noise across constructions ). Third , the children 's strong bias to say "No " in response to uncontrolled factors is troublesome . Just the opposite is

usually reported in experimentalinvestigations of both children and adults (see, e.g., Grimshaw and Rosen 1990, 190). That is, subjects generally manifest a bias to say "Yes" when they are confusedor when they find sentencesdifficult to comprehend. The fact that the picture verification studies conducted by Philip and his colleagues evoked an unusualresponsebias from children is unexplained. This finding suggests that some feature of the task conflicts with children 's expectations about the circum -

stancescorresponding to the "Yes" responseson the meaning they assignto the test sentences . It is tempting to concludethat some feature(s) of the experimentaldesign encouragederrors that would have not occurred otherwise. A final problem with the Symmetrical Account is that it maintains that children find sentenceswith the universal quantifier ambiguous, whereasthey are unambiguous for adults. Attributing ambiguity to children's grammarscreatesa seriousproblem for the account, however . Because children have both interpretations , they can never ex-

punge the symmetrical interpretation. There are two ways to expunge an incorrect analysis: either parents provide correction when children assign the nonadult interpretation , or children notice that adults' behavior does not match their own . Neither source of error detection

is available

for the " symmetry

child ," however . First , cor -

rection by adults will never be forthcoming, becausewhenever children assign the symmetrical interpretation, they do so in circumstancesthat make the sentencetrue on the adult interpretation. (Notice that the symmetrical interpretation makes sentenceswith the universal quantifier true in a subsetof the circumstancescorresponding to the adult interpretation.) Second, the evidencechildren encounter will always be consistent with their preferred interpretation, the adultlike interpretation. Therefore , if the Symmetrical Account is correct , children will never converge on the adult grammar , because they cannot notice that adults allow only one of the interpretations that they allow . It follows that another tenet of the Symmetrical Account must be

abandoned, in light of considerationsof languagelearnability, namely, the view that children find sentenceswith a universal quantifier ambiguous. We have establishedthat the Symmetrical Account makes the wrong predictions about children's interpretation of relative-clause donkeysentencesand that a good casecan be made against the account as an explanation of children's nonadult linguistic behavior. Next, we ask whether there is a reasonablealternative explanation

for children's misunderstanding of sentences with a universalquantifier .

300

Chapter 35

35 . 7 An Alternative

Account

The Symmetrical Account claims that children prefer a particular nonadult reading of sentenceswith the universal quantifier. The dispreferred reading is the meaning assignedby adults (meaningl); the other reading at issue is the symmetrical interpretation (meaning2).

(12)

Is every farmer feeding a donkey? a. Meaning}: Every event with a farmer or a donkey in it is an event in which a farmer is feeding a donkey. b. Meaning2: Every farmer is feeding a donkey.

Sentence(12) should be unambiguousfor those children who have the adult grammar. By contrast, according to the SymmetricalAccount, (12) is ambiguous, with meaning! preferred (although we have shown that the latter claim is likely to be false). Let us start to design an experiment from the perspective of the Symmetrical Account. On this account, children are expected to make errors. Therefore, the experimental hypothesis is that children will not assignmeaningzalone; meaning1 will also be accessed . The null hypothesis is that children have only meaningz. Except for the experiment on bound pronouns , all of the previous experiments we have discussed associated meaning2, the adult interpretation , with the " No " response. To avoid type I errors, however , an experimenter who adopts the Symmetrical Account should choose to associate meaningz with the "Yes" response. Therefore , research by

Philip and others made the test sentencestrue on the adult interpretation. As noted earlier, in this design the target sentencescan be partitioned into background, assertion, possible outcome, and actual outcome. The assertioncorresponds to the actual outcome (the I'Yes" answer), and the possible outcome correspondsto the negative judgment (the "No" answer). For the question to be felicitous, the assertion must be in doubt at somepoint during the trial. Therefore, at somepoint during the trial , some outcome

other

than the actual one should

be conceivable . This is a

corollary of the condition of plausible dissent (i.e., one of plausible assent ). We summarize the experimentaldesign as follows: (13)

Background: Every farmer is feeding so-and-so. Assertion : a donkey Possible outcome : Some farmer feeds a dinosaur

Actual outcome : Every farmer feeds a donkey

The experimentalsetup used in the studiesreported by Philip (1995) did not conform to the experimental design outlined in (13). The main difference is the absenceof a possible, but not actual, outcome; no alternative to the actual outcome was under consideration. Therefore, the YesfNo questionswere not used felicitously. We contend that this was the source of many children 's nonadult responses in the extra object

condition, for example. On this scenario, children made nonadult responsesbecause the circumstanceswere inappropriate for a yes/no question (or a true/falsejudgment) on the adult interpretation. Placedin this predicament, children were led to suppose that another interpretation was intended . On the alternative interpretation , the ques-

tion concernedthe numerical correspondencebetween agents and objects. Children who inferred that the IIextra object " was relevant to their interpretation would conclude

that

the correct

answer

was

' IN 0 ." It would

also be clear

to these

children

that

Universal Quantification

301

the correct answer would be "Yes" if the extra object were removed . In short , the test questions were more felicitous on the symmetrical interpretation than on the adult interpretation . 35 .8 Loose Ends

Before we leave this discussionof experimental design, a few more remarks are in order. First, if the felicity conditions are satisfied, the presenceof an IIextra object" (e.g ., a donkey that was not fed by any of the farmers) should not influence the inter -

pretation children assignto the target sentences . Therefore; we are led to predict that children will respond correctly to the question "Is every farmer feeding a donkey?'! as long as it is askedin a felicitous context, regardlessof the number of donkeys in the situation. That is, if every farmer is feeding a donkey in the context, then children should respond affirmatively to the question. We should also consider why older children and adults do not fall victim as often as younger children to flaws in experimental design. We believe the reason is that older children and adults are simply better test-takers than young children. To be successfulin previous studies, subjectswere required to accommodatethe fact that the negation of the test sentences was not under consideration on the adult inter -

pretation. Presumably, older children and adults have learned to "seethrough" misleading circumstancesin which test sentencesare presented, although this leads to increased processing time , as Wason (1965) has shown . Many younger children are

apparently unable to perform the necessaryaccommodationsas rapidly and as successfully as older children and adults. One Anal comment on differencesbetween children and adults: Another population of adults- namely, adults with Broca's aphasia{also known as agrammatismor nonfluent aphasia)- are also prone to exhibit abnormal responsepatterns in circumstancesthat do not facilitate normal sentenceprocessing. For example, a nonfluent aphasic subject studied by Tyler (1985) manifested the same on-line responsepatterns as normal adults when presentedwith sentencesthat were both syntactically well formed and semanticallycoherent. However, this subjectmanifestedan aberrant profile when he encountered sentencesthat were syn,tactically well formed but semanticallyanomalous. Shankweileret al. (1989) interpret the Andings as evidence that

the syntactic processingcapabilitiesof Tyler' s patient are adequateunder ordinary conditions , where

both

syntactic

and semantic

cues converge

on an ap -

propriate analysis. But in adverse conditions syntactic processing may become derailed . (pp . 26 - 27 )

In light of the similar responsepatterns by children and nonfluent aphasicsin IIadverse" conditions, it might be expected that nonfluent aphasicswould give symmetrical

responses

in the same situations

that

evoke

them

from

children . Indeed ,

Avrutin and Philip (1994) found this to be the case. The observation that some young children and nonfluent aphasicsproduce the sameresponsesto quantificational sentencesin the samecontexts suggeststhat the difficulty is nonlinguistic for both groups and does not derive from children's nonadult linguistic representations, as the Symmetrical Account maintains .

302

Chapter 35

35.9 Conclusion

It has long been thought that children have difficulty understandingeven simple sentences with a universal quantifier. Severalaccountshave been offered, including the SymmetricalAccount, which is noteworthy becauseit incorporatesinsights from the syntax and semantics of generative grammar .

The Symmetrical Account clearly adopts the methodological assumptionsof the Competing Factors Model . The conclusion that children assign the symmetrical in -

terpretation to sentenceswith transitive verb phrasesis based on the reduction of nonadult responsesthey produce with other constructions (sentenceswith intransitive verb phrases, sentenceswith bare plural noun phrases, noun-incorporated structures, etc.). The fact that children continue to produce nonadult responsesas much as 40% to 50% of the time is attributed by Philip (1995) to "response strategies" (p. 109), "the effects of secondaryuncontrolled variables" (p. 107), or "strong carryover effects[that] were observedto confound the performanceof all subjects" (p. 124). From the vantage point of the Modularity Matching Model, accepting the Symmetrical Account would be tantamount to surrenderingthe fundamentalassumptions of the model : that children

and adults have access to the same cognitive

mechanisms .

These observations prompted us to investigate whether there was something in previous tasks, other than lack of grammatical knowledge, that might explain children 's errors . We discovered that those studies that had made the strongest case

for a linguistic account of children's errors had not satisfied the pragmatic felicity conditions for the task they had adopted, asking yes/no questions. In investigating children's linguistic competence, it is crucial to ensure that test sentencesare presented in felicitous contexts . In the contexts for yes/no questions, felicitous usage dictates that both the assertion and the negation of a target sentence should be under consideration . In tests of the Symmetrical Account , however , the target questions

were not felicitous, becausenothing in the task correspondedto the negative answer to the questions on the adult interpretation . In studies that elicited the symmetrical

interpretation in response to questions like "Is every farmer feeding a donkey?", children responded to pictures that led them to infer that the question of concern to the experimenter was whether or not there were equal numbers of farmers and donkeys. The answer to this was "No." From a methodological point of view, if the goal is to evoke optimal performancefrom children, it is important to present sentences in felicitous contexts, whatever their inherent complexity. Overly simplified contexts can actually impair children's performance, as compared to their level of performancein more complex but more appropriate contexts. (Hamburger and Crain (1982) make a related argument concerning children's performance in responseto sentences with restrictive

relative

clauses .)

In studies that met the felicity conditions we identified, children performed about as well as adults do in interpreting sentenceswith a universal quantifier; moreover, like adults, children produced such sentencesin appropriate contexts. For a report of the relevant findings, see Crain, Thornton, et al. 1996. On the basis of the findings from seven studies of 4 - to 6 - year -olds , we conclude

that children

do not lack gram -

matical competence. This opens the door for further studies of children's knowledge of universal quantification, and other aspects of quantification. We anticipate that these studies will offer additional results that are in keeping with the precepts of the theory of Universal Grammar .

Chapter 36 Donkey Sentences

This

chapter

concerns

sentences

like

( 1

)

( 2

relative

Relative

)

,

some

a

) ,

provide

clause

and

for

1991

) .

;

( On

it

( I

)

and

conditional

donkey

.

both

the

Kamp

The

farmer

not

the

universal

like

( I

farmer

truth

other

has

as

to

he

the

at

conditional

and

.

,

of

an

result

,

he

true

.)

Heim

binder

taken

to

- clause

sentences

to

Stokhof

donkeys

be

" unselective

relative

truth

according

the

to

is

-

the

and

one

has

relative

that

sentence

as

goal

both

reading

least

Reyle

One

Groenendijk

the

donkey

continuing

ensure

quantifier

a

and

Kamp

to

to

( see

for

As

.,

goals

strong

owns

feed

.

main

been

universal

donkey

,

.g

conditions

quantifier

the

( e

two

same

required

) ,

and

reading

had

must

is

1981

Theory

that

this

.

and

correspond

each

but

treating

same

.

donkey

sentences

,

it

1982

the

kinds

,

,

feeds

have

assigns

both

all

)

Heim

sentences

reading

nominals

( always

sentences

every

by

he

Representation

that

them

goals

receive

like

"

) -

sen

the

In

-

donkey

( 2

.

unselec

-

desired

.

. 1

The

Preferences

in

Interpretation

Symmetrical

tences

.

Account

Children

universal

.

should

rej

adopt

ect

relative

.

at

one

donkey

sentences

On

)

has

have

Chierchia

' s

of

dynamic

of

the

the

,

binding

.

from

truth

relative

used

in

on

in

a

relative

conditional

donkey

( i .e

.,

the

discourse

anaphora

are

.

According

the

not

;

both

) .

and

to

sentences

farmer

reading

clause

they

weak

should

sentences

strong

relative

,

the

every

children

-

a

donkey

to

which

these

( corresponding

donkey

- clause

binding

in

,

sen

with

correspond

context

Moreover

donkey

sentences

unselective

that

that

conditions

- clause

simple

of

based

conditions

assumption

- clause

of

contexts

( 1 )

donkeys

truth

the

same

account

in

sentences

same

relative

reading

is

reject

his

challenged

to

strong

assign

should

,

donkey

have

( 1995

the

sentences

they

all

extend

interpretation

only

donkey

,

- clause

should

children

not

should

Chierchia

mechanisms

but

relative

constructions

assign

example

,

)

symmetrical

analysis

- clause

For

least

distinguish

) .

the

35

the

should

Because

interpretation

feeds

( chapter

who

quantifier

sentences

ing

weak

donkey

bind

effect

like

feeds

with

donkey

of

feed

,

Discourse

semantics

feeds

the

these

tences

donkey

donkey

donkey

- clause

tively

of

sentences

may

achieves

analyses

farmer

he

relative

a

Beginning

of

a

every

owns

a

versions

conditional

conditions

which

owns

owns

.

linguistic

to

who

farmer

current

been

36

farmer

background

through

1993

sentences

Conditional

If

First

donkey

clause

Every

( 2

- clause

) .

conditional

strong

interpreted

read

using

to

Chierchia

-

the

,

304

Chapter 36

these

mechanisms

the

of

conditional

to

clause VP is

establish

definition

is

the its

repeated

comes

to

given

in

the

dynamic

weak

reading

conditional

for

more

customary

counterpart

in

the

consequent

; by

within

its

. A

reside

relative

, = . . In

scope

the ,

the

dynamic brief

- clause

donkey

metamorphosis indefinite

binding

sketch

NP

the

the

in

, material

of

sentences

from the

in

antecedent

the

essentials

of

, by

dynamic

main

the

clause

derivation

(3 ) . .

(3 )

In

the

Every

farmer

3y

( donkey

(y ) &

owns

(x , y ) ) ) =

b .

Vx

[ ( farmer

(x ) &

3y

( donkey

(y ) &

owns

(x , y ) ) ) ~

3y

( donkey

conditional

, as

Chierchia

than

adults

interpretations

world

more

36 . 2

The

test

a

experiment

this

exactly

existential , the

rep

-

interpretation

of

mechanisms factors

of , the

dy

-

strong

. and

children

conditional to

influenced

' judgments

by

general

world ' s experi

influence

of

so , children

apparatus

more

alternative

children

the

. If

them

the

. Because to

donkey

draw

about

subject

language

should

have

the

ale

,

1996

both

truth

Conway ) , we

-

general

' s judgments

.

- clause

judgment

of

saying

what

he

predict

what

would

the

story

Thornton an

and

task

to

of

, and conducted

relative

value

close

relative

same .

If

sentences

(4 )

the

tried

and

test

a.

test

two and

I know the

I know the

park

donkey

, to

main

had in

designed sen

innovation

of

child

in

story

. Kermit

what

he

corresponded

strong

interpretation

sentences

are

a lot park a lot .

given

to

donkey test in

conditional

the

weak

were

. Children

conditional

and

were

a re -

had

said

to

with

. The were

sentences

interpretation

available

tested

sentences sentences

donkey

four

stories

of

children

,

relative

the they

- clause

corresponding

virtually

identical

to . Typical

(4 ) .

about

boys

and

dogs

. Every

about

boys

and

dogs

. If

boy

who

has

a dog

takes

it

. a boy

has

a dog

, he

takes

it

-

the

happened

the

the

; Conway

donkey

already

happen remind

1995

experiment

conditional

. The

thought

sentences

stories

conditional

sentences

to b .

the

the

- clause

- clause

. The

only

rejected

rela ; tive and

( Crain , et of

: instead

at

the

sentence

- in

Conway

of

Frog

for

target

donkey

the

- clause

easily

be

the

.

protocols

were

adults

the

by

experience not

, the

judgments

be

by

; therefore

pragmatic

expect

may

as of

, Thornton

prediction

happen

The

extent

VP

cases

. Adult

should

understanding

the

the

would

most

requires

principles

variant

was

, Kermit

peated

same

Laura

a ; Crain 's

, using

story

the

bound

, by

would

sentences

y ) is - clause contrast

relative

reason

judgments

basic

with 1995

children

tences

we

(x , y ) ) ]

Experiment

Crain

to

lead

to reflect

collaboration

and

)

, their

degree in

knowledge

main . By

between

following

such

limited

directly

drawn

the

primarily

large

, then

( ambiguous

knowledge

may

determined a

available

be

the

. Acquiring

is

not

( feeds

(x , y )) ]

variable

in

, to

suggests , for

of

knowledge

interpretation

readily to

( i . e ., the

weak is

more

feeds

NP

influenced

di .stinctions

(x , y ) &

indefinite

the

sentences is

owns

pronoun

the to

being are

clearly

both

with

but

interpretation

In

, ( 3b ) , the

donkey binding

ence

( y) &

corresponds

sentences

it .

