Title: Parameter resetting and learnability in the acquisition of French and Spanish by adult native speakers of English
CITATION THUMBNAILS PAGE IMAGE ZOOMABLE
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/UF00102723/00001
 Material Information
Title: Parameter resetting and learnability in the acquisition of French and Spanish by adult native speakers of English
Physical Description: Book
Language: English
Creator: Ayoun, Dalila, 1963-
Copyright Date: 1992
 Record Information
Bibliographic ID: UF00102723
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: oclc - 27719744
ltuf - AJM5608

Full Text










PARAMETER RESETTING AND LEARNABILITY IN THE ACQUISITION OF
FRENCH AND SPANISH BY ADULT NATIVE SPEAKERS OF ENGLISH














By

DALILA AYOUN


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA


1992














ACKNOWLEDGEMENTS


I am very thankful to Dr. David Birdsong for agreeing

to direct my dissertation as well as for his support, guidan-

ce, challenge and friendship. Likewise I am appreciative of

the scholarly knowledge, advice and encouragement provided by

Dr. Jean Casagrande, Dr. John Lipski, and Dr. Ann Wehmeyer,

my committee members. Several other persons have made signifi-

cant contributions to the writing of this dissertation.

Special thanks go to David Johns who wrote the Turbo Pascal

software for both experimental tasks carried out in this

study. Sincere thanks go to all the students who participated

in the experiment and made that dissertation possible. I also

benefited from enlightening discussions with Dr. Andrea Tyler,

who was always very encouraging and generous with her time.

I am also grateful to a special friend, Leon Hounshell, for

his continuous support and faith in my abilities. And last,

but not least, my thanks and gratitude go to my husband,

Robert S. Clancy, for his patience, support and love through-

out my doctoral work at the University of Florida. He applied

all his logical mathematical thinking and computer science

expertise in running the necessary statistical analyses on

the Statistical Analysis Software program.














PREFACE


How do adults acquire a second language? This experimental

study addresses this very broad question from two perpec-

tives. Within a learnability perspective, the principle to be

investigated is the Subset Principle and the predictions that

can be made based on a superset/subset relation. Within a

parameter resetting perspective, the purpose of this study is

to investigate whether the following syntactic properties--

preposition stranding, dative alternation, dative passive and

Exceptional-Case marking--do cluster and how they may be

acquired. This study looks at the acquisition of English by

adult learners of French or Spanish.

Our point of departure is the principles and parameters

approach to Universal Grammar (UG), as realized in Government

and Binding (GB) Theory (Chomsky 1981c). "This theory assumes

that principles and parameters of UG constitute an innately

given body of knowledge which constrains first language (L1)

acquisition" (White 1989c). Assuming this approach to be

correct, the question arises as to whether UG is still avail-

able in second language acquisition.


iii








The parameter resetting approach and the learnability

perspective make both similar and competing predictions as

far as second language acquisition is concerned. First, both

approaches predict that the acquisition of the cluster of

syntactic properties will take place based on positive

evidence alone. Within a parameter resetting perspective we

assume that it is the positive evidence present in the input

that will trigger the acquisition of the properties subsumed

under a parameter. The acquisition of any property will

trigger the acquisition of the remaining properties. Further-

more, once a property has been acquired, the acquisition of

the other properties should follow whether or not there is any

kind of evidence available, positive or negative. Within a

learnability perspective, it is claimed that the learners

should entertain the smallest hypothesis compatible with the

positive evidence available to them in the input. However, the

parameter resetting approach predicts that the properties will

be acquired simultaneously as a cluster. In contrast, the

learnability perspective breaks up that cluster by predicting

that the properties will be acquired individually in a partic-

ular order, based on the positive evidence available to them.

Chapter 1 presents an overview of the learnability con-

straints and learning principles which may guide first lan-

guage acquisition, but not second language acquisition. Second

language learners face different situations depending on the

various grammatical properties of their native language and








the target language. It is shown that this fact opens up

various possibilities within a learnability perspective and

a parameter resetting perspective. Chapter 2 reviews the lin-

guistic theory concerning the cluster of syntactic properties

to be investigated in this study within a parameter resetting

perspective, and outline the predictions which can be made

within that perspective. Chapter 3 examines the relevant

second language literature and introduces the second perspec-

tive and its predictions adopted in this study, the learnabi-

lity perspective. Chapter 4 outlines the experimental study

which consists of a grammaticality judgment task and a correc-

tion task. Chapter 5 restates the hypotheses and predictions

made within the parameter resetting perspective and the learn-

ability perspective, and reports the results of both experi-

mental tasks. In Chapter 6, these results are summarized and

relevant conclusions are drawn.














TABLE OF CONTENTS


page
ACKNOWLEDGEMENTS ......................................... ii

PREFACE.................................................. iii

ABSTRACT ................................................ ix

CHAPTERS

1 INTRODUCTION......................................... 1

Learnability Theory in First Language Acquisition... 1
Learnability Theory in Second Language Acquisition... 5
Parameter Resetting in Second Language Acquisition.. 11


2 REVIEW OF LINGUISTIC THEORY......................... 15

Preposition Stranding .............................. 17
Exceptional-Case Marking............................ 20
Dative Alternation ................................. 22
Dative Passive ..................................... 24
Parameter Resetting Perspective..................... 25

3 REVIEW OF SECOND LANGUAGE ACQUISITION LITERATURE.... 32

Mazurkewich (1984a). The Acquisition of the Dative
Alternation by Second Language Learners and Lin-
guistic Theory..................................... 33
Mazurkewich (1985a). Syntactic Markedness and Lan-
guage Acquisition................................. 34
Le Compagnon (1984). Interference and Overgeneration
in Second Language Learning: The Acquisition of
English Dative Verbs by Native Speakers of
French............................................ 37
Hawkins (1987). Markedness and the Acquisition of
the English Dative Alternation by L2 Learners..... 38
Tanaka (1987). The Selective Use of Specific
Exemplars in Second Language Performance: The
Case of the Dative Alternation..................... 41
White (1987b). Markedness and Second Language Acqui-
sition: The question of Transfer.................. 44









Birdsong et al. (1984). Universals versus Transfer
Revisited........................................ .47
Van Buren and Sharwood Smith (1985). The Acquisition
of Preposition Stranding by Second Language Learners
and Parametric Variation........................... 48
Sheppard (1991). At Sea in SLA: Evidence of UG in the
Acquisition of French and English Verbs........... 50
Learnability Perspective............................. 52

4 THE EXPERIMENTAL STUDY............................... 58

Subjects........................................... 59
Population Attributes............................... 60
Methodology........................................ .60
Test condition................................... 60
Instruments. ..................................... 61
Stimuli............................................ 64
Pilot Study....................................... 66
Subjects. ....................................... 66
Methods........................................... 67
Results ......................................... 67

5 RESULTS AND INTERPRETATIONS........................ 70

Analyses........................................... .71
Results of the Grammaticality Judgment Task........ 72
Parameter resetting perspective................... 72
Overall results: French and Spanish groups........ 72
Results by groups................................ 76
Rejecting versus accepting....................... 84
Overall results .................................. 87
Analysis by sentence sets........................ 89
Peripheral issues................................ 99
Learnability perspective......................... 104
Results of the Correction Task .................... 109

6 SUMMARY AND CONCLUSION............................ 119

APPENDICES

A RESULTS OF THE GRAMMATICALITY JUDGMENT TASK....... 137

B RESULTS OF TIMING ................................. 143

C ITEM FREQUENCY ANALYSIS ........................... 150

D BACKGROUND DATA QUESTIONNAIRE....................... 155

E BACKGROUND DATA ANALYSIS .......................... 158

F BACKGROUND DATA CORRELATIONS...................... 162


vii









G STIMULI........................................... 167

H INSTRUCTIONS...................................... 170

I PROBABILITY TABLE ................................ 171


REFERENCES.... .......................................... 174

BIOGRAPHICAL SKETCH....................................... 186


viii













Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

PARAMETER RESETTING AND LEARNABILITY IN THE ACQUISITION OF
FRENCH AND SPANISH BY ADULT NATIVE SPEAKERS OF ENGLISH

By

Dalila Ayoun

August 1992



Chairperson: David Birdsong
Major Department: Romance Languages and Literatures

This dissertation investigates the second language ac-

quisition of a cluster of syntactic properties--preposition

stranding, Exceptional-Case marking, dative alternation and

dative passive--from a learnability perspective and a param-

eter resetting perspective. The experimental subjects are

adult native speakers of English at various stages of. acquisi-

tion of French or Spanish, as well as a control group of na-

tive speakers of French and Spanish. The subjects were asked

to perform a grammaticality judgment task and a correction

task to test several hypotheses. Within a parameter resetting

perspective, it is hypothesized that if the properties do

cluster, the acquisition of any property would trigger the

acquisition of the remaining properties based on positive

evidence. Within a learnability perspective the acquisitional








sequence of the proposed properties should show that second

language learners follow the Subset Principle. The stimuli

consist of 50 sentences randomly distributed. The French and

Spanish data include grammatical and ungrammatical sentences.

All the equivalent English sentences are grammatical. Instan-

ces of preposition stranding include ten grammatical senten-

ces with pied piping: five declaratives and five wh-interroga-

tives; and ten ungrammatical sentences with preposition

stranding: five declaratives and five wh-interrogatives.

Instances of Exceptional-Case marking and dative passive

include five grammatical and five ungrammatical sentences.

Instances of dative alternation include five double object

constructions, ungrammatical in French and Spanish, and five

prepositional constructions, grammatical in French and Span-

ish. The results do not indicate strong support for the param-

eter resetting hypothesis nor for the learnability hypothe-

sis. Subjects successfully acquire some of the properties

tested but fail to acquire others. However, it is argued that

the results show that subjects are acquiring the properties

tested based on positive evidence available to them. It is

concluded that direct or indirect negative evidence may be

necessary for second language learners to constrain an L2

grammar when the L1 is a superset of the L2.













CHAPTER I
INTRODUCTION


Learnability Theory in First Language Acquisition

Pinker (1989) argues that first language acquisition in

general can be seen as follows. The child hears a finite num-

ber of sentences symbolized by the X's in (1). He or she must

generalize from these finite sentences to an infinite number

of sentences which not only include but also go beyond the in-

put sample. This is represented by the circle with the arrow

pointing to it. The other circles symbolize the infinite num-

ber of hypotheses which are consistent with the input sample

but differ from it and from the correct hypothesis.

(1)








In the L1 acquisition and learnability literature, it is usu-

ally accepted that learnability constraints and learning prin-

ciples such as the Subset Principle (Berwick 1985; Wexler and

Manzini 1987) and the Uniqueness Principle (Wexler 1981; Roe-

per 1981; Berwick 1985) prevent the L1 learners from producing

over-inclusive grammars. Such constraints or principles allow







2

the children to formulate only conservative hypotheses, so

that language learning can proceed based on positive evidence

alone. According to Pinker (1989, pp. 5-6), there are three

different ways in which the child can formulate incorrect hy-

potheses which will be disconfirmed by positive evidence.

First "the child's language can be disjoint from the target

language" (p.5) as in (2a). Second, the language generated by

the child's hypothesis may intersect with the target language

as in (2b). Or, as claimed by the Subset Principle, the child

may entertain an hypothesis which is a subset of the target

language as in (2c). In all three cases the "+" symbol repre-

sents the positive evidence available to the child. That posi-

tive evidence allows the child to reject an incorrect hypoth-

esis.

(2) (a) (b) (c)



+ +




If the children were to formulate 'wild' hypotheses about the

input, they would need to have access to and be able to use

negative evidence in order to disconfirm these incorrect hy-

potheses. This is represented in (3): if the child overgener-

alizes by entertaining a grammar which generates a superset

of the target language, positive evidence will not be enough








3
to disconfirm the wrong hypotheses. Negative evidence will be

necessary.

(3)






Baker (1979) has pointed out a number of learnability

paradoxes which illustrate the fact that children do manage

to avoid, or unlearn, overgeneralization mistakes, when they

do not use negative evidence (Braine 1971, Brown 6 Hanlon

1979) or when negative evidence is ineffective (Pinker 1989).

The acquisition of the English dative alternation is one of

these learnability paradoxes. Mazurkewich and White (1984)

argue that children do overgeneralize until positive evidence

in the input allows them to realize the morphological and se-

mantic constraints which apply to dative alternation. They re-

port that 46.7% of the following ungrammatical sentences were

judged to be grammatical by the group of 9-year-olds:

(4) a. David suggested Jack the trip

b. Tom reported the police the accident

c. Cathy described Lisa the movie

There are also some anecdotal examples of younger children who

do not seem to be aware of the semantic constraints governing

the dative alternation:

(5) Mummy, open Hadwen the door (speaker aged 6;0)

(cited in Mazurkewich & White 1984, p.270).







4
In White (1987a), it is suggested that overgeneralization

of the double-object construction does occur and cannot be

dismissed as random performance error on the part of the

children. However, as Gropen et al. (1989) point out, Podor

(1985) had already suggested that the "reported instances of

overgeneralization in children's spontaneous speech are proba-

bly rare (e.g. compared to obvious overgeneralizations like

heated ; the reports include little information about sample

size, and there is therefore no way of estimating the relevant

baselines" (p.208-9). Furthermore, Fodor (1985) also mentions

possible methodological problems in the experimental studies

carried out by Mazurkewich and White (1984).

Gropen et al. (1989) claim that although children are not

invariably conservative, they do show conservative tendencies,

and their overgeneralizations are influenced by morpho-phono-

logical and semantic criteria. In other words, the importance

of the overgeneralizations presented by Mazurkewich and White

(1984) is downplayed by Fodor (1985) and Gropen et al. (1989).

The latter study also minimizes the conclusions reached by

White (1987)--to stress the regularity of first language ac-

quisition. Gropen et al. (1989) explain the lack of overgene-

ralization errors first by pointing to the evidence that

adultlike meanings are often acquired quite slowly (Gentner

1975, 1978, 1982; Bowerman 1978). Incorrect verb meanings com-

bined with correct rules would lead to syntactic errors

according to the theory, and as verb meanings are refined,








5
these syntactic errors should disappear (Oropen et al., p.

250). Their second explanation is based on the assumption that

children do not differ from adults after all and that adult

innovations in the use of dative alternation are very similar

to the so-called overgeneralizations produced by children in

that:

both adults' and children's innovations obey some
version of the possession-change constraint, and both
are rare, occurring against a background of conserva-
tism or the minimal exceptions to it licensed by nar-
row-range rules. Both children and adults, then, may
possess a broad-range dative rule mapping cause-to-go
forms onto cause-to-have forms, and may occasionally
use them productively for one-time innovations in ad-
dition to their main function of licensing and moti-
vating more low-level lexical alternations.
(Gropen et al., p.252)


Learnability Theory in Second Lanauaae Acauisition

Second language learners face different situations depend-

ing on the various grammatical properties of their native lan-

guage and the target language. For example, the NL and the TL

may permit such different properties (e.g., the NL is a head-

first language whereas the TL is a head-last language) so that

no persistent transfer effects from the NL to the TL are

attested. This is illustrated by the diagram in (8a).

(8a)







6

Or the LI and the L2 may partially overlap, leading the

learner to initially hypothesis that the L2 grammar is

identical to his/her L2 grammar. This is illustrated in (9a):

(9a)

LI L2




For example, French and English partially overlap as far as

dative alternation goes since English allows both preposi-

tional structures and double object structures, while French

allows only prepositional structures (see examples in 10 and

11). An English native speaker learning French may initially

believe that French allows both dative forms.

In terms of superset/subset relations, learners may need

to start with a subset and build a superset or, on the other

hand, they may have to go from a superset to a subset. For

example, French learners of English may start with [NP PP] as

the only possible dative construction and will have to learn

that English allows not only the [NP PP] structure but also

the [NP NP] structure.