( x ) &:

formula

there

feeds

[ ( farmer

resentation

If

a donkey

\ Ix

associated

namic

owns

a.

final

quantifier

who

to

Donkey Sentences

305

Figure36.1 Contextcorrespondin ~ to the weakinterpretation

Although the experiment featured stories that were acted out in real time, the out. come of the story corresponding to (4), which we call a weakcontext , is statically represented,in figure 36.1. The story depicted in this figure involved four boys and six dogs. One boy had one dog, one boy had two dogs, one boy had three dogs, and one boy had no dogs.! After the characterswere introduced, but before the story was presented, Kermit uttered the test sentence,either (4a) or (4b). The experimenterthen acted out the story, according to the protocol in (5). (5) Theseboys decidedto take their dogs to the park becauseit was sucha nice day. The boy with one dog preparedhis dog to go to the park by putting a leashon it . The boy with two dogs put a leashon one of his dogs (the dog that was awake), but the other dog was sleeping and could not go to the park. The boy with three dogs put a leashon one of his dogs becausethat dog was awake, but not on his other two dogs, who were asleep. The boy without any dogs went along with the other boys to the park. Thus, during the courseof the story, eachdog owner took only one dog to the park. Therefore, a child who assignsthe weak reading should say that the puppet' s statement is true; a child who assignsthe strong reading should say that the statementis false .

Fifteen children participated in the experiment. They ranged in age from 3;7 to 5;5 (mean 4;5). Six of the children had consistently given symmetrical responseto sentenceswith a universal quantifier and a transitive VP in an earlier experiment, using pictures. The results are as follows. Overall, children accepted(a relevant subsetOf)2 the relative-clause donkey sentencessignificantly more often than the conditional

306

Chapter 36

donkey

sentences

priate

to

P

<

. 05

a

similar

,

the

) .

The

6

of

The

not

all

children

no

child

tences

the

the

child

all

not

selective

4

either

the

The

by

the

surprising

of

since

it

is

all

widely

on

all

inter

that

-

children

- clause

targets

their

.

By

or

for

,

In

extremely

assigned

the

conditional

these

;

contrast

.

were

,

grammars

interpretation

trials

that

;

weak

weak

targets

strong

reading

accepted

the

in

children

strong

the

relative

- clause

the

. 12

manifested

evidence

With

relative

reading

the

than

-

6

) .

available

individual

weak

availability

.

was

assigned

responses

%

appro

=

accepted

stronger

alike

reading

always

assigned

.

even

for

children

,

60

not

- square

earlier

often

versus

sentences

weak

were

( chi

children

more

%

interpretation

,

that

significant

; ' these

( 89

the

contexts

responses

sentences

donkey

in

symmetrical

provide

of

) ,

constructions

donkey

strong

trials

binding

,

con

-

strong

donkey

sentences

fact

sen

involve

un

-

-

.

Conclusion

observation

donkey

that

sentences

pret

two

that

/ 28

was

the

sentences

types

targets

is

The

the

( 13

difference

subjects

two

%

The

given

- clause

targets

on

. 3

had

individual

the

conditional

:

46

.

demonstrated

reading

versus

donkey

from

treat

sistent

)

who

relative

conditional

with

/ 43

between

required

with

the

children

resists

universal

metrical

strategy

is

the

tactic

that

as

our

it

studies

quantificational

roughly

maintain

govern

demonstrate

Grammar

sentences

' s

that

. .

This

is

Here

view

,

have

,

,

,

and

they

as

is

the

all

support

The

even

data

-

consistent

syn

directly

-

universal

this

.

understand

-

Sym

same

to

to

inter

the

research

adhere

quantification

acquisition

clause

appropriate

assign

attempted

universal

such

that

children

correctly

language

the

relative

children

development

do

children

child

our

adults

we

which

binder

view

of

young

brings

In

this

,

understanding

of

to

unselective

language

On

as

.

interpretation

according

an

.

of

.

analyses

children

sentences Universal

view

to

these

to

it

35

Grammar

analyses

pertains

if

weak

account

chapter

strongest

Universal

the

an

as

in

the

of

accept

on

considered

semantic

constraints

point

to

theory

and

readily

explanation

quantifier

Account

with

with

children

conditional

data

did

( 37

interpretation

of

pretation

of

%

distinction

interpretation

36

86

strong

view

-

findings

complex

in

line

Chapter37 A PotentialDrawbackof the Task

There is no absolute guarantee that , if the experimenter makes an interpretation available to child subjects, they will generate that interpretation . For example, if a sentence is ambiguous for children , we cannot take consistent "No " responses to one reading as evidence that they cannot generate the reading that is associated with the "Yes" response. Rather, the preference for the alternative reading might be great enough that even the explicit presentation of the first reading does not make it readily available to children (or adults). This puts a limit on the effectiveness of the truth value judgment task in certain instances. In these cases, it is not always possible to tell whether a child 's pattern of responses indicates a strong preference for one reading of an ambiguous sentence or a genuine ,grammatical prohibition against the other reading . In this chapter, we will illustrate the problem with an example . In the next two chapters, we will illustrate some ways to overcome the problem , in specific cases. 37.1 Preferencesversus Principles The need to distinguish preferences and principles has not taken hold , even among researchers who work within the generative framework . Perhaps the following remarks will make the importance of this distinction clear. Generally , people use ambig uous sentences in contexts that are consistent with only one of their interpretations ; it is rare for a sentence to be used in a context that is appropriate for more than one of its interpretations . Although such circumstances are rare, consider an ambiguous sentence for which one interpretation would be strongly preferred if the context were consistent with both interpretations . In certain instances in ordinary life , the sentence will be used in contexts that are consistent only with its dispreferred interpretation . In such contexts , the perceiver is compelled to disregard the preferred interpretation and seek out the dispreferred interpretation . This may take some cognitive effort , but apparently it is something perceivers manage quite well for the most part . In ordinary contexts , then, the perceiver can correctly understand a sentence, whichever inter pretation is intended by the speaker, although the cognitive demands may be greater in certain contexts than in others . Children 's linguistic knowledge is rarely tested in ordinary contexts , however . In experiments with children , the contexts that researchers construct are generally consistent with more than one interpretation of the test sentence. One is the adult inter pretation ; the other is the interpretation prohibited by the principle being investigated . Lacking the principle , children may have access to both interpretations of the test sentence, but they could still favor one over the other . Suppose that children favor the interpretation that is consistent with the adult grammar . If the sentence is false on this interpretation in the (ambiguous ) experimental contexts , then children may reject

308

Chapter 37

it , because it is false on the interpretation that comes most easily to them . Therefore ,

we cannot infer from children's "No " responsesthat they do not commandthe interpretation associated with the J' Yes" response. The interpretation associated with the

JJYes " responsemay simply be less accessible . 37 .2 Plural NPs

Consider an example. In a situation in which there are three octopuses, each one holding a hamburger, what is the answerto the following question? (1) How many hamburgersdo the octopuseshave? Clearly, the question is ambiguous. If the questioner intends to ask, "How many hamburgersdoes eachoctopus have?", then the answeris 'lOne"; but if the questioner intends to ask, I'How many.hamburgersdo the octopuseshave altogether?" , then the correct answeris I'Three." When the plural NP the octopuses is interpreted as referring to the octopuses as a group (i.e., on the collective interpretation of the plural NP), "Three" is an appropriate answer. When the plural NP is given a distributive interpretation , however , 'lOne " is correct . As child subjects listen to the test sentence in (1), they will no doubt assign one or

the other interpretation of the plural NP the octopuses . It is well establishedthat adults rapidly resolve ambiguities in context. This is surely true of children, too..However, in the real world , the context surrounding an ambiguous sentenceis usually appropriate to only one of its interpretations. Therefore, listeners can ordinarily appeal to featuresof the conversationalcontext to resolve ambiguity. However, in the context

just described , the plural NP in the questioncan correctlybe assignedeither the distributive or the collective interpretation, Accordingly, either answer is correct. Nevertheless, if children re~olve ambiguities as adults do, then they will favor one interpretation of the plural NP over the other. If there is a preferencefor one interpretation over the other, they might consist~ntly give just one of the possible responsesto the question in (I ), becauseit is ambiguous. Supposewe want to know whether children have both interpretations . How would this be tested?

To begin, we would advise adopting the researchstrategy used in the strong crossover experiment , in which the questions were embedded as indirect questions . This enables the puppet , Kermit the Frog , to express both the question and its answer. That is, Kermit could give either the IIOne" answer or the IIThree" answer on

different experimentaltrials, as in (2). (2) a.

I know how many hamburgersthe octopuseshave. Three. or

b. I know how many hamburgersthe octopuseshave. One. Embedded questions permit the experimenter to present the alternative interpretations individually for evaluation by the child . For adults, both of Kermit the Frog 's state-

ments are true. That is, both (2a) and (2b) are accuratedescriptionsof the situation. A problem remains, however. Supposethat children have a strong preferencefor one interpretation

over the other . If they process the indirect question on-line , as

adults do, then they could decide on the correct answer beforeKermit produces the answer. A child's acceptance,of Kermit's description of the situation dependson the correspondencebetween Kermit's answerand the answerthe child expectsKermit to

Potential Drawback of the Task

309

give. If Kermit gives the expected answer, the .child will almost certainly say that Kermit said the right thing. However, if Kermit does not give the expected answer, then one of two events may take place in the child's mind. On the one hand, the child may be able to accessthe alternative interpretation. For example, if children generally favor the distributive interpretation, they may neverthelessbe able to mentally regroup, with the goal of making Kermit's statementtrue, conforming to the bias to resolve an ambiguity in a way that makesthe speakersay something true. It is easy to imagine that the collective interpretation will be readily available to children if they follow this bias, so despite their preferencefor the distributive interpretation, they will say "Yes"- that is, that Kermit said the right thing. On the other hand, it is easy to imagine that children have such a strong preference for one or the other interpretation of plural NPs that they will judge Kermit's answer to be wrong. On this scenario, if children prefer the distributive interpretation, but hear Kermit give a collective answer, they may not be able to retrieve the alternative interpretation. If not, they will say "No ." Given these considerations, supposethat children are found to consistently reject one kind of answer to embedded questions like (2a- b). Such a finding presents a dilemma. We cannot tell whether the interpretation that children reject is strongly dispreferred or whether it is not available at all at a particular stage of language development. Exactly this situation arose in Miyamoto and Crain's (1991) study the acquisition of the distributive and collective interpretation of plural NPs; it was found that the majority of children systematicallyrejected the collective interpretation of the plural pronoun they in sentencessimilar in form to (2a- b). In chapters 38 and 39, we describe two ways to resolve this dilemma.

Chapter38 Resolvingthe Dilemma: Control Sentences

As described in the previous chapter, the truth value judgment task faces the problem of distinguishing whether children 's responses reveal (a) a strong preference for one reading of an ambiguity over another possible reading or (b) the absence in children 's grammars of all but one reading of what constitutes an ambiguity in the adult grammar. There are two ways to overcome this limitation in certain cases: by including the right control sentences or by varying the context in which a single test sentence is presented . We discuss these strategies in this chapter and the next , respectively .

38.1 StrongCrossover In some of the experimentsdiscussedearlier, we were able to avoid the problem of distinguishing preferencesagainst a particular reading from its absencein child grammars. For example, consider the experiment on strong crossover that investigated whether or not children adhereto Principle C of the binding theory. For the experiment, the null hypothesis is that children lack Principle C. LackingPrinciple C, children could interpret crossover questions like (1) and bound variable questions like (2) in the sameway. That is, (1) would be ambiguous, just as (2) is. The experimental hypothesis is that (1) is unambiguousfor children, just as it is for adults, and that (2) is ambiguous, just as it is for adults. (1) Crossover I know who he thinks has the best smile. Grover and Yogi Bear. (2) Boundvariable I know who thinks he has the best smile. Grover and Yogi Bear. More specifically, the null hypothesis maintains that the child could interpret the pronoun hein (1) as a bound pronoun (meaning}), perhapswith multiple referents; the deictic interpretation of the pronoun (meaning2) would also be possible, according to which he refers to a single individual. On the experimentalhypothesis, the crossover question is unambiguous: only the deictic interpretation of the pronoun is allowed. The experiment was therefore set up as follows, where meaning} was associatedwith th e IIYes" response: . Ho: Children lack the constraint on strong crossover. Expected results: Children permit both meaningl and meaning2. . HI : Children know the constraint on strong crossover. Expected results: Children permit meaning2, but not meaningl' Supposethat questions like (1) were tested in isolation and that children consistently assigned the deictic interpretation of the pronoun. There are two ways of

312

Chapter 38

interpreting these data~ (a) they support the experimental hypothesis, or (b) the deictic interpretation of a pronoun is preferred over the bound variable interpretation. Both possibilities are consistentwith the hypothetical findings. This is where the control questions - the bound variable questions -

come into

play. On the null hypothesis, these questions should be interpreted in exactly the sameway as the crossover questions. In particular, since both questionsare ambiguous on the null hypothesis, both types of questionsshould allow a multiple referent interpretation of the pronoun at least some proportion of the time . Of course, the alternative , deictic interpretation of the pronoun will also be available ; the two inter pretations will thus compete . This means that the proportions of responses that are

allotted to one interpretation or the other will depend on the various factors that govern ambiguity resolution. Moreover, if the null hypothesis is correct, whatever pressures favor one interpretation over the other will hold for both crossover ques-

tions and bound variable questions. Therefore, if both sentencesare presentedin exactly the same contexts, the alternative interpretations of questions like (1) and (2) should be observed, to the sameextent. For example, if the multiple referent answer is accepted50% of the time in responseto the bound variable questions, then this answer should be accepted50% of the time in responseto crossoverquestions. .

Ho : Children

lack the constraint

on strong crossover .

Expected results: Bound variable questions : Crossover questions :

Deictic meaning

Bound variable meaning

50 %

50 %

Deictic meaning

Bound variable meaning

50 %

50 %

On the other hand, if the experimental hypothesis is correct, then subjectsknow the grammaticalprinciple; therefore, they should never assignthe interpretation prohibited by the principle (the bound variable interpretation). According to the experimental hypothesis, the crossover questions are unambiguous, in contrast to the bound variable questions , which are ambiguous . Whatever tendency there is to re-

spond to the ambiguous bound variable questions by assigning a multiple referent interpretation of the pronoun should be overridden by the grammatical principle, Principle C, in crossoverquestions. Therefore, we wpuld expect a different pattern of responses by subjects to crossover questions and bound variable questions . Subjects

will give the multiple referent answersomeproportion of the time to bound variable questions , but never to crossover questions . In short , whatever tendency subjects ex-

hibit to assign the bound variable interpretation to the ambiguous control sentences will be,obviated by the grammaticalprinciple prohibiting this interpretation of crossover questions. This contrastswith the prediction of the null hypothesis: that subjects will exhibit the same proportion of bound variable responses to both types of ques-

tions. The data from the experiment can unequivocally test between thesecompeting hypotheses, at least if subjectsrespond by giving the multiple referent interpretation of the bound variable questions a reasonable proportion of the time . .

HI : Children know the constraint on strong crossover .

Expected results: Bound variable questions : Crossover questions:

Deictic meaning

Bound variable meaning

50 %

50 %

Deictic meaning

Bound variable meaning

100 %

0%

Control Sentences

313

In the present case, both (1) and (2) could be presented in the same contexts , because the null hypothesis claimed that both sentences had the same range of meanings . In our presentation of the experiment , this enabled us to randomly match sentences with contexts . That is, the child could randomly decide which story to listen to on a particular trial (by choosing the bag of toys for the next story ), and the experimenter could choose a sentence of one type or the other to present on that trial . This eliminated any effects, favoring one reading or another , that could be due to the context . Also , because the same contexts could be used for either type of construction , the order of events could be held constant in all of the contexts . If the order of events favored the bound variable interpretation of one type of question , then} by hypothesis , this interpretation would be favored to the same extent for the other type of sentence. 38.2 DiscourseBinding

Severalfactors makeit impossibleto achievethe samedegreeof control in other cases. Nonetheless, even in thesecaseswe were able to distinguish grammaticalknowledge from preferencesto a large extent. Again, the strategy was to comparechildren's interpretations of control sentenceswith their interpretation of test sentences . The study of children's adherenceto the closure constraint on discoursebinding is a good example of using a different structure as a control. Recall that, in a test of the closure constraint, children were presentedwith linguistic items like (3) and (4). The null hypothesis was that children would permit an anaphoric link between the pronoun and the quantified NP in both (3) and (4). The experimentalhypothesis was that only (4) permits suchlinkages; they areprohibited in (3) by the closureconstraint. (3) No mouseca~ e to Simba's party. He wore a hat. (4) No mouseat Simba's party said he wore a hat. Items like (3) and (4) cannot be presentedin identical contexts, however, because(3) consistsof two sentences , and (4) of only one. In the experimental situations acted out for (3) and (4), however, the bound variable reading is associatedwith the "Yes" response. The order of events in both contexts also favors the bound variable reading- it is the last event acted out. If the order of events favors the bound variable interpretation of one type of sentence , then, by hypothesis, this interpretation should be favored to a similar extent for the other type of sentence. Supposethe findings of the experiment are as predicted by the experimentalhypothesis; that is, children reject any link between no mouseand he in (3), but allow it in (4). Nevertheless, it could be maintainedthat the lack of a bound variable reading in (3) results from some unexplained differencein the properties of (3) and (4). It is logically possible that (3) does allow an anaphoric link between the pronoun and the quantified NP but that this link is highly dispreferred. There could be some hidden differencebetween the two sentencetypes that results in a stronger preferencefor the bound variable reading in (4) than in (3). But this would be a post hoc explanation of the pattern of responsesthat the experimental hypothesis predicts. Moreover, anyone advocating such a position would be obliged to explain why there are strong parsing pressuresin one direction for one kind of examplebut in the opposite direction for the other kind of example. In the absenceof independent

314

Chapter 38

evidence for such a state of affairs, which we judge to be highly unlikely I the pattern of results can be taken as presumptive support for the principle under investigation .