(10) a. Mark gave the letter to Paul

b. Marc a donn6 la lettre & Paul

(11) a. Mark gave Paul the letter

b. Marc a donn6 Paul la lettre







7

In that case French is a subset of English, as can be seen in

(12):

(12)
E




An English learner of French will provide the example of a

learner proceeding from a superset to a subset. It is the lat-

ter possibility which does seem to raise the most interesting

questions in second language acquisition learnability since,

as White (1990) points out, learners do not seem to observe

the Subset Principle (van Buren 1987). Rather they tend to

adopt grammars which generate a superset of the structures

actually allowed in the L2. In other words, second language

learners whose native language constitutes a superset of the

target language, will produce or accept ungrammatical sen-

tences because they lack the principles and constraints

imposed on first language acquisition. This suggests, as White

(1990) proposes, that negative evidence may be useful or even

necessary. Birdsong (1989) argues that "negative evidence is

necessary in L2 acquisition in order to disconfirm certain

types of hypotheses about grammatical structure" (p.131). One

type of hypothesis which seems to require negative evidence

for disconfirmation is the free variation hypothesis where

the learner assumes that two forms coexist in free variation,

i.e. are equally possible.








8
Birdsong (1989) reports a personal anedocte about a Span-

ish native speaker barber who seemed to entertain a free-

variation hypothesis when referring to the word 'hair,'

whereby the plural and singular would be in free variation.

The English speaking customers would use singular pronouns

when referring to 'hair,' but the Spanish and French Canadian

speaking customers would use singular and plural pronouns.

These facts may have led the barber to entertain a free-

variation hypothesis concerning plurality. Schachter (1986)

suggests the case of uninflected verb complements in which the

subject of the complement differs from the subject of the main

sentence. Schachter argues that the examples in (13a) could

induce the learner to hypothesize the pattern V + NP + V with

an optional to between the NP and the second verb. The learner

could produce grammatical but also ungrammatical structures

as in (13b):

(13a) Data available to the learner:

I want him to go I made him go

John expects me to go We had him polish the floor

They told the men to leave They let the children cross

the street
We helped her find a place I helped him wash the dishes


(13b) Data produced by the learner

I want him to go I made him go

* She made him to go I want him go







9

In the absence of negative evidence, the learner would have

"to figure out which nonoccurring sentences could not occur!"

(Schachter 1986, p. 219).

It is usually agreed (see e.g. Birdsong 1989, p.184) that

L1 learners do not entertain free variation hypotheses because

they are guided by the Uniqueness Principle (Wexler 1981; Roe-

per 1981; Berwick 1985) or one of its derivatives, the Princi-

ple of Contrast (Clark 1987). Under these learning principles,

learning is constrained by the assumption that if there is a

difference in form there is a difference in meaning. It fol-

lows that when two forms appear to be synonymous, one preempts

the other.

Thus, if second language learners were guided by the same

learning principles, and if they happened to entertain a free-

variation hypothesis, they should be able to preempt one over

the other. Unfortunately, this does not appear to be the case,

since as Birdsong (1989) speculates, the Uniqueness Principle,

"which represents a procedural linchpin in the mechanisms of

hypothesis testing without negative evidence, is rejected in

the course of linguistic development as the result of height-

ened metalinguistic awareness, and may not be applied by adult

L2 learners" (p.131).

It is then concluded that second language learners would

need negative evidence to preempt the wrong forms. Otherwise,

they would not know which non-occurring form cannot occur and







10

they would continue to entertain free-variation hypotheses,

ultimately leading to fossilization.

Carroll & Swain (1991) found that, when taught the morpho-

phonological and semantic constraint on English dative alter-

nation, Spanish learners were able to benefit from positive

input plus negative feedback, at least temporarily. All their

experimental groups, who were exposed to different types of

negative evidence, performed significantly better than the

comparison group (it is not specified who the subjects in the

comparison group were), who did not receive any feedback. The

subjects were divided in four groups. The subjects in Group

A were given an explicit hypothesis about how alternation

works and were told that this hypothesis was wrong. This type

of feedback was called Explicit Hypothesis Rejection. Subjects

in Group B were simply told that they were wrong whenever they

made a mistake. This was called the Explicit Utterance Rejec-

tion. Subjects in Group C were given a reformulated correct

response. This feedback was considered to be modelling and was

called Implicit Negative Feedback. Whenever subjects in Group

D made a mistake, they were asked if they were sure of their

response. This feedback was called Indirect Metalinguistic

Feedback.

As far as dative alternation is concerned, English con-

stitutes a superset, allowing both the [NP PP] and the [NP

NP] structures, whereas Spanish allows only the former:









(13) a. John gave a book to Mary
b. Juan le ha dado un libro a Maria

(14) a. John gave Mary a book

b. Juan le ha dado Maria un libro

Thus some type of negative evidence may indeed prove to be

particularly useful, if not necessary, in superset/subset

cases where the learner may need negative evidence to adopt

the correct L2 grammar. White (1991) attempted to determine

whether the L2 learner would transfer L1 argument structure

in the L2 when the structure in question does not occur in the

L2 input. The subjects were three groups of English children

undergoing partial immersion, total immersion, and submersion

in French. A preference task and a judgment task were used to

test double-object structures and adjacency (which included

adverb placement). As predicted, it was found that LI suboate-

gorisations of dative verbs were accepted in the L2 by all

experimental groups, confirming results reported in White

(1987b). When the L2 was the superset, the Partial Immersion

and Submersion groups were conservative, tending to prefer the

forms which were acceptable in both the L1 and L2. However,

these results were only based on a preference task and were

not replicated with a grammaticality judgment task.


Parameter Resettina in Second Lanauaae Acquisition

One of the goals of linguistic theory within a generative

framework is to explain language in terms of principles of







12

Universal Oramar rather than language-specific rules. "In a

highly idealized picture of language acquisition, Universal

Orammar is taken to be a characterization of the child's pre-

linguistic state" (Chomsky 1981, p.31). In addition, a number

of these abstract principles of Universal Grammar are associ-

ated with parameters, so that "the grammar of a particular

language can be regarded as simply a specification of values

of parameters of UO, nothing more" (Chomsky 1981c, p.31).

Thus languages vary depending on which parameter setting is

adopted. One of the first principles to be proposed was the

principle of structure-dependency which seems to be common to

all languages. The principle of structure-dependency asserts

that "operations on sentences such as movement require a

knowledge of the structural relationships of words rather than

their liner sequence" (Cook 1988, p.6). In first language ac-

quisition, the parameter values are set by positive evidence

alone and the acquisition of any property triggers the acqui-

sition of the remaining properties. The setting of a parameter

to a particular value has drastic effects on language. For

example, the Head Parameter specifies the position of heads

of phrasal categories. All languages have phrases (PP, VP, NP,

etc) which contain heads (P, V, N, etc) and the position of

the head within the phrase varies from language to language

yielding languages as different as English and Japanese.

English is head-initial since the complements appear to the

right of phrasal heads, whereas Japanese is head-final because







13

the complements are to the left of heads. This choice between

head-first or head-last is reflected in the respective word

orders SVO and 80V, in the order within NPs, with relative

clauses either preceding or following their head, as well as

in the orders of other phrase types. "Ideally, we hope to find

that complexities of properties differentiating otherwise

similar languages are reducible to a single parameter, fixed

in one or another way" (Chomsky 1981c, p.6).

This concept of parametric variation opens various possi-

bilities in second language acquisition since the first lan-

guage and the target language may have similar or different

parameter settings. The question is whether the second lan-

guage learners are able to adopt a different parameter setting

when the parameter values differ for the L1 and the L2. In

other words, are second language learners able to reset param-

eters to the appropriate L2 value based on the available posi-

tive evidence? As White (1989) argues:

Since many parameters link clusters of syntactic
properties, one needs to look for evidence of
clustering in second language acquisition.
Clustering should be expected both in the case
where the learner carries over the LI value of a
parameter and where the L2 value is achieved.
Learners who transfer the LI value of the parameter
ought to transfer the whole cluster of properties.
When learners acquire a parameter setting relevant
to L2, all properties associated with it should be
acquired together. (p.82)

The second language acquisition literature offers many

examples of previous attempts to show evidence of parameter

resetting. Phinney (1987), White (1985, 1986), Liceras







14

(1986) and Hilles (1986) investigated whether second language

learners acquired all the properties of the pro-drop

parameter. It appeared that there was transfer of the Ll

parameter setting, and sometimes, resetting to the L2 value.

But as far as clustering is concerned, only some of the

properties associated with the parameter were evidenced in

second language acquisition.

We propose to study the acquisition of a potential param-

eter, the Oblique-Case parameter, which holds a cluster of

apparently unrelated phenomena: dative alternation, dative

passive, preposition stranding and Exceptional-Case marking.

The experimental subjects will be three groups of English

native speakers learning French and three groups of English

native speakers learning Spanish. A parameter resetting per-

spective will be adopted in an attempt to find out if the

properties will cluster and how they may be acquired.

Chapter 2 will review the linguistic theory concerning the

cluster of syntactic properties to be investigated in this

study within a parameter resetting perspective, and will out-

line the predictions which can be made within that perspec-

tive. Chapter 3 will review the relevant second language ac-

quisition literature and will introduce the second perspective

and its predictions adopted in this study, the learnability

perspective. We will see that the parameter resetting perspec-

tive and the learnability perspective make similar and compe-

ting predictions.












CHAPTER 2
REVIEW OF LINGUISTIC THEORY


This chapter reviews the four syntactic properties to be

tested in this experimental study:

1 Preposition stranding

2 Exceptional-Case marking

3 Dative alternation

4 Dative passive.

Kayne (1981, 1983) claims that these apparently unrelated phe-

nomena of French syntax can be explained in a unified account

within the framework of Government and Binding adopted here.

And Chomsky (1981c, p.297) suggests that all four properties

may be subsumed under a single parameter. If this is indeed

the case, we will be able to test the acquisition of these

syntactic properties by second language learners of French

and Spanish (we will see that French and Spanish exhibit the

same characteristics as far as preposition stranding, Excep-

tional-Case marking, dative alternation and dative passive are

concerned) from a parameter resetting perspective.

The following assumptions of Case Theory are necessary:

a According to Chomsky (1981c, p.170), the fundamental

properties of Case assignment are

(1) (i) NP is nominative if governed by AGR







16

(ii) NP is objective if governed by V with

the subcategorisation feature: ____P (transitive)

(iii) NP is oblique if governed by P

(iv) NP is genitive in [np __X]

(v) NP is inherently Case-marked as determined by

properties of its [__.N] governor

(i)-(iv) correspond to structural Case; (v) is inherent case

and is only found in double NP constructions such as

(a) Mary sent Bill a letter

b According to Chomsky (1981c, p.172), "in English, only

the [-N] categories, verb and preposition, are Case-assigners.

In OB, it is assumed that verbs assign objective Case, and

that prepositions assign oblique Case apart from marked

properties". Chomsky (1981c) assumes with Kayne (1979) that

in English both verbs and prepositions assign objective Case,

because the richer Case systems have been lost. However,

Chomsky (1981c, p.293) suggests a slight modification of that

proposal: the Case system may be lost in English only within

VP, whereas PPs that are immediate constituents of 8' have in-

herent Case-marking.

c English prepositions have the property of assigning

objective Case, which is the Case normally assigned by a verb

to its direct object. In contrast, French prepositions assign

only oblique Case. Contrary to English, they do not have the

property of assigning both Cases. Kayne (1983) argues that it

is this distinction between French prepositions and English








17

prepositions that accounts for various properties of French

and English, as well as Spanish, which like French, does not

allow preposition stranding, dative alternation, Exceptional-

Case marking, nor dative passive. Thus English prepositions

have the property of assigning oblique Case as well as objec-

tive Case, whereas French and Spanish (which is similar to

French in that respect) prepositions assign only oblique Case.

Let us see how these facts may provided a unified account of

the syntactic properties to be tested in this study.

Preposition Stranding

English, but not French or Spanish allows preposition strand-

ing in wh-constructions and in passives:

(9) a. which candidate did you vote for?

b. quel candidate as-tu vot6 pour?

c. .cuil candidate has votado por?

(10) a. John was voted against by almost everybody

b. Jean a 6et vot6 centre par presque tous

c. Juan fue votado contra por casi todos

Kayne (1979b) adopts the following approach in terms of

Case theory:

inherent Case is assigned at D-Structure

structural Case is assigned at S-Structure

English has lost the inherent Case system: prepositions

do not assign inherent oblique Case but rather structural

objective Case, as do verbs

French prepositions assign oblique Case









The permissibility of preposition stranding in English would

then follow from the presumably marked loss of oblique Case

assignment in English. Weinberg and Hornstein (1978) note that

stranding is restricted to PPs that are within the VP. They

propose that a reanalysis rule within the VP creates a complex

verb including the preposition, the NP object of the preposi-

tion becoming the object of the complex derived verb, hence

accessible to movement in the normal way; movement of the

object of a preposition that is not reanalysed in this way is

excluded by a filter that blocks empty oblique elements as-

suming that P assigns oblique Case while verbs assign objecti-

ve Case (Chomsky 1981c, p.292).

Kayne (1981) argues that what imposes reanalysis is the

Empty Category Principle, rather than Hornstein and Neinberg's

Filter since French does have reanalysis rules: one reanalysis

rule involves causatives and another reanalysis rule involves

idiomatic expressions. So rather than interpreting the lack

of preposition stranding in French as resulting from the ab-

sence of a reanalysis rule, Kayne suggests that it results

from the absence of a reanalysis rule involving prepositions.

French lacks such a reanalysis rule involving prepositions

because, according to Chomsky (1981c, p.293), a reanalysis

between two lexical categories is possible only if they assign

Case in the same way. Kayne (1981) modifies that reanalysis

principle within the framework of government: reanalysis

between two lexical categories is possible only if they govern







19

in the same way. We have already seen that in English P and

V govern in the same way, whereas in French they do not. A re-

analysis rule involving prepositions is therefore impossible,

and this explains why preposition stranding is ungrammatical

in French. V-V reanalysis is possible in French because it in-

volves the same categories. And V-P reanalysis is possible in

English because V and P govern the same way, but not in

French, because it involves two different categories which

govern differently.

Thus it is concluded that although preposition stranding

occurs in English, it is not allowed in French or in any other

Romance Languages (Zribi-Hertz 1984, Pollock 1989). King and

Roberge (1990) attempt to challenge that conclusion by arguing

that the following sentences are instances of preposition

stranding in Prince Edward Island Canadian French (PEIF):

(11) a. Qui-ce tu vas a Ottawa a-travers-de?

Lit.: 'Who are you going to Ottawa through?'

b. Quoi-ce tu travailles dessus?

'What are you working on?'

c. Qui-ce tu as fait le g&teau pour?

'Who did you make the cake for?'

d. Le ciment a 6t6 march dedans avant d'&tre sec

'The concrete was walked on before drying'

e. Tu connais pas la fille que je te parle de

'You do not know the girl that I am talking to you

about'








20

PEIF is in a minority position with respect to English in the

province of Prince Edward Island and it is reasonable to as-

sume that this dialect of French is strongly influenced by

English. King and Roberge (1990) themselves indicate that PEIF

includes a number of prepositions of English origin contrary

to Qu6bec French, for instance, which has borrowed verbs but

no prepositions from English. So the fact that PEIF exhibits

instances of preposition stranding can be attributed to struc-

tural borrowing. It is limited to that dialect of French and

cannot be extended to other varieties of French or Standard

French. Therefore we may still assume that the preposition

stranding, Exceptional-Case marking, dative alternation and

dative passive may constitute a cluster of syntactic proper-

ties subsumed under the Oblique-Case parameter. However, if

preposition stranding was indeed acquired as part of the PEIF

grammar, and if it could be demonstrated that PEIF preposi-

tions actually assign Case the same way English prepositions

do, then these instances of preposition stranding would con-

stitute counterevidence to the Oblique-Case parameter.

Kayne shows that the distinctions between French and

English about Case assignment can be extended to explain

another syntactic difference between the two languages:

Exceptional-Case marking.