38.3 Plural NPs Reconsidered We conclude this chapter by returning to the problematic caseintroduced in the previous chapter: plural NPs. In the study under discussion, children were presented with indirect questionslike the ones in (5). (5) a.

I know how many hamburgersthe octopuseshave. Three. or b. I know how many hamburgersthe octopuseshave. One.

The "puppet, Kermit the Frog, expressedboth the question and its answer. Kermit gave the distributive answer, "One," or the collective answer, "Three," on different trials. In the context, there were three octopuses, and each octopus was holding one hamburger. Therefore, Kermit's question/answer pair was correct on either trial. However, many children acceptedonly the distributive answer, "One," and rejected the collective answer. Are there control sentencesthat could be presented to children in order to tell whether or not the collective interpretation of plural NPs is available to them? One possibility was suggestedto us by Sus"an Carey and Gavin Huntley. They pointed out the existenceof purely "collective" NPs like army, family and forest. These collective NPs could easily be incorporated into the experiment. For example, different trials could comparethe plural NP treeswith its unambiguous, collective counterpart forest. (6) (a) I know how many kites got stuck in the trees. Four/One. V5. (b) I know how many kites got stuck in the forest. Four/One. Supposethat the context includes a forest consisting of four trees, each with a kite stuck in it . Comparing children's responsesto the alternativesin (6) would determine whether or not they had the collective interpretation of a plural NP available to them. If children continued to acceptonly a distributive answer, "One," with collective nouns like forest, then it could be inferred that they did not have the collective interpretation availableto them.

Chapter

39

Resolving

At

the

the

end

of

knowledge moto

of

only

one

way

to

discuss to

plural

dilemma : making

further

in

experiment

The

experimental

Ernie

. The

contest

to

Kermit

the

lifting

this

the

Ernie

see

cans

Bird

came

Frog

turned to

out

his

rescue

holding

on will

like

distributive

assign

earlier the

interpretation

this

of

also

24

the

-

) . However

chapter

, one

, we

only

will

permits

ambiguities

to

-

inter

discussed

not

allows

, beyond

us to

changes

children

return

to

a

in pragmatic

between

following

, so

the

ages

of

:

in the

context

started

to to

a variant , Kermit the

adhere

to in

sen call

distributive all

condition

four

them the

the

cans

. At

entire

by

this set

original

of test

, then

they

is Principle of

is exactly

of

ambiguity condition what

,

himself

.

point

, Big

four

cans .

sentence

is pragmatically

collective

. This

condition

interpretation

principle

the

" They

will

distributive

this

sentence

, we

' s description

. This

a

cans .

. Therefore

sentences

context

two

the

drop lift

lifting

test

and in

of

lift

of

ambiguous

interpretation

the

the to

in

Bird

engaged

interpretation

interpretation

discourse

distributive

, using

managed

with

who

succeeded

, Big

characters

a distributive

, but not on condition .

the

them

attempted

he

they

characters

, both

. Following

he

processing

. 1 Children collective

of

two

. First

happened

and

situation

in

parts

experiment

together

matches

( 1 ) involved

two

collective

heavy

reading collective

adults

that

tested like

favors

the

cans ." In this

are

Crain

thought

the

too

the

four

extremely

cans . Each

contest

, and

collective this the

, described

sensitive

had

he

of

be

the call

interpretation

should

to

readings

are

context on

the

described

If children

Success

what , the

win

. In

just

's

Miya

both

al . 1992

maneuver

, but

to

out

odd

to

. This

additional

sentences

more

condition

wanted

are

propriate fore , we

lift

described

sentences

context

by

of

have

children

NPs

cans .

acted

cans ." Here

et

. We

circumstances

corresponding

could

control

the

and

four

was

who

, Kermit

" They

tion

that

distributive

The

the

trial

story

of

trials

of

plural

availability

; Avrutin

most

study

of

Events

Miyamoto

lifting

. The

the

1992

of

children

4 ; 11 ), using

is pragmatically

still

Again

are

four

; it

by age

investigate

on

in

availability

Collective

a dilemma

Miyamoto

means

Context

interpretations

evoked

ambiguous

1 : that

facing

to

changes

in

versus

They

see

, by

the

chapter

6 ; 0 (mean

(1)

tence

for

preferred

left

designed

(also

the

collective

successfully

the

second

are

NPs was

39 . 1 Individual

3 ; 0 and

) was

way

are

were and

resolve

point made context .

A

37 , we

another

that

: Varying

distributive ( 1991

reading

probe

those

the

Crain

pretations

us

chapter

of

and

Dilemma

: ap -

. There

will

-

prefer

Referential resolu and happened

the

-

316

in

Chapter 39

the

experiment

tive

.

condition

both

89

true

In

and

the

ment

They

the

are

was

Kermit

interpretation

the

appropriate

.

younger

time

,

as

bias

to

the

First

say

,

I ' Yes

younger

, "

pretation

to

The

take

utive

interpretation

far

the

39

. 2

should

collective

this

distrib

-

,

because

it

and

the

made

children

'

about

s

which

older

ones

in

speaker

'

.

Finally

s

,

greater

:

ral

the

is

children

%

of

ex

strength

of

true

the

.

Second

,

inter

children

than

-

could

older

of

-

their

collective

younger

discussion

the

possible

utterance

tenacity

further

both

context

84

sev

over

with

the

finding

has

a

access

if

condition

outcome

but

,

interesting

children

,

can

pronouns

distributive

from

likely

Crain

a

'

s

findings

invite

us

operator

( Heim

,

to

( D

Lasnik

in

debt

,

39

. 4

and

children

the

infer

)

issue

that

to

1991

.

would

.

The

children

.

the

This

is

,

the

' s

like

distrib

the

experiment

for

children

,

express

)

Crain

on

sentences

to

operator

May

section

influence

ambiguous

-

and

Miyamoto

considerable

whether

,

but

the

.

not

A

-

point

by

of

A

vrulin

observation

that

decisions

about

which

.

the

for

to

the

.

the

the

real

interpretation

world

)

the

,

are

and

the

is

true

of

(

question

1991

)

sentence

is

superset

of

be

the

-

ambiguous

to

.

Third

,

superset

the

pose

circum

must

relation

a

study

following

the

alter

the

-

child

' s

interpreta

-

circumstances

corre

-

.

.

experience

For

s

the

corresponding

the

in

'

if

subsetjsuperset

meaning

and

.

in

circumstances

a

Crain

arise

sentence

in

interpretation

distributive

-

,

sentence

relationship

,

First

Second

sole

the

alternative

,

Miyamoto

would

.

adults

which

subsetjsuperset

from

child

children

about

according

findings

leamabilityproblern

( for

Fortunately

terpretation

the

This

older

most

owes

against

to

a

and

the

decision

interpretation

do

NPs

to

problem

sponding

in

model

of

children

plural

making

distributive

has

hypothesis

,

.

reported

) ,

interpretations

tion

-

interpretation

their

another

in

Success

plural

assign

ask

adults

initial

produced

differ

Referential

experiment

conspire

for

chil

Considerations

leamability

native

mental

basis

sentence

that

of

with

the

context

stances

s

,

,

to

show

distributive

and

Learnability

We

'

collective

the

study

children

the

covert

1994

to

,

-

.

the

interpretation

age

statement

than

of

of

discourse

between

' s

concerned

Miyamoto

(

the

state

children

separately

the

s

Accordingly

interpretation

contributor

might

degree

a

for

main

all

favor

afield

Thornton

match

the

s

for

less

hypothesize

departure

and

a

being

greater

,

,

%

seems

too

adults

70

Principle

addition

the

by

'

children

possibility

us

In

a

the

last

data

might

to

adhere

of

this

.

'

that

"

on

-

were

Kermit

children

events

sentence

assign

Crain

Kermit

younger

children

the

,

evidence

,

lifting

interpretations

rejected

to

.

,

distributive

compared

planations

However

and

the

5

"

infelicity

to

is

collec

interpretation

cans

.

This

Presumably

two

pragmatic

was

Analyzing

than

.

.

the

sentences

collective

two

interpretation

time

of

children

Miyamoto

and

this

the

value

the

lifted

on

of

in

the

.

of

collective

truth

discourse

assign

findings

false

%

sentences

because

on

character

Success

the

.

the

to

The

the

Despite

true

of

was

test

,

condition

the

for

statement

this

Referential

tendency

model

"

the

surprising

each

represented

.

too

in

70

on

to

not

where

cans

of

decide

some

' s

is

statement

Principle

to

interpretation

s

context

them

mental

four

'

the

there

This

,

lifting

Kermit

utive

.

felicitous

discourse

caused

affirmatively

time

condition

guided

of

the

pragmatically

rejected

were

responded

of

distributive

"

dren

Children

%

collective

If

a

interpretations

child

will

example

initially

provide

,

the

of

hypothesizes

ample

child

NPs

the

evidence

could

plural

encounter

are

not

distributive

for

the

a

in

alternative

sentence

-

,

like

Varying the Context " Gennaro

and David

have 50 cents ," in a context

317

in which , say , Gennaro has 10 cents

and David has 40 cents. Since the child was expecting the total to be $1.00/ it will be

clear that the distributive interpretation yields the wrong results for such sentence/ meaning pairs. Similarly, if the child initially hypothesizes the collective interpretation, there will be ample evidencefor the alternative, distributive interpretation. For example, the child could encounterthe sentence"Gennaro and David have 50 cents" in a context

in which

Gennaro

and David

each have 50 cents .

39.3 PrincipleB Reconsidered This chapter concludeswith another example of the researchstrategy for studying children's grammars, according to which the context is modified while the structure of the test sentencesis held constant. The example comes from a study by A vrutin and Thornton (1994) designed to test children's knowledge of a covert quantificational operator associatedwith plural NPs. This study is based on the observation made by Wexler and Chien (1985) and

many others that someyoung children incorrectly acceptsentence(2) as a description of a story in which Papa Bear covers himself (with , say, a blanket ), but does not cover

another male characterwho is salient in the discoursecontext (and similarly for other, analogoussentences ). (2) PapaBear covered him. Recall that the same children who accept (2) reject (3) in the same context , that is,

when PapaBear covers himself (Thornton 1990; also seeAvrutin and Wexler 1992). (3)

I know who covered him . Papa,Bear.

Evidently , children who allow an illicit interpretation of certain sentences with refer ring expressions, like Papa Bear, adhere to the relevant constraint on interpretation

(i.e., Principle B of the binding theory) when a quantificational NP and a pronoun are related , as in (3), that is, when the pronoun is bound by the NP . In other words ,

children distinguish between bound and referential pronouns, in that they fail to observe an adultlike restriction on the interpretation of referential pronouns . It is worth

remarking that many languagesmake an overt distinction between referential pronouns and bound pronouns. In Spanish, for example, bound p~onouns are no! phonetically realizedin certain syntactic contexts, whereaspronouns are overt if they are used referentially. Some children learning English have incorporated this distinction into their grammar, as well; although their grammarsfollow the natural seamsof the theory of Universal Grammar, they are not the seamsof the target language. Pronouns are not the only linguistic elementsthat can be either overt or covert. Operators share this chameleon-like quality. There were overt operators in the sentences children were asked to respond to in Wexler and Chien's studies (e.g., Every bearcoveredhim). In the study by A vrutin and Thornton, children respondedto sentences with conjoined NPs (e.g., The Smur! and the Troll coveredhim), which are asso-

ciated with a covert operator in certain cases , according to Heim, Lasnik, and May' s (1991) account. A covert operator is attached to a conjoined NP when it is being assigneda distributive interpretation. A distributive interpretation, in turn, is appropriate in a context where the membersof the group being referred to by the pronoun are analyzed individually . For example, consider the conjoined NP the Smur! and the

318

Chapter 39

Troll in (4), and imagine a context in which two characters , Grover and Big Bird, ask the Smurf and the Troll to cover them. (4)

TheSmurf and the Troll covered them.

It would be appropriate to assignthe distributive interpretation to the conjoined NP, the Smurf and the Troll, in a context in which the Smurf covered Grover and Big Bird with a blanket at one time and the Troll covered them at another time. In sucha context, call it the distributive context, we would expect the covert distributive operator to be attached to the subject NP. The distributive operator would be absent in another context, where the Srourf and the Troll together covered Big Bird and Grover. In this context, call it the collective context, the conjoined NP the Smurfand the Troll is given a collective or group interpretation. To test for the distributive operator in children's linguistic representations , A vrutin and Thornton used sentenceslike (4), presentedin distributive contexts and in collective contexts.2 The two contexts were alike in all essentialrespects. In the story that was presentedto children, the Smurf and the Troll refuse to cover Grover and Big Bird, becausethey need their blankets for themselves. In the collective context, having refused to cover Grover and Big Bird, the Smurf and the Troll cover themselves together with a big blanket. In the distributive context, there were two separate events. In the first event, the Smurf refuses~o cover Grover and Big Bird and covers himself, instead. The secondevent repeatsthis scenario, with the Troll saying he is unable to cover Grover and Big Bird and covering himself instead. Adults' grammar prohibits themfrom referring to the Smurf and the Troll in (4), regardless of whether the conjoined NP is given a collective or a distributive interpretation. But we have seenthat, for some children, it makesa differencewhether an NP contains an operator or not. The children who accept an anaphoric relation between a pronoun and a referring expression (as in (2)), but not between a pronoun and an NP with an operator (as in (3)), are ideal candidatesto test the proposal that conjoined NPs are assigneda covert operator in some instances. If the proposal that an unseendistributive operator attachesto the subjectNP in (4) in a distributive context is correct, then some children should allow coreferencein the collective context, but not in the distributive context. Thesechildren should prohibit themfrom referring to the Smurf and the Troll in the distributive context, owing to the presenceof the unseenoperator. This is exactly what A vrutin and Thornton found. Twelve out of 33 children were identified as relevant subjects. That is, these children were sensitive to the presenceof an operator, distinguishing sentenceslike (2) from sentenceslike (3), and they demonstratedknowledge of the collective and distributive distinction in a separatecontrol condition.3 These 12 children acceptedthe collective context of (4) 93% of the time, but the distributive context only 27% of the time.4 A vrutin and Thornton's study underscores three points about investigating children's knowledge of linguistic principles. First, the findings demonstratethe utility of studies of child language for evaluating theoretical proposals. Second, the study illustrates children's sensitivity to variations in pragmatic context. With the sentence held constant, children assigneddifferent linguistic analysesto the conjoined NPs depending on featuresof the discoursecontext. Third, the findings illustrate a linguistic distinction that is hidden in adult English but overtly manifested in many of the world 's languages. Therefore, the findings represent another challengefor the Input Matching Model.

Chapter40 Conclusion

Recentadvancesin linguistic theory have led investigators to hypothesize that more linguistic knowledge is innately specifiedthan was previously thought. Much of this knowledge is seento be encodedin principles and parameters. Innate linguistic principles, in turn, assistin the formation of a variety of linguistic constructions, certain properties of which must be learned from experience (see Crain , McKee , and Emiliani

1990; Crain and Fodor 1993). If not all aspectsof a linguistic construction are innately given, we might expect some delay in its acquisition. However, we should expect it to emergejust as soon as the relevant learning has taken place. Unfortunately, a great many researchfindings do not comport well with the expectation that linguistic knowledge should emerge early in development. Worse still, apparentviolations of putatively innate linguistic principles have occasionallybeen reported (e.g., Jakubowicz 1984; Lust 1981; Matthei

1981, 1982; Phinney 1981; Roeper 1986;

Solan and Roeper 1978; Tavakolian 1978, 1981). In addition, longitudinal studies of children's spontaneousproductions suggest that language develops graduallyI such that many linguistic constructions- even ones that receive special assistancefrom innate principles- are masteredquite late. Slow acquisitionand a high proportion of errors by children in experimentalstudies seemout of step with recent findings in the literature on child languagedevelopment, however. Children have been found to perform almost flawlessly in responseto sentences that require complex syntactic and semantic representations. This has been true especially if the linguistic principles are arguably part of the child's innate linguistic knowledge, that is, if the principles under investigation are linguistic universals. In addition, apparentgaps in children's knowledge of other linguistic phenomena have often proved to be artifacts of inappropriate experimental methodology. On the basis of these findings, the Modularity Matching Model maintains not only that children have accessto Universal Grammar, as adults do, but also that the principles of Universal Grammar have primacy over nonlinguistic factors for children, just as they do for adults (Fodor 1983; Crain and Steedman 1985). Owing to the modular

architectureof the languageapparatus,grammaticalknowledge preemptsnonlinguistic factors. Nonlinguistic factors do not compete with linguistic knowledge pursuant to linguistic behavior. In fact, nonlinguistic factors exert little influence on behavior. Their role is limited to circumstancesin which linguistic knowledge fails to yield a well-formed and interpretable analysisof the input. Ordinarily , however, children are expectedto processlinguistic information as well as adults do. High error rates continue to be accepted, however, by researcherswho view linguistic behavior as an aggregateof linguistic and nonlinguistic factors. According to this viewpoint , the Competing Factors Model , the influence of factors extraneous to

grammarmay be enough to makeungrammaticalsentencesacceptablefor children, or

320

Chapter 40

these factors may conspire to make children reject grammatical sentences . In this book, we have questioned this viewpoint and its methodological foundation. We have tried to show that apparentgaps in children's syntactic knowledge are artifacts of experimental design- the introduction of nonsyntactic demandscausedchildren to make errors. This conclusion is supported by evidence that children's errors disappear, or are greatly reduced, when tasksthat minimize the nonlinguistic burdens of languageprocessingare used to assesstheir linguistic knowledge. These observations underscore the need for a considered discussion of research --strategies and empirical tools in child language studies. The researchstrategiesand designsintroduced in this book are offered in the hope that this need can eventually be met, and in the hope that they may eventually bring researchfindings in line with the expectations of linguistic theory. To conclude, we hope . that this book will aid students and researchersin designing new experimental studies. We have tried to offer specific guidelines for researchdesign, introducing researchstrategies that we believe will further advanceunderstandingof children's universal mastery of the syntactic and semanticprinciples of Universal Grammar.