Exceptional-Case Marking

(12) a. I believe [John to be the most intelligent of all]

b. Je crois [Jean Etre le plus intelligent de tous]







21
c. Creo [Juan ser I maAs intelligent do todos]

(13) a. Quel gargon crois-tu [t 6tre 1e plus intelligent de

tous]

According to Kayne (1981), a comparative analysis of the

complementizer systems of French and English will bring out

the relation between preposition stranding and ZCM. Ds and

for are both infinitival complementizers. There are two major

differences between them:

foL can be followed by a lexical subject of the

infinitive but de cannot:

(14) *ce serait doumage do quelque chose lui arriver

(15) it would be a pity for something to happen to

him

Ad is compatible with control, for is not

(16) ce serait donmage de partir maintenant

(17) it would be a pity for to leave now
Example (14) is a straight violation of the Case Filter if we

assume that de cannot assign Case to the subject of the infin-

itive whereas fqo can. Example (12a) is usually explained by

assuming that 8'-deletion takes place; this is a marked prop-

erty of the English verb believe and similar verbs. However,

Kayne (1981) assumes that there is no 8'-deletion, but rather

a zero complementiser 0, a preposition which governs and

assigns Case, a complementiser analogous to flo. This is

essentially the analysis of Chomsky and Lasnik (1977). If this

is right, then English and French do not differ in the








22

structures assigned to the examples (12)(a) and (b). In each

case there is a 0-complementizer.

Kayne suggests that, once again, the basic difference

between French and English is that an English preposition

(here 0) assigns Case and governs in the same structural way

as the verb, whereas in French a preposition governs and as-

signs Case only inherently to an NP and its subcategorization

frame. Thus 0 neither governs nor assigns Case to the subject

of the infinitival construction in (12)(b).

Therefore, a single principle accounts for the differences

between French and English with regard to preposition strand-

ing and Exceptional-Case marking. That principle can be in-

formally stated as a difference in the way English and French

prepositions govern. English prepositions have the property

of assigning objective Case as well as oblique Case, whereas

French prepositions can only assign oblique Case.

Dative Alternation

(18) a. John gave Mary a book

b. John has left his children a great deal of money

c. They sent John a registered letter

(19) a. Jean a donn6 Marie un livre

b. Juan le ha dado Maria un libro

c. Jean a laiss6 ses enfants beaucoup d'argent

d. Juan le ha dejado su nifo much dinero

e. Ils ont envoy Jean une lettre recommand6e

f. Le han mandado Juan una carta registrada







23
(20) a. John gave Mary a book

b. John gave a book to Mary

Oehrle (1976) proposed a base analysis for (3) without NP-

movement, and without a preposition present at every stage of

the derivation. Kayne agrees that there is no NP-movement,

but argues that a preposition is present at every stage of the

derivation, in a way more in the spirit of trace theory than

would a to-deletion rule.

(21) Jean a essay de chanter

The rule of equi-NP deletion that derived (21) from '. .. de

Jean/lui chanter ..' by deleting the subject of the infini-

tive, has given way to the idea that (21) contains an empty

pronominal subject at every stage of the derivation: de PRO

chanter. Similarly, it is suggested that (22) contains an

empty preposition Pe, at every stage of its derivation:

(22) V [pp Pe NP] NP

The difference between French and English regarding dative

alternation can now be explained in terms of Case theory:

(23) An empty preposition cannot be the source of Case

Thus in (20) Mary is not directly assigned Case by Pe, but by

V through percolation. This is possible because P and V assign

Case in the same way as seen before. It follows that the dou-

ble object construction is ungrammatical in French: it is a

violation of the Case Filter since only one NP is assigned

Case. Case cannot be assigned by percolation through V since








24

in French P assigns oblique Case whereas V assigns objective

Case.

Dative Passive

(24) a. John was given a book by Mary

b. Jean a 6t6 donn6 un livre par Marie

c. Juan fue dado un libro por Maria

Postulating an empty preposition in the active form in (25):

(25) John gave Mary a book

leads us to conclude that the passive form in (24) is an in-

stance of preposition strand-ng. Then the ungrammaticality of

(24)(b) and (c) simply follows from the impossibility of prep-

osition stranding in French and Spanish.

Kayne extends the analysis seen in (1) to preposition

stranding in passives on the basis of the following assump-

tions: a Case-marked trace is identified as a variable. Fol-

lowing Siegel (1974), NP-mvt leaves Case behind on the trace;

reanalysis follows inherent Case-assignment. Following these

assumptions, in English, lacking inherent Case-assignment in

VP, reanalysis will permit the passive, since no structural

Case is assigned by the passive participle.

Chomsky (1981c, p.297) concludes that Kayne's analysis of

preposition stranding and Exceptional-Case marking "thus shows

how a single parameter of UG accounts for a variety of prop-

erties of the two language types, while providing further

evidence for the projection principle, successive cyclic move-

ment and several other principles".








25
Chomsky does point out that the proposed analysis of ECH

raises a few problems and wonders whether an analysis assum-

ing a O-complementier is really the correct one. But he con-

cludes that "Kayne's approach to unifying preposition strand-

ing and Zxceptional-Case marking is, however, sufficiently

attractive so that an attempt to resolve the remaining diffi-

culties surely seems in order" (1981c, p.298). Thus the Oov-

vernment and Binding (OB) Theory provides a sound analysis of

these syntactic properties which may be subsumed under a

parameter.


Parameter resetting perspective

Much of the grammatical theorizing done in second lan-

guage research has focused on parameter resetting following

the principles and parameters approach within Universal Gram-

mar. Under this view, a certain number of parameters are set

to the value of the first language being acquired. The ques-

tion to be investigated in second language acquisition is the

resetting of parameters when the value of the second language

differs from the value of the native language.

a The pro-drop parameter

The most studied parameter is the pro-drop or null subject

parameter (Chomsky 1981c, Jaeggli 1982, Rizsi 1982) and the

second language acquisition literature offers many examples

of previous attempts to show evidence of parameter resetting

(Phinney 1987, White 1985, 1986, Liceras 1986, Hilles 1986).







26

In general, it appears that that there was transfer of the L1

parameter setting, and sometimes, resetting to the L2 value.

But as far as clustering is concerned, only some of the prop-

erties associated with the parameter were evidenced in second

language acquisition.

However, as White (1989) points out, there is some disa-

greement as to what properties are actually subsumed under

the pro-drop parameter. One version assumed by Chomsky (1981),

Liceras (1988b), Phinney (1987), Rizzi (1982) and White (1985,

1986) claims that [+ pro-drop] languages such as Spanish and

Italian exhibit the following properties:

null subjects

no pleonastic pronouns

rich verbal agreement

subject-verb inversion in declaratives

that-trace sequences

Further [- pro-drop] languages such as English exhibit these

properties:

lexical subjects

pleonastic pronouns

lack of rich agreement

no subject-verb inversion in declaratives

no that-trace sequences

According to the version assumed by Hilles (1986) and Hyams

(1986), [+ pro-drop] languages would have the following

properties:









null subjects

no pleonastic pronouns

auxiliaries and main verbs from one category

whereas [- pro-drop] languages would exhibit the following

characteristics:

lexical subjects

pleonastic pronouns

auxiliaries are distinct from main verbs

White (1989, p.85) argues that the main difference between the

two versions is that the first version assumes that subject-

verb inversion and that-trace sequences are part of the

[+ pro-drop] value of the parameter whereas the second version

does not. But Chao (1981), Safir (1982) and Hyams (1983) have

also suggested that the Verb-Subject word order may not be

part of the pro-drop parameter. This lack of consensus as to

what properties cluster under the pro-drop parameter weakens

its status as a putative parameter.

White (1985) tested the acquisition of the cluster of

properties subsumed under the pro-drop parameter by native

speakers of Spanish learning English as a second language. The

results of a grammaticality judgment task reveal that the

Spanish subjects accepted many ungrammatical English sentences

as grammatical, whereas the French-speaking controls did not,

and that the subjects' performance improved with increasing

levels of proficiency. Referring to White (1985, 1986), White

(1987c) concludes that parametricc variation offers a







28

potential explanation of transfer errors with the native

speakers of pro-drop languages transferring that parameter to

the L2 so that LI influence falls within a universals-oriented

approach to the L2 acquisition" (White 1987c, p.240).