Notes

Chapter

1

1

.

For

2

.

Whether

3

.

a

survey

of

or

will

not

For

an

take

experimental

1988

Chierchia

methods

adults

up

here

introduction

Radford

mend

not

learning

see

second

McDaniel

,

language

McKee

have

,

access

and

to

Cairns

the

1996

LAD

.

is

a

matter

of

debate

that

we

.

to

provides

and

,

a

syntax

a

McConnell

more

within

general

-

Ginet

the

generative

framework

introduction

1990

.

and

Larson

For

an

and

,

we

recommend

introduction

Segal

to

1995

semantic

Haegeman

theory

1994

,

we

recom

;

-

.

Chapter 2 1. There are even caveats to this expectation. For example, it may turn out that not every language exhibits the structural prerequisitesfor the application of the principle in question. 2. Some constraints that meet this criterion are stated in positive terms. When they take this form, they generally indicate what must be the case. For examp~e, Principle A of the binding theory statesthat an anaphor(e.g., a reflexive pronoun) must have a local antecedent. This is logically equivalent to the negative statementthat an anaphor cannot fail to have a local antecedent. Clearly, then, the removal of this constraint from the grammar would result in an increasein sentence / meaning pairs, to include ones in which a reflexive appearswithout a local antecedent. 3. In later chapters, we present evidencethat augmenting one's mental model in this way is quite difficult, both for children and for adults. 4. Although in most casesthe grammaticalknowledge of adults"is not at issue, it is usually "importantto incorporate adult controls in an experiment, to ensurethat the task is tapping linguistic knowledge (see chapter29). Chapter 3 1. The form in which this knowledge is encoded is subject to debate, however. For example, some researchershave argued that constraintsare not statementswithin the theory of grammar, but are part of the architectureof the human parsing apparatus(seeFodor and Crain 1987). 2. Recallthat constraintsneed not be statednegatively to be prohibitions against certain sentence/meaning pairs. Seenote 2, chapter 2. 3. To learn constraints on meaning, rather than on form, children would need negative semanticevidence; that isl they would need to be informed in some way that it is illicit to assign certain interpretations to sentences . 4. The observation about the necessity of an abundant supply of negative evidencealso presentsa challenge to conservativelearning strategiesthat invoke special"cuesI' or "triggers." If any of the necessary ingredients for grammar formation are not abundant in the input, then some learnerswould not encounter them, and theselearnerswould fall short of the target grammar. Chapter 4 1. In parts II and III, the term null hypothesisis used differently, to refer to one set of possible experimental outcomes, namely, outcomesthat are not expectedaccording to the model under investigation. At that point, we will use lowercaseto distinguish that use of the term from its presentuse.

322

Notes to Pages37- 69

Chapter5 1. At least, the LAD is no longer in service in the analysisof adults' first language. There is some suggestive evidencethat the LAD may be availablefor secondlanguagelearning. 2. As expectedon the Continuity Hypothesis, there is a crosslinguisticparallel to the structure in (6) in the Paduan

dialect

of Italian .

3. We are grateful to William Snyder for this idea, and for pointing out its negative empirical consequences .

4. Lacking real-world knowledge, children may also be more consistent in their judgments than adults are

.

Chapter6 1. A mode is a measureof central tendency, like the meanor median. A mode is a peak in the distribution of scores. The mode is a useful description of distributions of scoreswith more than one peak. 2. If the true component in the observationsmade by children is the sameas the true component in the observationsmadeby adults, then the overall distribution will be normal, although the magnitudeof the error will be greater for children; hence, the contribution of the true component to children's linguistic behavior will be less. For example, in a task in which children are askedto judge whether sentencesare true or false, almost no children (or adults ) should indicate that all of the test items are true or all are false .

3. Parametric statistics are used to evaluate quantitative measures of distributions

of observations drawn

from populations that exhibit certain special characteristics , and where the sample size is sufficiently large. Parametric statistics (e.g., the t test, analysis of variance) are contrasted with nonparametric statistics (e.g., the chi-squaretest), which are more appropriate for populations that do not mee.t the assumptionsunderlying the useof parametricstatistics, or where the samplesize is small. 4. Parametricstatistics may be applied to the findings of experimentsadopting the truth value judgment task, provided there are enough subjects and items. This may seem surprising becausethe data from such a task are on a nominal scale. Thesedata can be converted to an interval scale, however, byaver aging acrosstrials. According to the Central Limit Theorem, the resulting data approximate a normal distribution

in the limit .

5. We would add the caveat that there are circumstancesin which children's grammatical knowledge is overruled

or goes unused . We will

discuss several

of these circumstances

. Also , see Fodor

1984 for a re -

lated discussionof adult linguistic judgments. 6. Here we are simply following conventional wisdom, as advocated, for example, by the criterion of attributing knowledge of a grammaticalprocessif it is manifestedat least 90% of the time in obligatory contexts

.

Chapter 7 1. Children's responsesto ambiguous control sentencesalso serve an important function: they are the yardstick against which to compare children's responsesto sentenceswithin the jurisdiction of the constraint

.

2. More.over, when the results of a statistical test are significant, it is inferred that other subjectswho are similar to the ones sampledby the experimenter would also be influenced by the true component of behavior in a similar test of their linguistic knowledge. 3. This expectation may seem counterintuitive

to readers who are familiar with findings from the liter '-"

ature on adult sentenceprocessing, where longer reaction time is usually associatedwith processing difficulty. Chapter

8

1. The other order of main and subordinate clauses was tested for both temporal conjunctions in all of the studies under consideration . 2. Not all adults share the intuition that the event mentioned in the relative clause is conceptually prior to the event mentioned in the main clause. We have queried hundreds of undergraduate and graduate students , however , and most agree with this intuition .

Notes to Pages69- 112 Chapter

9

1 . Moreover We

, it

leave

should

aside

be

the

, experimenters

can

in acting

be used

Chapter

appropriate

pragmatic

2 . Moreover

in

the

context

prerequisites often

out

the

out

the

make

for it

content

clear

of the

for

the

this

statement

that

figures

sentences

bear

to

by

say

the

that

the

turtle

tickled

the

horse

.

bear .

that

are

not

mentioned

in

the

test

sentences

that

are

not

mentioned

in

the

test

sentences

.

10

1 . Moreover

, experimenters

can be used 2 . Of

course

,

also

permits 4 . Of

course

accept

the

sentences

reading

, and

, if

it

can

be

Factors

clear the

, but

that

do

test

was

for

at

rates

is such do

see the

approaching

) context reason that

not

if

this

the

them

makes

the

interpretation

could

such

because

sentences

interpretation

Model

of

reject

reading

Factors

necessity

100 %

was

. ; adults

this

prefer

Competing

not

.

a different

children the

figures

(ambiguous

context

), then Model

that sentences

meaningl in the

the

shown

interpretation

Competing

it of

JIYes /' response

one

adult

Chapter

would

the

make

content

reject

only

(i .e ., the the

children

with

3 . Adults

often

in acting

associated

that

is false

be maintained

a demonstration

the

constraint

false . in

; but

the

context

advocates

of

.

11

1 . Of

course

, the

conditions

statistical

' s judgments

sentences

of

grammatical

guistic

Big

Principle

interpret

the

pronoun

hypo

Philip

_thesize

( 1995

Chapter

by

Bird

C . In

in

show

provided

( I ) as if it meant

3 . Children

coreference

; they

antecedent

violates

using

raw

scores

, not

these

mean

percentages

for

the

two

response

only

the

that

context

ate the apple

that

Big

Bird

with

the

representation

its

like

them

is , the

true

an

not

the

pronoun

lead

the pillow

additional

denotation

( 1 ) do (if

pragmatic

touched

provide

as coreferring nonadult

sentences

find

) . That

when

experiments

to

they

show

that

encouraged . Although

linguistic

they

refers

to

children

such

only

to

true , this

antecedent

find

the

lin -

interpret

sentence

, children

too

always

.

in

addition

to

an

adultlike

representation

, according

).

12

1 . To

evaluate

studies and

this

conception

investigating

those

review

the

weiler

Chapter

the

types

relevant

on

, and

that

the

verbal

kinds

of

unambiguous

lesser

demands

pose

literature

differences 1991

of

both

sentence

memory

working

here . For

processing

Shankweiler

, also

and

memory

on

discussion

and

1986

, we

that

memory

see Bar - Shalom

Crain

system

sentences

have

resources

empirical

, Crain

of

. Space

research

, and

conducted

are . costly

on

Shankweiler

a number

memory

does

not

of

resources permit

the

effects

of

1991

, Crain

, and

,

us

to

working Shank

.

13

1 . Of

course

, using

presupposition avoids

set

a restrictive that

adding

2 . Presumably

in

performed

.

2 . Children

to

323

new

, the

immediately elevated

relative

some

horses

entities

to

( c ) versions before

reading

the

in

to

was

both

this

way

been

model

easier

ambiguity

times

in

already

mental

were

the

clause

had

past

, as dictated

repair

because

encountered

experiments

is infelicitous

raced

the

by they

; the

the

Principle

did

not

region

, as compared

. It requires barn ; but

to

of

require

with

accommodation

accommodating

the

sentences

Parsimony

with

the way

of a contrast

operator the

this

.

construction focus

of in

only

definite

resulted

determiner

the . 3 . The of

alternative prepositional

number

findings

5 . In keeping low

sentences

with

suppose

phrases

of regressions

4 . Similar

with

is to

have the

working containing

that

in by

these

been

relative

- span

subjects

with

only . However

subjects

obtained

Modularity

memory

low

sentences

with Matching

capacity clauses

did .

in the

only - VP

special Model as well

did

not

initially , this

sentences

populations

with

the

fails

NP - attachment

to

explain

the

analysis significant

.

of adults

, Ni , Crain as people

pursue

account

, and high

, such

as Broca

Shankweiler

( 1996

capacity

' s aphasics ) found

in processing

. that

people

unambiguous

-

324

Notesto Pages 112- 125

C hap

1.

fer

Any

14

model

of

formation that

of

and

the

, are

by about does

3 .

is

the

affairs

we

of

affairs

This

argument

Even is

if

not

to

sentences

to

given

ering

hypotheses

important

to

being

made

a large not

bottom to

be

a

. Roughly

the

interpretation

which

Chapter

1.

Children

In

may

assign

of

modularity may

times

be

an

maximal

set

the

-

options

of

langu

"age

the

state

options

, with

. That

is , we

.

the

. learners

that

shall

cannot

be

task explain

the

adult

this

why

refer

of

the

case

. In

and

options

these

values

we

are

options

that .

number

made

at

consid

-

semantic

evidence the

being

sentence of

formulating

positive

set are

sentence

to

adults

in to

the

subjects

of

) interpretive basis

, the

a range

certain

ones

a contrast

, even

hearers

, we

have

and

old

the

of

a sentence

is

of

. The

then

commit

-

consequence

.

verification

now

setting

( universal on

falsifiability

forces

for may only

commitments

judge

. For

of

size

more

necessarily

. We

available

supplant

) options

set , the

for

short

be

would

the

chapters

regard

reader

on

the to

a

description

"

" test

judge

the

" extra

the

true

probably

"

of the

truth

value

characters

as

mentioned

in

the

sentence judgment

relevant

to

section

14 . 1 ,

.

a

tendency

to

strong

are in

omission

tolerate

the

in

. If

errors

view

appear

as

it

is , the

" main

effectsl

. It

have

by

their

of

errors

"

but

long

and

from

the

be

may

the on

and

no

assumption .

will be

kind

some

similar that

in

adults

adults

to

- by

- construction

I ' group

.

attempt

adults

should same

children

should

complex

responses adults

to

too

and

effect

be

by

or

sentence

follows

would

there

representation

too

incorrect

children

responses

is

a similar

proportion

different

it the

context will

made

grammatical

interpreting

errors

, but

, the

in

any because

discourse of

response

more of

the

sources

incorrect

make

assign

sentence

factors

true

incorrect

point

of

use

deletion

according

to

is

interesting

mance

errors

optional

form

also

underlies

2

was

wanna

and

where null

cases

on . That

the

preference

are

not

of

in

in

the

, when

I ' wanna

various

forms

preference which

speech

for

linguistic of

want of

contraction

contraction

contracted the

, contraction

occur

contract

as

producing

sentences

obligatory

should

," proves

reduced

and

to

to

adults

to

be

forms

principles

either

is where

do form or

is not

wanna children

, ,

.

possibility is , even

for

, in

to

known

consequence

. However

Model

' s tendency

phenomenon

. Another

complementizers

Matching

speculate

people

the

. This

adults

complementizer

Modularity

some

"

chapter

complementizers

or

to , in

to children

a

forms

from

contracted both

of the

" reduced

example

contraction

the

be

would

should

to . An

and

quite

it these

cannot the

."

permissible want

to

they

nonlinguistic

makes

, but

statistical

reason comprehend

, that

out

, children

constructions

interactions

the

that

children

some

invoke

, however turn

. From

for cannot

well

interpretation

inflated

. In

if they

may

ultimately

make

The

errors

because

matching

linguistic

It

strategy commit

confuse

interpretations

, children

parameter

a limited

between

picture

of

to

about

hypothesize

values

- particular

picture the

sentence

make

, children

it

This

also

4 .

this

that

imagination

also

cases

to

or

the

, perhaps

such

nature

3 .

the

is

available

to

of other

15

a sentence

2 .

a picture

not

setting

between

parameter

contrast

. In

sentences

options

parameter nevertheless

with

is

of

idea

in

may

correlation

would

grammatical

the

particular

, not

writer

the

. Some

some

ambiguous

a

additional

language in

competing

is

,

with

transfer of

Parsimony entail

therefore

regularity

interpretive

, new

people

picture

evidence

of

target

or

that

in -

line

the

with

that

appear

importanl

among

sufficient

difference

the

is

the

not

structurally

. As

begin

in

may

a hearerl

facilitate

model

new

in

.

( language the

that

of on

important

larger

description

relies

they

commitments

say

, the

, but

, children

: the

. It

alternative child

derived

selects

language

the

.

by

parser

and

Principle

discourse

the

. To

mental

the

disconnrmed of

them

of

example

semantic

speaker

sentences

, but

with

setting

observe

- hand

false

task

an

additional

of

to

right

is

parameter

number

is

theory

contrast

is

of

the

include

ments

This

for

. In

to

target makes

in

, by

extended

the

Grammar

. There

, however

negative

of

are

processing

a

ambiguous

theory

child

in

informed

ambiguous them

time

that not

in

Universal

the

adults

are

, for

model

of

their

is

the

between

align

domain

the

existence

bring

models

to

options

in

whose to

guidelines

the

both

, where

presupposes

information

of

within

settings

confronts

mental

of

the

, that

describing

order

attempt

available

parameter

that

necessarily

any

6 .

are

learners

available

S . It

with

events

interpretations

events

be

or 1 in the

flow

Following

avoid

however may

that

assigned

.

to and

entail

case

state

assume

4 .

is

the

continuously

discourse

Grammar

between with

must

parser

not

individuals appropriately

interfere

individuals

Universal

of

to

the

the

contains

modified

. Mismatches

apt

in

adopted

Such

be

perceivers

participants

This

to

interlocutor

a reader

ments

that

have

informationl

2 .

discourse

will

that when

the

the

parser

grammar

' s preferences of

a particular

may

cause

language

genuine does

perfor not

permit

-

Notesto Pages 125- 177

325

one to use a certain reducedform, the languageuser could, in times of stress, respond in a way that is consistent with the grammar of another language. In fact, evidencefor this comesfrom an experiment on elicitation of bound pronouns (Thornton and Crain 1989). In the first few trials of the experiment, when children were trying to accessthe structuremandatedby the experimentalsituation, somechildren produced null pronouns, asking questionslike "Who thinks_ is the skinniest?" insteadof "Who thinks he is the skinniest?" The result is not a parsing preference, since the responseis not a viable option in English; "Avoid Pronoun" is not a grammaticaloption. However, this performanceerror could, in some sense,be a reflection of the parser's inner workings. Chapter

16

1. For researchers testing children 's acquisition of American Sign Language (ASL )[ however , use of puppets is problematic , because it is very difficult to make a puppet sign . Yet , for all the reasons presented , it remains desirable to have a puppet interact with the child . To counteract this problem , Lillo -Martin and her colleagues have one of the experimenters (the one who would play the role of the puppet i~ tests using a spoken language ) actually dress up as the character , wear face paint , and so on . This modification has been very successful. Presumably , children do not envisage the dressed-up character as an adult with this modification of the task. 2 . If it is feasible, the sessions can be videotaped . In general , we do not videotape every experimental session. We do try to videotape several sessions of each experiment , however , to document the experi ment and for use in classes and at conferences .

Chapter

17

1. In cases where the construction of interest occurs frequently , presumably transcripts of children ' s spon taneous productions can yield sufficient data for analysis . 2. Some of the alternative explanations of negative findings apply more forcefully to comprehension tasks and to the Competing Factors Model . In studies of elicited production , children 's productions can be used to make inferences about their intended messages. Moreover , on the Modularity Matching Model , statistical power is not an issue, because the experimental hypothesis is the total absence of certain forms of behavior by all subjects, both children and adults . This leaves, as alternatives to the experi mental hypothesis , factors such as parsing strategies and memory limitations ;' these will therefore be the focus of the rest of our remarks .

Chapter

18

1. On the account proposed by McDaniel , Cairns , and Hsu (1991 ), children have a different processing system from adults , however . They suggest that children generate a coordinate structure as a default structure , because the appropriate control structure is too complex to process. 2. This study elicited both subject and object gap relative clauses.

Chapter

20

1. The machinery we adopt is one notation that is employed by linguists to indicate the pertinent structural relationships in the examples under consideration . It is these structural relationships that are important , and not the notation itself . 2. It was important that the experimenter point out the relevant characteristics of the picture to the child so that the child did not focus on details irrelevant to the experiment . 3. This example is introduced for purposes of illustration . In practice , we avoid experimental items that suggest violence .

Chapter

21

1. More specifically , contraction is prohibited across a wh -trace that is Case-marked (Chomsky 1980 ; Jaeggli 1980 ). An alternative explanation of the facts offered by Snyder and Rothstein (1992) is that a null , Case-assigning complementizer is responsible for blocking the contraction . The details of the syntactic analysis are not important for the discussion in this chapter .

326

Notes to Pages 177- 196

2. The constraint is most easily studied in dialects of American English, where wanna contraction is widespread. In British and other varieties of English, the difference between the contracted and noncontractedforms may not be so obvious. The point holds, however. Questions extracting from subject position should not exhibit contraction. Chapter 22 1. By saying that the wh-phraseis in the intermediateCP, we intend to remain neutral at this point about the position of the medial-wh in the phrasestructure. 2. The term medial-wh is not used here in the sameway it is used in de Villiers, Roeper, and Vainikka 1990 and subsequentpapers. In those works, medial-wh refers to the wh-phrase in the embedded SpecCPin questionswith two wh-phrasessuchas Whendid CookieMonster know what to bake?We use medial-wh only to refer to children's long-distancewh-questionsthat contain a copy of the wh-phrasein the embeddedCP. 3. In the Minimalist Program, being developedby Chomsky since 1992, the ECP no longer has a role to play. Many of the facts previously explainedby the ECP are now explainedby JIshortest move." 4. As we explain later, there were in fact children who always used the complementizer. We argue that this is a consequenceof their grammar. Children override the parsing preferenceto reduce forms in order to generatesyntactic representationsthat are in keeping with their current grammar. 5. Strictly speaking, since want is an exceptional-Case-marking verb, questions from the wanna contraction paradigm cannot shedlight on the status of the medial-wh in all infinitival clauses. 6. The obvious question is why there should be a preferenceto insert a complementizerin factive structures, when there is a preferenceto deletecomplementizersin declarativesand wh-questions. The answer is not clear, but the Modularity Matching Model would predict that it arisesfor grammaticalreasons. Factivesare generally thought to contain a null operator in SpecCP(or someother projection) (Melvold 1991; Watanabe 1993). It may be that since the operator is null, there is some requirement that the headbe overt (for this idea, seeSpeas1994). 7. The column labeledJIPartial " in table 22.1 indicatespartial movement questions. In addition to the data summarizedin table 22.1, another set of long-distancequestionswas elicited from eachchild. In these questions, the adverb really was inserted into the lead-in (e.g., 'IAsk the snail what he thinks really . . ."), with the hope that this would elicit more complementizers(seeThornton 1990). The questionsdid not differ significantly in form from those reported in table 22.1. 8. One standard view is that in subject extraction questions, the trace in the intermediate SpecCPis a proper governor for the tracein subjectposition of the embeddedclauses.If a complementizeris present, however, antecedentgovernment by the intermediatetrace is blocked (see~asnik and Uriagereka1988; Haegeman1994). 9. There are also other possible scenarios.Children do not necessarilyhave to passthrough the outlined stagesin turn. If children are sensitiveto the input in the form of adult questions, they will realize that in addition to being necessaryonly for subject extraction questions, spec-head agreementdoes not need to be expressedovertly . If they make both of these observationssimultaneously, they will not pass through the second stage, in which they think English spec-head agreementis like the French version in being realizedovertly . 10. It may be that the verb want takes a CP complement. SeeSnyder and Rothstein 1992 for arguments supporting this view. 11. There are certain syntactic environments in which that-trace effects appear to be suspended - for example, casesinvolving a topicalized adverb (Culicover 1991; Rizzi 1996). This is illustrated in (i) and (ii) (examplesfrom JulienMusolino). (i) (ii)

"'Which amendmentdo you think that t will becomelaw? Which amendmentdo you think that next year will becomelaw?

12. In the dialectsof Germanand Romanianthat allow: a copy of the wh-phrasein the intermediateCP, the structure is apparently optional. If children were to start out thinking that the medial-wh was optional, however, they would be faced with a leamability problem. What evidencecould inform them that the medial-wh is not permissible? All of the positive evidencewould be consistentwith their hypothesized grammar. 13. We say "SpecCP" here, becausewe include partial movement questions as a form expressing overt spec-headagreement. But unlike medial-wh questionsand questionswith a that complementizer, partial

Notes to Pages196- 226 movement to

14

.

questions

filled

CPs

These

figures

tions

and

Chapter

1 .

are

These

determine

of

1 , 500

are

One

In

the

)

of

world

my

question

in

SpecCP

.

For

this

reason

,

we

sometimes

refer

types

and

do

not

differentiate

between

medial

- wh

ques

-

.

in

children

' s

represent

a

from

found

14

only

D

- linked

spontaneous

productions

grammatical

,

phenomenon

children

eight

in

the

examples

not

.

CHILDES

of

but

For

not

database

questions

like

with

example

sufficient

,

in

an

( MacWhinney

those

in

( 3b

frequency

extensive and

)

and

( 4b

) ,

search Snow

with

1985 two

),

auxil

-

to

) .

sentences

not

signal

spec

movement

is

Notice

their

incompatibility

that

Who

the

with hell

broke

certain

my

modifications

window

?

is

,

fine

,

but

such

as

Which

boy

the

hell

the

hell

.

declarative

does

that

- phrases

1987

? is

' s

Stepwise

wh

( Pesetsky

window

children

,

signal - head

is

such

that

as

spec

agreement

,

necessary

so

" I

think

- head as

that

the

agreement

mentioned

that

in

there

Spaceman

has

is

a

likes

taken

place

chapter

22

proper

beans

,"

. However

the

,

presence

of

individual

a

children

.

head

governor

for

the

trace

in

subject

.

25

There

are

modifications

example

,

if

an

ambiguous

to

control

owing for

Caution lack

a

obey

1993

a

be

or

linguistic

of

B

Wexler

1990

Chapter

,

,

would

not

.

from of

a

overcome

this

have this

to .

task

(a )

,

it

limitation

that

are

infer

that same

of

experiments

has

in

principle

pronouns

in

The

results

example

a

,

at

least

for acting

they

in

one out

were

general

some

cases

. For

interpretation a

of

sentence

overriding

strategy

their will

an

that

is pref

serve

as

the

.

the

pragmatic

deficiency

preference

interpretation

interpretations

For

a

reasonable

production

either

,

be

interpreting

do

namely

can

children

on

in

they

knowledge

expressions

it

elicited

however

results

of

,

the

that

that

avoid

constraint

when

lack

task

consistently

grammar

,

constraint

(b )

- out

establish

but

linguistic

used

Principle )

,

adult

variants

must

dren

the

to

foundation

act

can

sentence

in

erence

the

experimenter

unambiguous

to

attested

utterances

diagnostic in

Chapter

been

.

argued

processing

It

may

that

capacity

needed

to

anaphorically

look

children

as ' s

( Grodzinsky

apply

Principle

linked

to

though

chil

apparent

and

B

Reinhart

properly

referential

to

NPs

-

failure

certain

( Chien

and

) .

26

This

statement

that

an

sons

of

for are

know

, not

Indeed

.

the

to

the a

clauses

.

evidence

other ,

that

in

trials

all

more

general

all

the

of

with the

more

formulation any

relevant

the NP

will

formulation

of

one

.

accepts

1990 ,

) .

the

Other

cases

of testing

including

ate

the

backward sentences hamburger

claim

that

studies restrictive

principle ) .

that

There

involve

are

C

but

states two

pronouns

Principle

anaphora

when

children

have

similar clauses

pronominals

. as

rea

Second

far

as

is

and not

a

effects

limited

verbal

of

temporal by

itself

working adverbial

convincing

. " He the

washed Smurf

1. Other possiblebackground/assertionpairs are given in (i)- (iii ). Background: Robocop thinks the Troll is such-and-such. Assertion : the best jumper

~

revealed

relative of

like

youn

have

Chapter27 (i)

of

c - commanding examples

general

interpretation

trials " He

,

if

al

constructions

in testing

,

et

extrasentennal

disallow

participated

First

the

coindexed

(6 ) .

( Crain

' s

of not

to

expected -

linguistic

they

children

subcase ( i .e .,

(6 ) .

in is

children

a

counter

formulation

capacity

on

participated

run

preference -

Thus

is free in

that

memory

memory

be

statement

languages

such

C

must

using

there

Fifty

Principle

r - expression

working

3 .

- phrase .

.

position

2 .

across

they

negative

verbs

used

1 .

wh

positions

questions

whether

complementizer

2 .

calculated

( 1990

broke

1 .

extra

Camp

movement

structures

to

or

4 .

the

filled

23

iary

3 .

contain

than

partial

Stromswold

2 .

may

rather

327

Luke was

Skywalker inside

the

," fence

and ."

32

children

,

we

328

It

Notesto Pages 226- 262 (ii)

Background: Robocop thinks such-and-such. Assertion: the Troll is the best jumper

(iii)

Background: Robocop doesso-and-so. Assertion: thinks the Troll is the best jumper

is

not

necessarily

different

pair

in

( i )

jumping

the

were

-

for

.

28

. 2

Wh

Our

move

an

yet

another

way

happened

position

"

known

it

.

to

as

moves

.

but

the

story

example

Troll

the

is

story

Chapter

would

,

have

the

best

it

33

if

at

be

/

something

should

a

besides

possible

reasons

quite

assertion

else

be

discusses

to

background

outcome

for

selecting

one

Recall

that

make

there

the

should

test

be

sentence

a

false

record

.

at

There

the

is

conclusion

no

of

such

record

in

SpecCP

,

the

specifier

of

1990

and

the

CP

projection

.

The

subject

is

.

Yes

"

.

The

in

the

a

.

He

,

pronounj

that

)

was

actually

their

The

the

conducted

experiment

we

check

at

on

say

did

' 1

not

be

out

children

this

strong

strong

he

Since

to

with

said

has

,

we

set

access

the

equated

crossover

.

children

have

is

who

possible

expect

,

know

on

also

that

laid

constraint

would

Yogi

of

,

,

appears

-

to

however

,

since

the

-

the

best

sentence

up

the

a

the

the

correct

multiple

refer

-

of

7

question

structures

crossover

.

serves

thumb

be

we

many

subject

-

way

construction

-

allow

remaining

sessions

.

Unfortunately

his

or

knowing

.

a

10

%

one

they

did

true

in

the

.

a

assertion

have

,

is

that

happened

not

true

This

discourse

is

in

the

worked

here

configuration

is

because

will

to

want

show

,

*

however

,

Iwhi

( i . e

these

of

in

enough

.

. ,

.

.

/ lHe

children

the

cases

on

battery

here

where

children

.

only

For

this

a

.

In

some

reason

,

.

,

and

1996

in

.

such

the

We

circumstance

be

/ lYes

"

.

in

generally

use

designs

should

the

tested

part

test

children

Hsu

we

crossover

we

control

margin

which

within

each

- subjects

.

The

availability

responses

of

to

be

certain

well

read

above

10

-

%

,

.

bound

testing

not

As

figure

would

the

be

what

would

discourse

designs

and

the

error

you

crossover

the

more

own

in

including

,

her

We

participated

,

.

Hsu

what

specific

ourselves

children

other

as

of

,

must

embedded

to

tell

format

was

critical

- subjects

1996

is

relevant

me

the

only

with

between

McKee

' not

strong

This

becomes

again

versus

,

.

subjects

it

it

"

) .

report

20

,

is

Let

This

the

assume

food

. "

"

in

usual

We

test

analysis

experiment

1996

"

.

,

. "

test

embedded

food

.

.

food

to

than

to

data

subjects

Gordon

best

binding

aim

the

best

is

best

who

like

the

and

the

know

needed

subjects

to

the

within

rule

the

"}

sentence

have

something

that

have

would

conduct

each

no

we

- in

have

assertion

test

Bear

said

was

fewer

in

see

because

is

the

Bear

included

discussion

lead

Bear

related

to

appropriate

Yogi

tested

,

The

the

Yogi

have

question

structures

be

.

could

and

we

negate

and

pair

Frog

of

,

planning

in

would

have

,

simply

answer

negation

and

can

of

not

to

design

Frog

answer

however

want

experimental

the

do

Grover

indirect

circumstances

designs

,

cannot

Grover

an

of

children

There

might

the

referent

said

/

that

battery

are

In

Kermit

response

we

the

Grover

normal

,

question

experiment

large

pated

1991

conducting

experimenter

who

multiple

"

/ lHe

because

] .

this

the

.

children

a

Yes

case

Kermit

said

however

said

"

this

of

.

,

questions

For

and

assertion

made

Five

Crain

independently

.

the

ings

were

playing

. "

,

as

that

story

crossover

Turtle

interpretation

For

Thornton

McKee

experimenter

ambiguous

response

Notice

is

The

Ninja

be

we

and

control

of

response

.

In

in

McDaniel

experimental

consisting

.

in

jumper

) ,

the

point

best

For

that

some

the

( iii

.

.

previously

time

additional

being

7

a

before

same

because

.

in

to

SpecIP

interpretation

ent

6

think

at

( i ) -

.

As

adult

.

is

others

really

( reported

the

food

5

,

Troll

defective

experiment

would

.

would

before

in

chapter

30

"

4

the

pairs

this

.

in

adult

.

as

over

Ilwhat

- phrases

crossover

3

the

in

29

about

.

of

Robocop

,

that

indicating

Chapter

2

one

developed

then

But

pair

is

trial

positioned

.

think

experiment

Chapter

1

,

.

assertion

choose

have

28

The

figure

.

we

cooking

to

/

the

1

,

Robocop

Chapter

to

one

selected

say

background

1

wrong

from

variable

children

complete

session

' s

the

.

knowledge

bound

The

other

of

variable

2

children

crossover

control

partici

in

for

one

two

- clause

-

clause

-

Notes to Pages262- 318

329

Chapter 31 1. We have chosento use the term anaphoricallylinkedrather than coreferential here becausewe will be discussingcasesin which the preceding NP contains a quantifier, such as everybearor no bear. Sincethese quantificational NPs are not referring expressions, it seemsinappropriate to label the relationship between theseexpressionsand a pronoun one of "coreference." 2. This finding was madeby Chien and Wexler (1990) for sentenceslike 'Everybearis washingher, in which the NP everybearis a quantificationaINP. (Also see A vrutin and Wexler 1992.) The finding has been confirmed by other researchers , and by Thornton (1990) for the operator who.

Chapter32 1. The force of the argument is blunted , however , by the fact that Chien and Wexler (1990 ) controlled for intonation

(p . 234 , En. 4 ).

Chapter35 1. Becausethe universal quantifier is downward entailing on the nominal constituent it combineswith, the disjunction of nominals , farmer and donkey, in the restrictor clause entails that a conjunction of events must satisfy the conditions stated in the nuclear scope. Any eveI)t in which there is a farmer and any

event in which there is a donkey must be an event such that a farmer is feeding a donkey.

Chapter36 1. As observed by Hamburger and Crain (1982), children find restrictive relative clausesto be most felicitous when they are used to restrict from a larger set. Therefore, eachtrial included an extra character(a boy , in this case) who did not have the object mentioned in the relative clause. This extra character was also included

on the trials

in which

the test sentence

was a conditional

.

2. The test materialswere divided along another dimension, which hinged on the nature of possession .

ChlJpter39 1. If children lacked this Principle of ReferentialSuccess , then their responsewould be dictated by the two remaining factors in the resolution of ambiguities: (a) the preferencefor one interpretation over another; and (b) the bias to say "Yes " (i .e., accept either interpretation

when it makes a true statement about the

discourse context). Assuming that both interpretations are equally available to children, they should give "Yes" responsesto test sentenceslike J'They are lifting four cans" in both the distributive and collective

conditions

. This is not what

occurred , however

.

2. Children heard exactly the samesentencepresentedtwice, once in a distributive context and once in a collective context . So that the game would be more interesting for the children , however , the stories used to test the distributive and collective contexts did not use exactly the same characters. Never theless, the stories were identical in all important respects. We have used the same characters in the text just for purposes of exposition .

3. In the control condition, children were presentedwith contexts that were ambiguousbetween a collective and a distributive

interpretation , but heard a test sentence appropriate for only one of these inter -

pretations. For example, two turtles might be set out in the workspace, eachwith two pet bugs in front of it . Then Kermit might say, "{ know how many bugs they have. Four." This particular utterancewould test whether or not children could accept a collective interpretation of the context . To test the distributive

context, on other trials Kermit would say, "I know how many bugs they have. Two." 4. One might ask why these children acceptedthe distributive context as much as 27% of the time. Presumably, this happenedbecausethe context is responsiblefor generating the syntactic representation that contains a distributive

operator . It may have been that in some of the stories , the context was not

sufficientto causechildren to generatethe operator. If so, it is highly unlikely that children's acceptances of (4) in the distributive context representviolations of Principle B.