In the White (1986) study, the experimental group (32 na-

tive speakers of Spanish, 2 native speakers of Italian) and

the control group (37 native speakers of French) performed a

grammaticality judgment task and written question formation

task. The first task tested three of the properties assumed

to cluster with the first version of the parameter: null

subjects, subject-verb inversion and that-trace effects. The

results show that the Spanish speakers were more likely than

the French subjects to accept the subjectless sentences. No

difference was found for the other two properties. White sug-

gests that this result may be explained by the choice of items

which may have been inappropriate. However, it must be con-

cluded that the properties did not cluster in the interlan-

guage of the subjects tested.

Phinney (1987), who also assumed the first version of the

pro-drop parameter, looked at two properties--subject pro-

nouns and the agreement system--in Spanish native speakers

learning English (ESL group) and English native speakers

learning Spanish (88L). The results of written production data

(i.e. free compositions) show that the two language groups

differ as far as evidence for parameter resetting is con-

cerned. Verbal agreement was quite accurate in all groups and







29

the ESL groups did omit many more pleonastic pronouns than

referential pronouns. In contrast, SSL groups correctly

omitted referential as well as pleonastic pronoun subjects in

Spanish and did not attempt to use a lexical pronoun as sub-

ject of impersonal constructions which suggests that they were

successful in resetting the parameter to the target language

value.

Hilles (1986), who adopted the second version of the param-

eter, used production data from a 12-year old native speaker

of Spanish learning English as a second language. It was found

that pronoun omission declined over the period of the study

(10 months) in parallel with the emergence of auxiliary verbs

as a distinct category. This suggests that the parameter was

being reset to the L2 value. However, some caution is in order

since this result is based on a single case study. More re-

search is needed to determine whether this result can be rep-

licated with a greater number of subjects.

Thus, the results obtained by White (1985, 1986), Phinney

(1987) and Hilles (1986) show some evidence for transfer of

the LI parameter setting as well as resetting to the L2 value.

However, only some of the properties subsumed under the pro-

drop parameter appear to cluster in second language acquisi-

tion.

b The Oblique-Case parameter

The present experimental study will investigate whether the

proposed syntactic properties subsumed under what we may call








30
the oblique-Case parameter, discussed above, do cluster and

how they may be acquired. English would be [+ structural Case]

since English prepositions can assign structural Case, whereas

French and Spanish would be [- structural Case] since French

and Spanish prepositions can only assign inherent Case.

The hypotheses to be tested are the following:

A if the properties do cluster, the acquisition of any

property will trigger the acquisition of the remaining

properties.

B the French properties and the Spanish properties will

be acquired in the same way.

In A it is implied that if the properties do cluster, the ac-

quisition of any property, irrespective of what that property

may be, will trigger the acquisition of the other properties.

So if for instance dative alternation is acquired first, it

will then trigger the acquisition of dative passive alterna-

tion, dative passive and Exceptional-Case marking. Hypothesis

A also contains the assumption that the acquisition of the

first property will take place based on positive evidence

alone. The acquisition of any one property should trigger the

acquisition of the other properties even in the absence of

positive or negative evidence.

Hypothesis B is based on the crosslinguistic similarity of

French and Spanish. The experiment will test whether English

learners of French and English learners of Spanish acquire the







31

same properties as a cluster and following the same acquisi-

tional sequence.

We have thus introduced the parameter resetting perspec-

tive and its predictions. Chapter 3 will review the relevant

second language acquisition literature and introduce the sec-

ond perspective adopted in this experimental study: the learn-

ability perspective and its predictions as far as the acquisi-

tion of the proposed cluster of syntactic properties.













CHAPTER 3
REVIEW OF THE SECOND LANGUAGE ACQUISITION LITERATURE

Lightbown and White (1987, p.501) stress the fact that
"until very recently, little attempt was made to relate re-

search on learner language to specific hypotheses within a

linguistic theory. In the 1980s there has been a considerable
revival of interest in Chomskyan linguistic theory as a source
of hypotheses about second language acquisition," of which
the following studies are examples. In other words, we are

testing the relevance of Universal Orammar (UG) to second
language acquisition. UG consists of principles and parame-
ters guiding first language acquisition. However, these prin-
ciples and parameters are constantly being refined. It is in
this sense that we can say that the theory of second language
acquisition depends on linguistic theory.

The following studies are relevant to the present experi-
mental study in that all of them tested one or more of the
four syntactic properties which may be subsumed under a param-
eter. Mazurkewich (1984a), Le Compagnon (1984), White (1987a),
Hawkins (1987) and Tanaka (1987) looked at the acquisition of
dative alternation in English as a second language, while in
White (1987b) the target language was French. Mazurkewich
(1985b) looked at the acquisition of English dative








33

passive by Inuktitut and French native speakers. White

(1987b), Birdsong et al. (1984), van Buren and Sharwood Smith

(1985) and Bardovi-Harlig (1988) investigated the acquisition

of preposition stranding. Sheppard (1991) is--to the best of

my knowledge--the only study which investigated the acquisi-

tion of Exceptional-Case marking by both English and French

native speakers.



Mazurkewich (1984a). The Acquisition of the Dative
Alternation by Second Language Learners and Linguistic
Theory

Mazurkewich (1984a), using a test that elicited intuitive

judgments, found that French-speaking and Inuktitut-speaking

students learning English acquired the unmarked dative prepo-

sitional phrase complements before the marked double-NP com-

plements. The type 1 sentences were generally judged to be

grammatical by all subjects, whereas Type 2 sentences shows

only an increasing judgment of acceptability across the three

levels in the Inuit groups and the French groups show less

systematic acquisition of that structure compared to the Inuit

groups.

Type 1:

(26) a. Peter threw a football to Philip

b. Diane baked a cake for Nicole

Type 2:

(27) a. Peter threw Philip a football

b. Diane baked Nicole a cake








34

It is argued that this acquisitional sequence supports the

theory of markedness which claims that unmarked structures,

which are part of core grammar, are acquired before marked

structures, which are peripheral rules of the core grammar.

However, that claim is disputed by Kellerman (1985), who sug-

gests a number of weaknesses in Mazurkewich (1984a) regarding

(a) the examples chosen to illustrate dative alternation, (b)

the choice of subjects in the experiment and the elicitation

format used, and especially (c) the presentation and statisti-

cal treatment of the data.

White (1985) remarks that the Inuit subjects at all levels

consistently produced a higher percentage of marked construc-

tions than unmarked constructions, which seems to be incon-

sistent with Mazurkewich's claim that the unmarked forms pre-

cede the marked forms developmentally. White (1985, p.315)

argues that "learners who have already been made aware of the

possibility of marked constructions in Ll may also assume that

they occur in L2." This may not necessarily be the case for

all learners or for all marked properties. Learners may assume

that Ll marked constructions are not allowed in the L2.



Mazurkewich (1985a). Syntactic Markedness
and Language Acquisition


Mazurkewich (1985a) focused on the acquisition of passive

dative structures by adolescents and young adults who were na-

tive French- and native Inuktitut-speaking students. Data were







35

elicited on the acquisition of dative structures in a passive

context using stimulus sentences containing to- and for-dative

verbs:

(28) a. A football was thrown to Philip

b. A cake was baked for Nicole

(29) a. Philip was thrown a football

b. Nicole was baked a cake

Data were also elicited using nonalternating dative verbs

which allow only passivised direct objects:

(30) a. A trip was suggested to Ruth

b. A house was designed for Claire

In general, the results indicate that passive structures

in which the direct object is passivised are acquired before

passive structures in which the indirect object is passivi-

sed. The same groups of subjects were used to elicit data tes-

ting the acquisition of dative questions in a passive context.

Once again, the results show that the unmarked question forms

are acquired before the marked ones. It is argued that the re-

sults support the predictions made by the theory of markedness

as outlined in that study.