References

Altmann, G., and Steedman , M . (1988). Interaction with context during human sentenceprocessing. Cognition , 30 , 191 - 238 .

Amidon, A ., and Carey, P. (1972). Why five-year-olds cannot understand "before" and "after." Journalof Verbal Learning and Verbal Behavior, 11, 417 - 423 .

Atkinson, M . (1996). Now, hang on a minute: Some reflections on emerging orthodoxies. In H. Clahsen (Ed.), Generative perspectiveson language acquisition. Amsterdam : John Benjamins .

Avrutin , S. (1994). Psycholinguisticinvestigationsin the theory of reference . Doctoral dissertation, MIT , Cambridge, MA . Avrutin , 5., Crain, S., Miyamoto , Y., and Wexler, K. (1992). Who knows everythingabout what everyone knows? (A study of quantificationin childgrammar). Paperpresentedat the Workshop on the Acquisition of WH , University

of Massachusetts , Amherst .

Avrutin , 5., and Philip, W. (1994). Quantificationin agrammaticaphasia . Paperpresentedat the Annual Meeting of the Linguistic Society of America , Boston .

Avrutin , 5., and Thornton, R. (1994). Distributivity and binding in child grammar. LinguisticInquiry, 25, 265 - 271 .

Avrutin , 5., and Wexler, K. (1992). Development of Principle B in Russian: Coindexation at LF and coreference. Language Acquisition , 2, 259 - 306 .

Bar-Shalom, E., Crain, S., and Shankweiler, D. (1991). A comparisonof comprehensionand production abilities of good and poor readers. Applied Psycholinguistics , 14, 197- 227. Bellugi, U. (1967). Theacquisitionof negation . Doctoral dissertation, Harvard University, Cambridge, MA . Bellugi, U. (1971). Simplification in children's language. In R. Huxley and E. Ingram (Eds.), Language acquisition: Models and methodology . New York: Academic Press. Berko, J. (1958). The child's learning of English morphology. Word, 14, 150- 177. Bloom , P. (1990 ). Subjectless sentences in child grammar . Linguistic Inquiry , 21, 491 - 504 .

Bloom, P. (1993). Grammatical continuity in language development: The case of subjectlesssentences . Linguistic Inquiry , 24, 721- 734.

Borer, H., and Wexler, K. (1987). The maturation of syntax. In T. Roeper and E. Williams (Eds.), Parameter setting. Dordrecht : Reidel .

Borer, H., and Wexler, K. (1992). Bi-unique relations and the maturation of grammaticalprinciples. Natural Language & Linguistic Theory, 10, 147- 190 .

Bowerman, M . F. (1987). Commentary: Mechanisms of language acquisition. In B. MacWhinney (Ed.), Mechanismsof languageacquisition . New York: Springer-Verlag. Bowerman, M . F. (1988). The "no negative evidence" problem: How do children avoid constructing an overly generalgrammar? In J. A . Hawkins (Ed.), Explaininglanguageuniversals . Oxford : Blackwell. Braine, M . (1971 ). On two types of models of the internalization ontogenesisof grammar . New York : Academic Press.

of grammar . In D . Slobin (Ed.), The

Brown, R., and Hanlon, C. (1970). Derivational complexity and order of acquisition in child speech. In J. Hayes (Ed.), Cognitionand thedevelopment of language . New York: Wiley . Cazden, C. (1972). Child languageand education . New York: Holt, Rinehart and Winston. Chien, Y.-C., and Wexler, K. (1990). Children's knowledge of locality conditions in binding as evidencefor the modularity of syntax and pragmatics. Language Acquisition, 1, 225- 295. Chierchia, G. (1992). Anaphora and dynamic binding. Linguisticsand Philosophy , 15, 111- 183. Chierchia, G. (1995). Dynamicsof meaning : Anaphora, presupposition and syntactictheory. Chicago: University of Chicago Press. Chierchia, G., and McConnell-Ginet, S. (1990). Meaning andgrammar. Cambridge, MA : MIT Press.

332

References

Chomsky , C. (1969). Theacquisition ofsyntaxin children from5 to 10. Cambridge , MA: MIT Press . Chomsky , N. (1965). Aspects of thetheoryof syntax . Cambridge , MA: MIT Press . Chomsky , N. (1971). Problems of knowledge andfreedom . New York: Pantheon . Chomsky , N. (1975). Thelogicalstructure of linguistictheory . New York: Plenum . Chomsky , N. (1980). Rulesandrepresentations . NewYork: ColumbiaUniversityPress . Chomsky , N. (1981). Lectures ongovernment andbinding . Dordrecht : Foris. Chomsky , N. (1986). Knowledge of language : Its nature , origin,anduse . NewYork: Praeger . Chomsky , N., andMiller, G. (1963). Introductionto theformalanalysisof naturallanguages . In R. D. Luce, R. Bush,andE. Galanter(Eds.), Handbook of mathematical psychology (Vol. 2). NewYork: JohnWiley. Chung,S. (1994). Wh-agreement and"referentiality " in Chamorro . Linguistic Inquiry,25, 1- 45. Chung, S., andMcCloskey , J. (1987). Government , barriers , and smallclausesin ModernIrish. Linguistic Inquiry, 18, 173- 237. Cinque , G. (1990). TypesofA-dependencies . Cambridge , MA: MIT Press . Clark, E. V. (1971). On theacquisitionof themeaningof "before" and"after." Journalof VerbalLearning and VerbalBehavior , 10, 266- 275. Comorovski , I. (1989). Discourse -linking and the Wh-IslandConstraint . In J. Carterand R.-M. Dechaine (Eds.), Proceedings ofNELS19. Amherst:Universityof Massachusetts , GLSA. Conway,L. (1997). Excavating semantics . Doctoraldissertation , Universityof Connecticut , Storrs. Conway, L., andCrain, S. (1995a ). Donkeyanaphora in child grammar . In J. N. Beckman (Ed.), NELS25. Amherst:Universityof Massachusetts , GLSA. Conway,L., andCrain, S. (199Sb ). Dynamicacquisition . In D. MaclaughlinandS. McEwen(Eds.), Proceed ingsof the 19thAnnualBostonUniversityConference on Language Development . Somerville , MA: Cascadilla Press . Cook, V. J., andNewson , M. (1996). Chomsky 's Universal Grammar (2nded.). Oxford: Blackwell . Crain, S. (1982). Temporalterms: Masteryby agefive. Papers andReports on ChildLanguage Development , 21, 33- 38. Crain, S. (1991). Language acquisitionin the absence of experience . Behavioral andBrainSciences , 14, 597650. Crain, S. (1992). TheSemantic SubsetPrinciplein the acquisition of quantification . Paperpresentedat the Workshopon theAcquisitionof Wh-ExtractionandRelatedWork on Quantification , Universityof Massachusetts , Amherst. Crain, S. (1993). Semantic subsetutions . Invitedpaperat TheCenterfor CognitiveScience Conference : Early CognitionandtheTransitionto Language , Universityof Texas , Austin. Crain, S., Conway,L., andThornton,R. (1994). D-quantification in childlanguage . In J. Fuller, H. Han, and D. Parkinson (Eds.), ESCOL'94: Proceedings of the11thEastern StatesConference onLinguistics . Ithaca , NY: CornellUniversity,Department of Linguistics . Crain, S., andFodor,J. D. (1984). On theinnateness of Subjacency . In G. Alvarez, B. Brodie,andT. McCoy, (Eds.), ESCOL'84: Proceedings of theFirstEastern StatesConference onLinguistics . Columbus : TheOhio StateUniversity,Department of Linguistics . Crain, S., and Fodor, J. D. (1985). How cangrammarshelp parsers ? In D. R. Dowty, L. Karttunen , and A. Zwicky (Eds.), Naturallanguage parsing : Psychological , computational , and theoretical perspectives . Cambridge : Cambridge UniversityPress . Crain, S., andFodor,J. D. (1987). Sentence matchingandovergeneration . Cognition , 26, 123- 169. Crain, 5'1andFodorlJ. D. (1993). Competence andperformance . In E. Dromi (Ed.), Language andcognition : A developmental perspective . Norwood, NJ: Ablex. Crain, S., andHamburger , H. (1992). Semantics , knowledgeandNP modification . In R. Levine(Ed.), Formal grammar , theoryandimplementation (Vol. 2). Vancouver : Universityof BritishColumbiaPress . Crain, S., andMcKee,C. (1985). Theacquisition of structuralrestrictions on anaphora . In S. Bermanl ]. Choe, andJ. McDonough(Eds.), Proceedings of NELS15. Amherst : Universityof Massachusetts , GLSA. Crain, S., McKeelC., andEmiliani , M. (1990). Visitingrelativesin Italy. In L. FrazierandJ. deVilliers(Eds.), Language processing andlanguage acquisition . Dordrecht : Kluwer. Crain, S., andNakayama , M. (1987). Structuredependence in grammarformation . Language , 63, 522- 543. Crain, 5'1NL W., andConwaYIL. (1994). Leaminglparsingandmodularity.In C. Clifton, L. FrazierlandK. Rayner(Eds.), Perspectives onsentence processing . Hillsdale , NJ: Lawrence Erlbaum . Crainl5'1NL W'I Shankweiler , D'I Conway,L., andBrazelD. (1996). Meaning , memory,andmodularity.In C. Schutze(Ed')1 MIT occasional papersin linguistics9: Proceedings of NELS26 Sentence Processing Workshop . Cambridgel MA: MIT, Department of LinguisticsandPhilosophy I MITWPL.

References

333

Crain, S., and Philip, W . (1993). Globalsemanticdependencies in child language . Paper presentedat the 16th GLOW Colloquium, Lund, Sweden. Crain, S., and Shankweiler, D. (1991). Modularity and learning to read. In I. G. Mattingly and M . StuddertKennedy (Eds.), Modularity and themotor theoryof speech perception . Hillsdale, NJ: LawrenceErlbaum. Crain, S., Shankweiler, D., Macaruso, Po, and Bar-Shalom, E. (1990). Working memory and comprehension of spoken sentences : Investigations of children with reading disorder. In G. Vallar and T. Shallice (Eds.), Neuropsychological disordersof sharf-term memory. Cambridge: Cambridge University Press. Crain, S., and Steedman, M . (1985). On not being led up the garden path: The use of context by the psychological parser. In D. R. Dowty , L. Karttunen, and A . Zwicky (Eds.), Natural languageparsing: Psychological , computational , and theoreticalperspectives . Cambridge: CambridgeUniversity Press. Crain , S., Thornton , R., Boster , C., Conway , L., Lillo -Martin , D ., and Woodams , E. (1996 ). Quantification

without qualification. Language Acquisition, 5, 83- 153. Crain, S., Thornton, R., and Murasugi, K. (1987). Capturing the evasivepassive. Paperpresentedat the 12th Annual Boston University Conferenceon LanguageDevelopment, Boston. Crain, S., and Wexler, K. (forthcoming). Methodology in the study of language acquisition: A modular approach. In T. Bhatia and W. Ritchie (Eds.), Handbookon languageacquisition . San Diego, CA: Academic

Press .

Culicover, P. (1991). Polarity, inversion, and focus in English. In G. F. Westphal, B. Ao, and H.-R. Chae (Eds.), ESCOL '91: Proceedings of the Eighth EasternStatesConference on Linguistics . Columbus: The Ohio State University, Department of Linguistics. Daneman, M ., and Carpenter, P. A . (1980). Individual differencesin working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 450 - 466 .

Davidson, D. (1967). Truth and meaning. Synthese , 17, 304- 323. Deutsch, W ., Koster, C., and Koster, J. (1986). What can we learn from children's errors in understanding anaphora? Linguistics , 24, 203- 225. de Villiers, J., Roeper, T., and Vainikka, A . (1990). The acquisition of long-distancerules. In L. Frazier and J. de Villiers (Eds.), Languageprocessing and languageacquisition . Dordrecht: Kluwer. de Villiers, J. Tager-Flusberg, H ., Hakuta, K., and Cohen, M . (1979). Children's comprehensionof relative clauses. Journal of Psycholinguistic Research, 8, 499 - 518 .

Dobrovie-Sorin, C. (1990). Clitic doubling, wh-movement, and quantification in Romanian. Linguistic Inquiry , 21, 351 - 398 .

Eisenberg, S. L., and Cairns, H. S. (1994). The developmentof infinitives from three to five. Journalof Child Language, 21, 713 - 734 . Erreich , A . (1984 ). Learning how to ask: Patterns of inversion in yes/ no and wh -questions . Journal of Child Language, 11, 579 - 592 .

Ferreiro, E., Othenin-Girard, C., Chipman, H. H., and Sinclair, H. (1976). How do children handle relative clauses ? ArchivesdePsychologie , 44, 229- 266. Flynn, s. (1986a). A parameter -setting modelof L2 acquisition : Experimentalstudiesin anaphora . Dordrecht: Reidel .

Flynn, S. (1986b). Production vs. comprehension: Differencesin underlying competences . Studiesin Second LanguageAcquisition, 8, 135- 164. Fodor, J. A . (1983). Themodularityof mind. Cambridge, MA : MIT Press. Fodor, J. A . (1984). Observation reconsidered. Philosophyand Science , 51, 23- 43. [Reprinted in A . Goldman (Ed.), Readingsin philosophyand cognitivescience . Cambridge, MA : MIT Press, 1993.] Fodor, J. A ., and Pylyshyn, Z. (1988). Connectionismand cognitive architecture: A critical analysis. Cognition, 28, 3- 71. [Reprinted in A . Goldman (Ed.), Readingsin philosophyand cognitivescience . Cambridge, MA : MIT Press, 1993 .]

Fodor, J. D., and Crain, S. (1987). Simplicity and generality of rules in language acquisition. In B. MacWhinney (Ed.), Mechanismsof languageacquisition . Hillsdale, NJ: LawrenceErlbaum. Frazier, L. (1978). On comprehending sentences : Syntacticparsingstrategies . Doctoral dissertation, University of Connecticut

, Storrs .

Frazier, L ., and Fodor , J. D . (1978 ). The sausage machine : A new two -stage parsing model . Cognition, 6, 291 - 325 .

Frazier, L., and Rayner, K. (1982). Making and correcting errors during sentencecomprehension: Eye movementsin the analysisof structurally ambiguoussentences . CognitivePsychology , 14, 178- 210. Frege, G. (1893). Ober Sinn und Bedeutung. Zeitschriftfur Philosophieund philosophische Kritik, 100, 25- 50. [Translated as Frege 1952 .]

334

References

Frege , G. (1952). On senseandreference . In P. GeachandM. Black(Eds.), Translations fromthephilosophical writingsof GottlobFrege . Oxford: BasilBlackwell . Gleitman , l . (1990). Thestructuralsources of verbmeanings . Language Acquisition , I, 3- 55. Goodluck , H. (1991). Language acquisition : A linguisticintroduction . Oxford: Blackwell . Goodluck , H., andTavakolian , S. L. (1982). Competence andprocessing in children 's grammarof relative clauses . Cognition , II , 1- 27. Gordon, P. (1996). The truth-valuejudgmenttask. In D. McDaniel , C. McKee, and H. S. Carins(Eds.), Methods for assessing children 's syntax . Cambridge , MA: MIT Press . Gorrell, P., Crain, S., andFodor, J. D. (1989). Contextualinformationandtemporalterms.Journalof Child Language , 16, 623- 632. Grice, H. P. (1975). Logicandconversation . In P. ColeandJ. Morgan(Eds.), Syntaxandsemantics 3: Speech acts . NewYork: AcademicPress . Grimshaw , J., andRosen , S. T. (1990). Knowledgeandobedience : Thedevelopmental statusof the binding theory. Linguistic Inquiry, 21, 187- 222. Grodzinsky , Y. (1990). Theoretical perspectives onlanguage deficits . Cambridge , MA: MIT Press . Grodzinsky , Y., andReinhart , T. (1993). Theinnateness of bindingandthe development of coreference :A reply to GrimshawandRosen . Linguistic Inquiry,24, 69- 103. Groenendijk , J., andStokhof , M. (1991). Dynamicpredicatelogic. Linguistics andPhilosophy , 14, 39- 100. Guasti, M. T. (1996). Acquisitionof Italianinterrogatives . In H. Clahsen(Ed.), Generative perspectives on language acquisition . Amsterdam : JohnBenjamins . Guasti,M. T., Dubugnon , C., Hassan -Shlonsky , S., andSchnitter , M. (1995). Quelques aspects del'acquisition desphrases relatives [Someaspectsof the acquisitionof relativeclauses ]. Ms., DIPSCO , Milan, and theUniversityof Geneva . Guasti, M. T., andShlonsky , U. (1995). The acquisitionof Frenchrelativeclausesreconsidered . Language Acquisition , 4, 257- 277. Guasti , M. T., Thornton,R., andWexler, K. (1995). Negationin children 's questions : Thecaseof English . In D. MaclaughlinandS. McEwen(Eds.), Proceedings of the19thAnnualBostonUniversityConference on Language Development . Somerville , MA: Cascadilla Press . Haegeman , L. (1994). Introduction to Government andBindingTheory(2nd ed.). Oxford: Blackwell . Hamburger , H. (1981). A deletionaheadof its time. Cognition , 8, 389- 416. Hamburger , H., andCrain, S. (1982). Relativeacquisition . In SiKuczaj(Ed.), Language development : Syntax andsemantics . Hillsdale , NJ: Lawrence Erlbaum . Hamburger , H., andCrain, S. (1984). Acquisitionof cognitivecompiling . Cognition , 17, 85- 136. Hamburger , H., andWexler, K. (1973). Identifiabilityof a classof transformational grammars . In K. J. J. Hintikka, J. M. E. Moravcsik , and P. Suppes(Eds .), Approaches to naturallanguage . Dordrecht : Reidel. Heim, I. (1982). Thesemantics of definiteandindefinite NPs. Doctoraldissertation , Universityof Massachu setts,Amherst. Heim, I., Lasnik , H., andMay, R. (1991). Reciprocityandplurality. Linguistic Inquiry,22, 63- 102. Hirsh-Pasek , K., andGolinkoff, R. M. (1996). Theoriginsof grammar : Evidence from earlylanguage compre hension . Cambridge , MA: MIT Press . Hirsh-Pasek ; K., Golinkoff,R., Hermon , G., andKaufman , D. (1995). Evidence fromcomprehension for early knowledgeof pronouns . In E. Clark(Ed.), TheProceedings of the26thAnnualChildLanguage Research Forum . Stanford , CA: CSLI.[Distributedby Cambridge UniversityPress .] Hornstein , N., andlightfoot, D. (1981). Explanation in linguistics . London:Longman . Hornstein , N., Rosen , S. To, andUriagereka , J. (1995). Integralpredications . In J. Camacho , L. Choueiri , and M. Watanabe (Eds.), Proceedings of the14thWestCoastConference onFormalLinguistics . Stanford , CA: CSLI.[Distributedby Cambridge UniversityPress .] Hsu, J. R., andHsu, L. M. (1996). Issuesin designingresearch andevaluatingdatapertainingto children 's syntacticknowledge . In D. McDaniel , C. McKee, H. S. Cairns(Eds.), Methods for assessing children 's syntax . Cambridge , MA: MIT Press . Hyams , N. (1986). Language acquisition andthetheoryof parameters . 0 ordrecht: Reidel . lnhelder , B., andPiaget , J. (1964). Theearlygrowthof logicin thechild. London:RoutledgeandKeganPaul. Jakubowicz , C. (1984). On markedness andbindingprinciples . In C. JonesandP. Sells(Eds.), Proceedings of NELS14. Amherst:Universityof Massachusetts , GLSA. Jaeggli , O. (1980). Remarks on to contraction . Linguistic Inquiry, 11, 239- 245. Kamp,H. (1981). A theoryof truth andsemantic interpretation . In J. Groenendijk , T. Janssen , andM. Stokhof (Eds.), Formalmethods in thestudyof language . Amsterdam : Mathematical Centre .

References

335

Kamp, H., and Reyle, U. (1993). From discourseto logic: Introductionto modeltheoretic semanticsof natural language , formal logicand DiscourseRepresentation Theory. Dordrecht: Kluwer. Kripke, S. (1972). Naming and necessity. In D. Davidson and G. Harman (Eds.), Semanticsfor natural language . Dordrecht: Reidel. Labelle, M . (1990). Predication, wh-movement, and the developmentof relative clauses . Language Acquisition, 1, 95- 119. Larson, R., and Segal, G. (1995). Knowledgeof meaning . Cambridge, MA : MIT Press. Lasnik, H. (1990). Essayson restrictiveness and learnability. Dordrecht: Kluwer. Lasnik, H., and Crain, S. (1985). On the acquisition of pronominal reference. Lingua, 65, 135- 154. Lasnik, H., and Uriagereka, J. (1988). A coursein GB syntax. Cambridge, MA : MIT Press. Lebeaux, D. (1988). Languageacquisitionand the form of the grammar. Doctoral dissertation, University of Massachusetts , Amherst. Lee, H., and Wexler, K. (1987). The acquisitionof reflexivesand pronounsin Koreanfrom the cross -linguistic perspective . Paper presented at the 12th Annual Boston University Conference on Language Development, Boston. Lewis, D. (1975). Adverbs of quantification. In E. Keenan (Ed.), Formal semanticsof natural language . Cambridge: Cambridge University Press. Lewis, D. (1979). Scorekeepingin a languagegame. In R. Bauerle, U. Egli, and A . von Stechow(Eds.), Seman ticsfrom differentpointsof view. Berlin: Springer-Verlag. Lust, B. (1981). Constraints on anaphorain child language: A prediction for a universal. In S. L. Tavakolian (Ed.), Languageacquisitionand linguistictheory. Cambridge, MA : MIT Press. Lust, B., Chien, Y.-C., and Flynn, S. (1987). What children know: Comparisonsof experimentalmethods for the study of first languageacquisition. In B. Lust (Ed.), Studiesin the acquisitionof anaphora(Vol . 2). Dordrecht: Reidel. Lust, B., Eisele, J., and Mazuka, R. (1992). The binding theory module: Evidencefrom first languageacquisition for Principle C. Language , 68, 333- 358. Lust, B., Loveland, K., and Kornet, R. (1980). The development of anaphorain first language: Syntactic and pragmatic constraints. LinguisticAnalysis, 6, 359- 391. MacWhinney, B., and Bates, E. (1989). Thecross -linguisticstudy of sentence processing . New York: Cambridge University Press. MacWhinney, B., and Snow, C. (1985). The Child Language Data Exchange System. Journal of Child Language , 12, 271- 296. Marcus, G. (1993). Negative evidencein languageacquisition. Cognition, 46, 53- 85. Martin, R. (1993). Short-term memory and sentenceprocessing: Evidencefrom neuropsychology. Memory and Cognition, 21, 176- 183. Matthei, E. H. (1981). Children's interpretations of sentencescontaining reciprocals. In S. L. Tavakolian (Ed.), Languageacquisitionand linguistictheory. Cambridge, MA : MIT Press. Matthei, E. H. (1982). The acquisition of prenominal modifiers. Cognition, 11, 301- 332. Mayer, J. W ., Erreich, A ., and Valian, V . (1978). Transformations, basicoperationsand languageacquisition. Cognition, 6, 1- 13. McCloskey, J. (1979). Transformational syntaxand modeltheoreticsemantics . Dordrecht: Reidel. McDaniel, D. (1986). Conditionson wh-chains. Doctoral dissertation, City University of New York. McDaniel, D., Cairns, H. 5., and Hsu, J. R. (1990). Binding principles in the grammars of young chil-

dren. Language Acquisition , 1, 121- 139. McDaniel, D ., Cairns, H. S., and Hsu, J. R. (1991). Control principles in the grammarsof young children. LanguageAcquisition, 1, 297- 335. McDaniel, D., Chiu, B., and Maxfield, T. (1995). Parameters ' for wh-movement types. Natural Language& LinguisticTheory, 13, 709- 753. McDaniel, D., and Maxfield, T. (1992a). The nature of the anti-c-command requirement: Evidence from young children. LinguisticInquiry, 23, 667- 671. -

McDaniel

,

358

D

and

.

Maxfield

,

T

.

-

( 1992b

) .

Principle

B

~

d

contrastive

stress

.

Language

Acquisition

,

2

H

.

,

337

-

.

McDaniel

,

D

luck

. ,

and

and

McDaniel

,

E

-

. ,

.

Language

D

Hughes

McKee

M

. ,

.

and

,

.

( 1993

( Eds

McKee

,

C

Rochement

and

Development

,

A

.

C

.

( 1996

Zukowski

.

) .

. ) ,

) .

Which

children

Island

Children

( Eds

Somerville

did

constraints

' s

. ) ,

,

.

MA

Proceedings

:

Cascadilla

they

show

Dordrecht

oblique

relatives

of

know

:

the

Press

.

20th

.

strong

Kluwer

In

Annual

crossover

?

In

Good

-

.

A

.

Stringfellow

Boston

,

University

D

.

Cahana

- Amitay

Conference

,

on

336

References

McDaniel, D., McKee, C./ and Cairns, H. S. (Eds). (1996). Methodsfor assessing children's syntax. Cambridgel MA : MIT

Press .

McKee, C. (1988). Italian children's masteryof binding. Doctoral dissertation, University of Connecticut, Storrs .

McKee, C. (1992). A comparison of pronouns and anaphors in Italian and English acquisition. Language Acquisition, 2, 21- 54. McKee , C. (1996 ). On -line methods . In D . McDaniel , C. McKee , and H . S. Cairns (Eds.), Methods for assessing

children's syntax. Cambridge, MA : MIT Press. McKee, C., and Emiliani, M . (1992). II clitico: ( 'e ma non si vede. Natural Language& LinguisticTheory, 10, 415 - 438 .

McKee, C., Nicol, J., and McDaniel, D. (1993). Children's application of binding during sentenceprocessing . Language and Cognitive Processes , 8, 265 - 290 .

McNeill , D. (1970). The acquisitionof language : The study of developmental psycholinguistics . New York: Harper and Row .

Melvold , J. (1991). Factivity and definiteness. In L. L. S. Cheng and H. Demirdache(Eds.), More paperson wh-movement : MIT working papersin linguistics(V 01. 15). Cambridge, MA : MIT , Department of Linguistics and Philosophy, MITWPL . ' Miyamoto , Y . (1992 ). The collective and distributive interpretation in child grammar : A study on quantification . General examination , University of Connecticut , Storrs .

Miyamoto , Y., and Crain, S. (1991). Children's interpretationof plural pronouns : Collectivevs. distributive interpretation . Paper presented at the 16th Annual Boston University Conference on Language Development, Boston. Morgan, J. L., and Travis, L. L. (1989). Limits on negative information in language input. Journalof Child Language, 16, 531 - 552 .

Naigles, L. (1990). Children use syntax to learn verb meanings. Journalof Child Language , 17, 357- 374. Nakayama, M . (1987). Performance factors in subject-auxiliary inversion by children. Journal of Child Language, 14, 113 - 125 .

Nelson K. G., Carskaddon, G., and Bonvillian, J. D. (1973). Syntax acquisition: Impact of environmental variation in adult verbal interaction with the child. Child Development , 44, 497- 504. Ni, W ., Crain, S.~ and Shankweiler, D. (1996). Sidesteppinggarden paths: Assessingthe contributions of syntax, semanticsand plausibility in resolving ambiguities. Languageand Cognitive Processes , II , 283 - 334 .

Parsons, T. (1990). Eventsin thesemanticsof English. Cambridge, MA : MIT Press. Pesetsky, D. (1987). Wh-in-situ: Movement and unselectivebinding. In E. Reulandand A . ter Meulen (Eds.), Therepresentation of (in)de:finiteness . Cambridge, MA : MIT Press. Philip, W. (1991). Spreadingin the acquisition of universal quantifiers. In D. Bates (Ed.), Proceedings of the Tenth West Coast Conference on Formal Linguistics . Stanford, CA : CSLI. [Distributed by Cambridge University Press.]

Philip, W . (1995). Event quantificationin the acquisitionof universal quantification . Doctoral dissertation, University

of Massachusetts , Amherst .

Phinney, M . (1981). Syntacticconstraintsand the acquisitionof embedded sententialcomplements . Doctoral dissertation , University

of Massachusetts , Amherst .

Pinker, S. (1984). Languagelearnabilityand languagedevelopment . Cambridge, MA : Harvard University Press. Pinker , S.' (1990 ). Language acquisition . In D . N . Osherson and H . Lasnik (Eds.), Language: An invitation to

cognitivescience(V 01. 1). Cambridge, MA : MIT Press. P~tter, M "., and Lombardi, L. (1990). Regenerationin the short-term recall of sentences . Journalof Memory and Language, 29, 633 - 654 .

Radfo~d, A . (1988). Transformational grammar. Cambridge: CambridgeUniversity Press. Radford, A . (1990). Syntactictheoryand the acquisitionof Englishsyntax: Thenatureof early childgrammarsof English. Oxford : Basil Blackwell. Reinhart, T. (1983a). Anaphoraandsemanticinterpretation . Chicago: University of Chicago Press. Reinhart, T. (1983b). Coreferenceand bound anaphora: A restatementof the anaphoraquestion. Linguistics and Philosophy , 6, 47- 88. Reinhart, T. (1986). Center and periphery in the grammar of anaphora. In B. Lust (Ed.), Studiesin the acquisition of anaphora( V01. 1). Dordrecht: Reidel. Rizzi, L. (1990). RelativizedMinimality . Cambridge, MA : MIT Press. Rizzi, L. (1993/ 1994). Some notes on linguistic theory and language development: The case of root infinitives . Language Acquisition , 3, 371 - 395 .

References

337

Rizzi, L. (1996). Thefinestructure of theleftperiphery . In A. BenettiandL. Rizzi(Eds.), Parameters andfunctionalheads . New York: Oxford UniversityPress . Roeper , T. (1972). Approaches to acquisition theory , with datafrom Germanchildren . Doctoraldissertation , HarvardUniversity, Cambridge , MA. Roeper , T. (1982). On the importance of syntaxandthe logicaluseof evidencein languageacquisition . In S. Kuczaj(Ed.), Language development : Syntaxandsemantics (Vol. 1). Hillsdale , NJ: Lawrence Erlbaum . Roeper , T. (1986). How children_acquirebound variables . In B. Lust (Ed.), Studiesin the acquisition of anavhora (Vol. 1). Dordrecht : Reidel . Roeper , T., and de Villiers, J. (1991a ). The emergence of boundvariablestructures . In T. M~ field and B. Plunkett(Eds.), Universityof Massachusetts occasional papers : Papersin the acquisition of WHo Amherst:Universityof Massachusetts , GLSA. Roeper , T., and de Villiers, J. (1991b). Orderedparameters in the acquisitionof wh-questions . In J. Weissenborn , H. Goodluck , andT. Roeper(Eds.), Theoretical issues in language acquisition : Continuityand change in development . Hillsdale , NJ: Lawrence Erlbaum . Roeper , T., Rooth, M., Mallis, L., andAkiyama , S. (1985). Theproblem of emptycategories andboundvariables in language acquisition . Ms., Universityof Massachusetts , Amherst. Roeper , T., andWeissenborn , J. (1990). How to makeparameters work. In L. FrazierandJ. deVilliers(Eds.), Language proceesing andlanguage acquisition . Dordrecht : Kluwer. Rosen , T. J., andRosen , S. T. (1995). Inferringthe innateness of syntacticknowledge . In E. Clark(Ed.), The Proceedings of the26thAnnualChildLanguage Research Forum . Stanford , CA: CSLI. [Distributedby Cambridge UniversityPress .] Russell , B. (1948). Humanknowledge : Its scope andlimits. London:Allen andUnwin. Sarma , J. (1991). Theacquisition of wh-quesfions in English . Doctoraldissertation , Universityof Connecticut , Storrs. Schaeffer , J. (1996). Objectscrambling in DutchandItalianchildlanguage . Doctoraldissertation , Universityof California , LosAngeles . Selkirk,E. (1972). Thephrase phonology of English andFrench . Doctoral dissertation , MIT, Cambridge , MA. Shankweiler , D., and Crain, S. (1986). Language mechanisms andreadingdisorder : A modularapproach . Cognition , 24, 139- 168. Shankweiler , D., Crain, S., Gorrell, P., and Tuller, B. (1989). Receptionof languagein agrammatism . Language andCognitive Processes , 4, 1- 33. Sheldon , A. (1974). The role of parallelfunctionin the acquisitionof relativeclauses in English . Journalof VerbalLearning andVerbalBehavior , 13, 272- 281. Sigurj6nsd6ttir , S., Hyams, N., and Chien, Y.-C. (1988). The acquisitionof reflexivesand pronounsby Icelandicchildren . Papers andReports onChildLanguage Development , 27, 97- 106. Snyder , W., and Rothstein , S. (1992). A note on contraction , caseand complementizers . TheLinguistic Review , 9, 251- 266. Solan , L. (1983). Pronominal reference : Child'language andthetheoryofgrammar . Dordrecht : Reidel . Solan , L., and Roeper , T. (1978). Children 's use of syntacticstructurein interpretingrelativeclauses . In H. Goodluckand S. Lawrence(Eds.), Universityof Massachusetts occasional papers4: Papers in the structure anddevelopment of childlanguage . Amherst:Universityof Massachusetts , GLSA. Speas , M. (1994). Null argumentsin a theory of economyof projection . In E. Benedictoand J. Runner (Eds.), Universityof Massachusetts occasional papers17: Functional projections . Amherst:Universityof Massachusetts , GLSA. Sperber , D., andWilson, D. (1986). Relevance . Oxford: Blackwell . Steedman , M., andAltmann, G. (1989). Ambiguityin context: A reply. Language andCognitive Processes , 4, (SpecialIssue ), 105- 122. Stowell, T., andBeghelli , F. (1995). Thedirection of quantifier movement . Paperpresented at the 17thGLOW Colloquium , Vienna. Strornswold , K. (1990). Learnability andfheacquisition of auxiliaries . Doctoraldissertation , MIT, Cambridge , MA. Szabolcsi , A. (1994). Thenounphrase . In F. KieferandK. Kiss(Eds.), Syntaxandsemantics 27: Thesyntactic structure of Hungarian . SanDiego, CA: Academic Press . Takahashi , M. (1991). Children ' s interpretation of sentences containingevery . In T. MaxfieldandB. Plunkett (Eds.), Universityof Massachusetts occasional papers : Papers in theacquisition of WHoAmherst: Universityof Massachusetts , GLSA. Tanenhaus , M. K., and Trueswell , J. C. (1995). Sentence comprehension . In J. Miller andP. Eimas(Eds.), Handbook of perception andcognition . Vol. 11, Speech , language , andcommunication . SanDiegolCA: AcademicPress .