In a study testing French learners of English as a second

language, Bardovi-Harlig (1988) also takes issue with the re-

sults of the Mazurkewich (1984b) study, arguing that the Inuk-

titut speakers'acquisition of preposition stranding over pied

piping does not support Masurkewich's claim since in contrast

to the French speakers, the Inuktitut speakers produced more







36

preposition stranding forms than pied piping form in early

stages. Furthermore, the possible transfer of pied piping by

French speakers is dismissed based on two very marginal inst-

ances of what Mazurkewich claims to be preposition stranding

in Canadian French, which goes against Vinet's (1979) analysis

of these examples:

(31) Le boss que je travaille pour

'The boss that I work for'

(32) Le gars quo j'ai sorti avec

'The guy that I went out with'

Vinet (1979) maintains that these structures in Quebec French

are not the result of preposition stranding, but of the fact

that these prepositions are intransitive. These constructions

are not only stylistically marked but restricted to 'avec' and

'pour'; 'sans' and 'apr6s' are also possible but much less

common. Furthermore, according to Vinet, these structures can-

not occur in questions (Vinet 1979, p.1):

(33) a. *Qui sors-tu avec?

b. *Qui tu sors avec?

'Who are you going out with?'

According to Vinet, the ungrammaticality of these interroga-

tive structures supports the intransitive preposition analy-

sis: prepositions cannot be stranded in French as a result of

Wh-movement.

Bardovi-Harlig (1988) found that the putatively marked

form, preposition stranding, was acquired before the unmarked








37

form, pied piping, in both wh-question and relative clauses.

It is suggested that this apparent counter-example to the

markedness hypothesis could be explained by positing a second

factor, salience, which would play a role in determining the

acquisition order. Bardovi-Harlig defined salience in terms

of availability of data: preposition stranding is more salient

than pied piping, being used more frequently, and therefore

standing out more prominently in the L2 input directed to the

learners. However, as suggested by Birdsong (1992, personal

communication), it could also be argued that pied piping is

more salient that preposition stranding. Since pied piping is

used less frequently that preposition stranding, its rare

occurrence may be more noticed, i.e. more salient.



Le Compagnon (1984). Interference and Overgeneration
in Second Language Learning: the Acquisition of English
Dative Verbs by Native Speakers of French

Le Compagnon (1984) studied the relationship between full

noun and pronominal indirect objects in dative alternation.

She elicited spontaneous production data in a longitudinal

study of two French adult native speakers, and a grammaticali-

ty judgment task with four subjects. The results of the lon-

gitudinal study show that all dative constructions where the

indirect object was a pronoun were of the form [NP NP], and

all dative constructions where the indirect object was a lexi-

cal NP, were of the form [NP PP]. For the grammaticality








38

judgment task, Le Compagnon replicated the study by Mazurke-

wich (1981), but she first presented subjects with sentences

with full dative NPs, and a week later, with the same sen-

tences where the full NPs had been replaced by pronouns.

Le Compagnon claims that her results tend to contradict the

results obtained by Mazurkewich (1984a), according to which

dative prepositional phrase complements would be acquired be-

fore double-NP complements: "If this were in fact the case,

then native speakers of French should not have allowed only

criticized indirect object pronouns with English to-datives

since such a theory predicts that they would have previously

learned the to + pronoun form" (Le Compagnon 1984, p.56).

Based on her results, Le Compagnon argues that the acqui-

sition process for English dative verbs is essentially identi-

cal for second language learners, and attributes the errors

made to false assumptions concerning both marked and unmarked

forms in English for which there is positive evidence in the

L1 and L2. This study does not provide any comparison data

between pied piping and preposition stranding.



Hawkins (1987). Markedness and the Acquisition
of the English Dative Alternation by L2 Learners


Hawkins (1987) investigated the acquisition of the English

dative alternation across a wider range of verbs than those

considered by Mazurkewich (1984a, 1984b, 1985) to include 36

verbs (alternating and non-alternating verbs, as well as to-







39
and for-verbs) with a group of 10 French native speakers by

eliciting a grammaticality judgment task and a sentence con-

struction task. Hawkins' results are very similar to those of

Mazurkewich, showing very clearly that the [NP PP] structures

are acquired before the [NP NP] structures for both to- and

fAo-verbs; the acquisition of the dative double object

construction with At-verbs preceding the acquisition of the

double object construction with for-verbs.

Hawkins adapted Le Compagnon's results to include refe-

rence to the distribution of at-and for-verbs: two Ss accepted

hardly any lexical IPs in the [NP NP] construction, whereas

two others accepted lexical NPs in the double object construc-

tion but only with to-verbs. The same two 8s accepted almost

all dative sentences with pronominal PPs, while the other two

accepted them only with foA-verbs. So it would appear that two

8s were making judgments based on the interaction of the dis-

tinction between lexical and pronominal dative NPs with the

distinction between to- and fgo-verbs, whereas the other two

made judgments only on the basis of the distinction between

lexical and pronominal NPs.

Then, combining his results with Le Compagnon's (1984),

Hawkins suggests a three step sequence of acquisition for

French learners of the English dative alternation: the learn-

ers would acquire the English dative alternation "by the pro-

gressive introduction of syntactic features into their

grammars. Learners begin with a broad distinction between







40

pronominal and lexical datives [...], which is later refined

by the introduction of a distinction between to- and Lor-verbs

[...], and later still is refined by the introduction of the

distinction between native and nonnative verb forms" (Hawkins

1987, p.46).

Hawkins claims that learners are guided by the one-to-one

principle: learners assume, until positive evidence to the

contrary, that each surface structure corresponds to a single

underlying representation. In other words, learners assume

that one of the surface constructions is initially prototypi-

cally pronominal and the other prototypically involves lexical

NPs. This may be explained by transfer from the subjects' LI.

Hawkins invokes learnability constraints to explain the

progressive acquisition of the dative alternation: it is pre-

sumably easier to test out one feature at a time when learning

a new construction than to try out the whole set of features.

Hawkins argues that the process of hypothesis testing is refi-

ned on the basis of positive evidence, and that the learning

complexity would define markedness in the dative alternation.

This analysis does not imply that UG is irrelevant in 8LA, but

takes issue with the particular version of UG proposed to ac-

count for the acquisition of the English dative alternation

in studies like those of Mazurkewich (1984a, 1984b, 1985).








41

Tanaka (1987). The Selective Use of Specific
Exemplars in Second Language Performance:
The Case of the Dative Alternation


This study looked at the selective use of give in the two

structures [NP NP] and [NP PP] in a translation task and a

judgment task by 266 Japanese college students within a

framework of transfer and markedness. The subjects were divi-

ded into three English proficiency groups, high, intermediate

and low, based on their score of the cloze test administered

beforehand. The translation test had the following three Ja-

panese sentences typed on separate cards:

a John wa sono shonen ni bike-o ataeta

[John] [the boy] [to] [a bike] [gave]

b John wa sono shonen ni punch-o ataeta

[John] [the boy] [to] [a punch] [gave]

c John wa sono shonen ni chance-o ataeta

[John] [the boy] [to] [a chance] [gave]

The judgment task consisted of six categories with three

items per category:

Category A:

l.a John gave the book to a boy

b John gave a boy the book

2.a Dr. Fox gave the comic books to boys

b Dr. Fox gave boys the comic books

3.a Jerry gave the expensive pen to a girl

b Jerry gave a girl the expensive pen









Category B:

1.a Harris gave a look at the document to Sharon

b Harris gave Sharon a look at the document

2.a Joel gave a turn to kick at the goal to Tom

b Joel gave Tom a turn to kick at the goal

3.a James gave a punch in the eye to Cathy

b James gave Cathy a punch in the eye

Category C:

1.a George gave a headache to Rose

b George gave Rose a headache

2.a Jinmy gave a party to Lucy

b Jimmy gave Lucy a party

3.a Alice gave a kiss to Clifford

b Alice gave Clifford a kiss

Category D:

1.a Overwork gave a heart attack to Bill

b Overwork gave Bill a heart attack

2.a Worry gave a nervous breakdown to Martha

b Norry gave Martha a nervous breakdown

3.a Her blonde hair gave a good appearance to Jane

b Her blonde hair gave Jane a good appearance

Category E:

1.a Thomas gave a kick to the ball

b Thomas gave the ball a kick

2.a Paul gave a pinch to her arm

b Paul gave her arm a pinch







43

3.a Oibson gave a pull to the rope

b Oibson gave the rope a pull

Category 7:

1.a Sharon gave a bike to the small boy

b Sharon gave the small boy a bike

2.a Iris gave some flowers to the girl

b Iris gave the girl some flowers

3.a Harris gave a toy to the baby

b Harris gave the baby a toy

The results of the translation task showed that the stu-

dents used both structures with equal frequency when dealing

with prototypical cases of dative givj, while the [(P PP] was

strongly preferred with cases deviating from the prototype.