338

References

Tavakolian, S. (1978). Children's comprehensionof pronominal subjectsand missing subjectsin complicated sentences . In H. Goodluck and L. Solan (Eds.), University of Massachusetts occasional papers4: Papersin thestructureand development of child language . Amherst: University of Massachusetts , GLSA. Tavakolian, S. L. (1981). The conjoined-clauseanalysisof relative clauses.In S. L. Tavakolian (Ed.), Language acquisitionand linguistictheory. Cambridge, MA : MIT Press. Thornton, R. (1990). Adventuresin long-distancemoving: The acquisitionof complexwh-questions . Doctoral dissertation, University of Connecticut, Storrs. Thornton, R. (1993). Childrenwho don't raise the negative . Paper presentedat the Annual Meeting of the Linguistic Society of America, Los Angeles. Thornton, R. (1995). Referentiality and wh-movement in child English: Juvenile D-linkuency. Language Acquisition, 4, 139- 175. Thornton, R. (1996). Elicited production. In D. McDaniel, C. McKee, and H. S. Cairns (Eds.), Methodsfor assessing children's syntax. Cambridge, MA : MIT Press. Thornton, R., and Crain, S. (1989). Children's useof boundpronouns . Paperpresentedat the Annual Meeting of the Linguistic Society of America, Washington, DC. Thornton, R., and Crain, S. (1994). Successfulcyclic movement. In T. Hoekstra and B. Schwartz (Eds.), Languageacquisitionstudiesin generativegrammar. Amsterdam: John Benjamins. Thornton, R., and Gavruseva, E. (1996). Children's split "whose-questions" and the structure of possessive NPs. Paperpresentedat the 21st Annual Boston University Conferenceon LanguageDevelopment, Boston. Thornton, R., and Wexler, K. (forthcoming). Children's interpretationof pronounsin PrincipleBand VP ellipsis structures . Cambridge, MA : MIT Press. Travis, L. (1984). Parameters and effectsof word ordervariation. Doctoral dissertation, MIT , Cambridge, MA . Tyler, L. (1985). Real-time comprehensionprocessesin agrammatism: A casestudy. Brain and Language , 26, 259- 275. van Hoek, K. (1995). Conceptualreferencepoints: A cognitive grammar account of pronominal anaphora constraints. Language , 71, 310- 340. Wason, P. (1965). Contexts of plausibledenial. Journalof VerbalLearningand VerbalBehavior , 4, 7- 11. Wason, P. (1980). The verification task and beyond. In D. R. Olson (Ed.), Thesocialfoundationsof language and thought. New York: Norton . Watanabe, A. (1993). Larsonian CP recursion, factive complements, and selection. In A . J. Schafer(Ed.), NELS23 ( Vol. 1). Amherst: University of Massachusetts , GLSA. Wexler, K., and Chien, Y.-C. (1985). The developmentof lexical anaphorsand pronouns. Papersand Reports on Child LanguageDevelopment , 24, 138- 149. Wexler, K., and Culicover, P. (1980). Formalprinciplesof languageacquisition . Cambridge, MA : MIT Press. Wilson, B., and Peters, A . M . (1984). What are you cDokin' on a hot?: A blind child's "violation" of universal constraints. Paper presentedat the 9th Boston University Conferenceon LanguageDevelopment, Boston.

Index

Altmannl

GerrYI

57

Ambiguous

also

' 52

34

from

27

,

29

,

Atkinsonl

,

the

- Shalom

,

Bellugi

,

Berko

,

Eva

,

Fillipo

,

,

,

Carey

,

Helen

,

Peter

,

Chien

,

98

Chierchia

Chiu

-

,

187

,

,

,

. ,

,

268

,

Daneman , Mo, 108 Davidson , Donald, 210, 294 Deutsch , Werner,268 deVilliers, Jill, 93, 152, 251, 293- 294 Discourse Representation Theory, 303 Dobrovie-Sorin, Carmen , 200

271

,

108

,

-

22

26

271

,

-

45

-

272

16

,

42

46

,

317

,

50

,

58

,

73

,

75

,

83

,

97

-

Eisele, Julie, 33, 35 , 36, 50- 51, 87- 91, 218 , 240, ,

152

,

284

-

286

244

,

Emiliani , Maria , 70 , 147 , 148 , 160 , 319 ,

,

247

3

,

,

Eve

,

,

,

,

Ileana

model

Condition

of

271

,

99

,

125

,

165

,

166

Errors . See also Performance

-

type II , 130, 227

201

,

-

33

,

36

,

25

,

33

193 - 195 , 200 - 202

Existential Closure. SeeUnselectiveBinding Extension (reference), 278

200

performance

,

32

-

,

,

assent

5

-

6

,

97

-

,

242

98

33

223

,

233

276

plausible

factors

275 , 300 -

falsification

,

of

98

UG-compatible, 37- 40, 126, 146, 171- 172, 174, 32

versus

Competition

,

200

200

Grammar

,

79

68

,

Competence

,

type 1, 129- 130, 180, 213 , 221, 223 , 234 , 240,

,

34

Comorovski

6

267

194

model

Cognitive

Erreich , Anne , 147 , 172

287

Guglielmo

,

318

192

Sandra

Coalition

260

155

314

,

,

Noam

Clark

-

Eisenberg, Sarita L., 148

,

,

316

304

Chomsky

,

3151

21

Gennaro

Carol

Chung

,

297, 304 Cook, Vivian, 5 Culicover, Peter, 72, 106

21

25

270

,

Cinque

,

,

Bonnie

,

,

A

,

Chomsky

167

301

21

,

68

Gaye

-

,

303

,

- Chin

268

,

21

148

Courtney

Yu

,

.,

34

,

,

201

22

Patricia

Carskaddon

Cazden

-

Conditionof plausibledissent(Russell 's Maxim), 225- 226, 232, 237, 240, 243, 245, 257, 260261, 271, 276- 282, 290, 300 Conjoined -clauseanalysis . SeeRelativeclauses Constraints on discourse , 16- 17, 24 on form, 11- 13 learningof, 24- 27 aslinguisticprinciples , 4, 10 on meaning , 13- 14 Context. SeealsoPerformance factors null, 72, 104- 107 specificversusgeneral , 104(seealsoPrincipleof Referential Success ) ContinuityHypothesis , 31, 35, 37- 38, 41, 121, 199 Conway,Laura , 42- 43, 50, 117- 119, 286, 292,

21

S

,

.,

,

,

Carpenter

271

Melissa

Roger

,

19

9

Martin

,

Cairns

D

,

,

Brown

stimuiusl

125

Hagit

Braine

the

147

Paul

Bowerman

of

147

John

,

See

200

,

Bonvillion

Borer

.

32

,

Ursula

,

212

126

,

Jean

Bloom

-

30

Elizabeth

Beghelli

211

68

2681

,

,

1031

283

SergeYI

Bates

531

poverty

MartinI

Avrutinl

-

factors

Arlenel

Argument

Condition

103

Performance

Amidonl

Bar

,

sentencesl

,

300

,

237

,

257

,

FerreirolEmilial142 FlynnlSuzannel 45- 46173- 77, 97- 98, 106 Fodor,Janet34141145, 68"1125, 154, 1611319 Fodor,Jerry, 29, 99- 1001104, 107, 319

340

Index

Frazierl

Frege

Lyn

,

,

,

,

Gavruseval

,

Golinkoff

,

Goodluck

,

Gordon

,

Gorrelll

72

Helen

,

145

571

,

Paul

Lust,Barbara , 33, 35, 36, 45- 46, 50- 51~73, 74- 75, 87- 91, 97- 98, 218,240,244,319

,

Peterl

.

40

1L

Roberta Mo , 32, 49, 58- 61, 145 34. 148 , 152 , 155

Paul

H

,

Lilal

Lewis,David, 152, 294- 295 Lightfoot, David, 19- 20, 27, 119 Lombardi , Linda, 75- 76 Loveland , K., 88- 89

154

278

Elena

Gleitman

Gricel

127

Gottlob

34

,

"

64

68

69

Grimshaw , Jane , 80- 84, Ill , 244, 269, 272- 274, 275, 299 Grodzinksy , Yosef, 83, 126, 268, 270 Groenendijk , Jeroen , 42, 290, 303 Guasti , MariaTeresa , 38- 39, 69, 143, 1481174, 199 Haegeman , Liliane,155 Hamburger , Henry, 68- 69, 70, 72, 106, 116, 124, 148, 151, 157- 159, 197, 302 Hanlon, Camille , 21 Heim, Irene, 42, 50, 295- 296, 303, 316- 317 Hirsh-Pasek , Kathy, 32, 49, 58- 61, 145, 269 Hornstein , Norbert, 19- 20, 27, 43, 119 . Hsu1JenniferR'I 155, 2681271 Huntley, Gavin, 314 HyamslNina, 29, 125, 268 Hypothesis . SeealsoModularityHypothesis , ContinuityHypothesis experimental vs. null, 129 Null 83, 124, 129, 270 structure -dependent vs. structure ~independent , 165- 168, 1731218

Maturation , 9- 10, 121 McCloskey , 194 - , -James McDaniel , Dana, 34, 35, 36, 58, 61- 65, 91- 92, 155, 160, 192, 251- 256, 257, 268, 269, 271 McKee , Cecile , 35, 58, 61- 65, 70, 91- 92, 147, 148, 160, 209, 219, 251- 256, 257, 269, 319 McNeill, David, 22 MacWhinney , Brianl32 Marcus , Gary121 Martin, Randi, 75- 76, 99, 101 Matthei, Edward , 123, 319 May, Robert,316- 317 Mayer, JudithW., 172 Maxfield, Thomas , 34, 35136, 192, 268 Mazuka , Reiko,33, 35, 36, 50- 51, 87- 91, 218, 240, 244 Medial-Wh. SeeWh-phrases Miller, George , 98 Miyamoto, YoichL122, 309, 315- 316 ModularityHypothesis , 29, 107, 111 Morgan, James , 21 Murasugi , Keiko, 122, 144, 147 Naigles

, Letitia

Nakayama Negative

Input Matching model, 7, 31- 33, 35, 37- 44, 146, 178, 199 Intension (meaning), 278 Jakubowicz, Celia, 268, 271, 319 Kamp, Hans, 42, 50, 303 Komet, R., 88- 89 Koster, Charlotte, 268

Koster,Jan,268 Kripke, Saul,278

Keith

Newson Ni

, Janet

Lebeaux , David, 126, 174, 197 Lee, Hyeonjin, 268

, 163

, 165

, 169

-

175

, 107

, 58 , 61

Scope

, 117

-

119

, 127

- 65 , 269

. See

Unselective

Binding

Overgeneration semantic, 26 syntactic, 27

Parameter null subject ,9 Parsons , Terence , 294 Parsing children , 111- 112 difficulties

, 98 -

garden

path

preferences 112

Labelle , Marie , 69, 148 Language Acquisition Device(LAD), 7, .37, 115, 117- 1201 121 Lasnik , Howard , 21, 60, 74, 125,180,188,316- 317

, 147

- 22 , 23

, 5

, 103

Nuclear

, 20

E ., 21

, Mark

, Weijia

Nicol

, 126

evidence

Nelson

Imitation. SeeTasks lnhelder , Barbel , 93, 293 Innateness hallmarksof, 9, 37, 61 Hypothesis , 5, 10, 23- 24 naturevs. nurture,37

, 145

, Mineharu

for

, 116

102

effects

, 124

, 98 , 100

one

of

a priori

Principle

of

Parsimony

Principle

of

Referential

strategies

107

, 112

, 84 - 85 , 111

, 94 , 103

Plausibility , 105

, 104 , 115

Success -

154 parser

-

, 313

Principle

universal

, 105

interpretation

, 30

106

, 111

, 116

, 103 -

112

-

117

, 315

, 115

-

, 281

- 316 120

,

-

Index Performance factors biasto makea sentence True(to say"Yes"), 3435, 52- 53, 84- 85, 116, 299 carryovereffects , 34, 94, 302 center(self) embedding , 34, 100 changingword order, 35, 255 directionality(linearorder), 73- 76, 89- 90, 218 errors(noise), 45, 84, 89- 90, 172, 197, 297, 298 experimental artifacts , 121 extralinguistic (realworld) knowledge , 30, 41, 43, 48, 52, 73- 77, 107- 111, 121 fatigue, 124, 126 general(uncontrolled ) influences , 4, 33, 298, 302 length, 124, 126 nonlinguisticcomplexity , 122- 124 numberof (animate ) NPs, 34, 155 numberof clauses , 33- 34 orderof mention(vs. conceptual order), 34, 67- 68 pragmaticlead(linguisticantecedent ), 74- 76, 8889, 144, 169, 224, 234, 236 presuppositions , 69, 72, 105, 116(seealsoRelative clauses ) slipsof thetongue, 124 strategies , 94, 302(seealsoParsing ) stress , 272- 273 tendencyto usereducedforms, 53, 71, 180, 188189 testtakingskills, 301 verbalworkingmemory , 48, 98- 102, 107- 111, 124, 270 warm-up effects , 124, 126 Pesetsky , David, 200 Philip, William, 34, 93- 95, 118, 293- 302 Phinney , Marianne , 71, 319 Piaget , Jean , 93, 293 Pinker , Steven , 6, 21, 31, 37, 72, 147 Possession inalienable vs. alienable , 43 Potter, Mary C., 75- 76 Povertyof the stimulus . SeeArgumentfromthe povertyof the stimulus Primarylinguisticdata, 11 Principlesversuspreferences , 60, 82- 85, 196- 198, 244, 307- 309 Pylyshyn , Zenon, 99- 100

main

- clause

vs . reduced

- relative

- clause

341

, 105

-

111

,

, 293

-

112 presuppositions Restrictor

of , 69 , 151 . See

Reyle

, Uwe

Rigid

Unselective

, Luigi

Roeper

, Sarah 269

, 188

, 194

-

195

, 199

- 201

, 93 , 123

, 125

, 154

, 251

, 257

T ., 25

, 272

Rosen

- 274

, T . John L

267

Russell

Sarma

, 226

, 148

, Lisa

Semantic Senten

, 148

, 177

Subset

Principle

. cejmeaning

pairs

Shankweiler

, Donald

Sheldon

, Amy

Shlonsky

, 117 . See

, 103

-

118

Constraints

, 107

, 126

, 127

, 301

, 69 , 148

, Ur

, 69

Sigurjonsd6ttir

, Sigridur

, Lawrence

Sperber

,

, 270

, Jeanette

Selkirk

, 244

, 299

, 25 - 26 , 33

- 268

, Jaya

Schaeffer

- 26 , 33 , 43 , 80 - 84 , Ill

, 275

, Bertrand

Solan

, 170

, 319

Rosen

Rule

160

, 278

, 38 , 125

, 'Thomas

294

-

, 42 , 50 , 303

designator

Rizzi

, 157 Binding

, DaI

, 268

, 75 , 218

, 319

} , 70

Statistics discrete

vs . continuous

in

reaction

times

in

research

normal

, 49

variables

studies

- 52 , 90 , 91 -

distributions

unimodal

, 57 - 58 , 79

, 56 - 58 92 , 250

, 255

, 46 - 49

vs . bimodal

distributions

, 46 - 47 , 92 ,

252 Steedman

, Mark

Stokhof

, Martin

Stowell

, Tim

Structure

, 57 , 103 , 42 , 290

107

, 116

, 319

, 200

- dependence

Subsetjsuperset 2991

-

, 303

. See

relations

Hypothesis

. 116

-

120

, 209

, 278

- 282

,

316

Szabolcsil

Annal

Takahashi

, Mari

Tanenhaus

40

, 93 , 293

, Michael

, 127

Tasks act

out

( figure

manipulation

) , 57 , 67 -

70 , 73 -

77 ,

152

Quantifiers . SeealsoUnselective Binding A-quantifiers (Adverbs), 293, 296 D-quantifiers (Determiners ), 293, 295- 296 Operators(quantificational NPs), 266, 272- 273, 317- 318

appropriateness cross eye

priming

movement

imitation judgment picture

Radford , Andrew, 10, 155 Rayner , Keith, 127 Reinhart , Tanya, 83, 267- 268, 270 Relativeclauses conjoined -clauseanalysis , 68- 70, 152- 153, 155157, 209

, 91 - 92 , 251

- modal

preferential transcripts Tavakolian

, 58 , 61 - 65 , 99

recording

, 71 -

, 99 , 107

-

111

, 127

77

, 79 - 82 verification

, 62 , 64 , 122

looking of

paradigm

spontaneous

, Susan

, 173

speech

, 34 , 69 , 75 , 148

218, 319 Travis, LisaD., 166 Travis, LisaL., 21

, 293

, 302

, 58 - 61 , 145 , 141 , 152

-

153

, 155

,

342

Index

Trueswell , John, 127 Tyler, Lorraine , 301 Universal Grammar , theory of; 3- 5, 9- 10. Seealso Errors , Constraints

UnselectiveBinding, 295- 296 Uriagereka, Juan, 43, 188 Vainikka

, Anne , 251

Valian , Virginia , 172 van Hoek , Karen , 25 , 33

Verbal working memory. SeePerformancefactors Wason , Peter , 225 , 301

Weissenborn , Jurgen, 125 Wexler , Kenneth , 6 , 9 , 25 - 26 , 29 , 38 - 39 , 58 , 6L 63 , 72 , 83 , 106 , 125 , 143 , 148 / 174 , 199 , 255 , 268 - 270 , 271 - 272 , 317

Wh-phrases bare versus full , 39 - 40 , 200 - 202 medial -wh , 37 - 38 , 40 , 187 , 201 Wilson

, Deirdre , 70