The judgment task was designed to investigate three different

constraints: discourse, semantic and perceptual. The results

indicate that the subjects were more sensitive to the percep-

tual than to the discourse constraint, which, according to

Tanaka (1987, p.83) suggests a theory based on scalar marked-

ness (Zobl 1983). Tanaka also claims that language learners

may perceive some items as more marked than others, and points

out that the degrees of markedness as perceived by the lan-

guage learners are an important consideration (Le Compagnon

1984). In addition, these degrees of markedness as perceived

by the language learners may differ from the degrees of mark-

edness as assigned by linguists.








44

White (1987b). Markedness and Second Language
Acquisition: The Question of Transfer


In White (1987b), English native speakers, adult and child

learners of French as a second language, were tested using

grammaticality judgment tasks on preposition stranding and

double object structures. Both structures are considered to

be marked. It was found that only the double object struc-

tures were accepted in French. Preposition stranding was not

transferred to the L2. White concludes that neither the hy-

pothesis that marked structures in the L1 will be a source of

transfer errors in the L2, nor the developmental hypothesis

which claims that all L2 learners will adopt the unmarked case

initially, are supported. However, White cites a number of

other studies where evidence of transfer of preposition

stranding was found, such as Harley (cited as personal commu-

nication in White 1987b). Selinker, Swain, and Dumas (1975)

report a number of production errors from English children in

French immersion programmes:

(1) Un chalet qu'on va aller a (p.146)

Liceras (1983), in a study of adult native speakers of

English learning Spanish, found that half of the beginner sub-

jects accepted preposition stranding; Tarallo and Myhill

(1983), studying the acquisition of relative clauses by native

speakers of English learning various languages, found that

preposition stranding was accepted in certain cases, espe-

cially by those learning German, Chinese and Japanese.








45
Ungrammatical relative clauses with stranding were judged

grammatical by native speakers of English learning various

languages where preposition stranding is not possible. Portu-

guese, for example, allows only pied piping:

(36) 0 homer con quo eu falei era americano

The man with whom I spoke was American

(37) 0 home que eu falei com era americano

The man that I spoke with was American

White suggests that prior knowledge is involved in SLA, "so

that data having triggered a marked form in another language

known to the learner may obscure the correct analysis of the

L2 data" (p.278). It is not clear why one marked structure was

transferred and not the other.

Liceras (1983) also found that native speakers of English

learning Spanish as a second language at the beginner's level

judged preposition stranding to be possible in Spanish, while

students at the intermediate and advanced level did not. In

addition, in an experiment involving grammaticality judgments

by native speakers of English learning French as a second lan-

guage (White 1983), ungrammatical French sentences with the

marked double-object construction (which is found in English

but not in French) were judged to be grammatical both by the

experimental group and by a control group which consisted of

native speakers of various languages, all of whom knew

English.







46
This suggests that the presence of a marked construction in
L1 or some other language that one already knows is indeed a

source of interference in that the subjects did transfer the

L1 property into the L2. This contradicts the developmental

hypothesis of markedness and supports the transfer hypothesis

as stated in White (1987b, p.266). The developmental hypothe-

sis predicts that "learners will acquire unmarked forms as a

necessary developmental stage before the acquisition of marked

forms. The L2 learner will not initially hypothesis marked

forms in the interlanguage, even where the L2 requires marked

forms, and marked forms will not be transferred from the Ll"

(p.266). White reports that it has been argued that second

language learners are less likely to transfer marked forms

than unmarked forms (Adj6mian and Liceras 1984, Hyltenstam

1984, Kellerman 1983, Liceras 1983, Rutherford 1982). Her

results partially support the claim that LI marked forms may

be transferred into the L2, although it remains unclear as to

what syntactic structures may be transferred and whether such

a transfer would depend on formal or psycholinguistic marked-

ness. A formal definition of markedness identifies marked

structures as the structures selected based on direct or

indirect evidence. The unmarked options are selected as such

in the absence of evidence to the contrary. A psycholinguistic

definition of markedness suggests that adult speakers know

what structures are marked in their L1, and that they will

avoid transferring them into the L2.








47

Birdsong et al. (1984). Universals versus
Transfer Revisited

Birdsong et al. (1984), focusing on native speakers of

English learning French as a second language, compared judg-

ment data from a nominal grammaticality task with data from

rank-ordering tasks. Using relative clauses, four types of de-

viances were tested: resumptive pronouns, preposition strand-

ing, resumptive pronoun + preposition stranding, preposition

stranding + relativizer deletion. These four types of sen-

tences are ungrammatical in French.

(a) Resumptive

a. II a cass6 les lunettes don't il EN a besoin

b. He broke the glasses which he needs THEM

(b) Stranding

a. Il a cass6 les lunettes qu'il lit AVEC

b. He broke the glasses that he reads with

(c) Resumptive + Stranding

a. II a cass6 les lunettes qu'il lit AVEC ELLES

b. He broke the glasses that he reads with THEM

(d) Stranding + Deletion of relativizer

a. II a cass6 les lunettes / il lit AVEC

In the first task, it was found that resumptive pronoun

structures are accepted with significantly more frequency than

preposition stranding. So it seems that not only subjects did

not transfer their NL rules, but they appear to accept more

often a rule which exists in neither the Ll or the L2. In the








48

second task, it appeared once again that subjects favored the

IL universal rule of resumptive pronouns to transfer rules of

preposition stranding and preposition stranding + deletion.



Van Buren and Sharwood Smith (1985). The Acquisition
of Preposition Stranding by Second Language Learners
and Parametric Variation


Forty-three university students in their second or third year

of English study at the University of Utrecht, and eighty five

school children who have had one year of English were tested.

In a second round of testing, forty six more second-year

students were given one of the tests administered to the first

round school children. In the first round, the univer-sity

students were presented with a Story Test task which con-

textualized the target sentences in short stories of a few

lines, and a Deaf Man task which asked subjects to judge the

responses of a deaf man who had not understand what he had

been told. Both tasks required subjects to accept or reject

English sentences with prepositions that were both correctly

and incorrectly stranded. Examples are given in (a) and (b):

(a) A(ccept) R(eject) 1) The room was still

A R 2) The inspector asked Mary a few

questions

A R 3) Where had she gone to ?









(b) JILL DEAF MAN

He immediately arrived WHICH SOLUTION DID HE

at the right solution IMMEDIATELY ARRIVE AT ? A R

(van Buren and Sharwood Smith 1985, p.32)

Preliminary results of the university tests show a signi-

ficant difference between the two tasks administered to the

subjects with the Deaf Man task proving to be difficult for

the majority of subjects. In the Story Test there were many

more wrong rejections than wrong acceptance. This may in fact

reflect a bias toward rejection.

School children were administered three subtests. The first

subtest asked subjects to form sentences with prepositions by

choosing words from a pool; the second subtest was a grammati-

cality judgment task involving preposition stranding and pied-

piping; the general result showed a marked preference for

pied-piping (60%); the last subtest was an alternative ac-

ceptability judgment task where subjects indicated stranding

or pied-piping possibilities by rearranging sentences in which

the preposition was unplaced. The overall tendency was also

in favor of pied-piping.

It is concluded that "the overwhelming presence of strand-

ing in the input is not reflected in either the school begin-

ners or the advanced university subjects' behavior and that

there are individual differences which suggest that learners

may have some freedom as to which of the three strategies hy-

pothesized by us they might choose, that is, pied-piping,








50

radical (i.e. 'promiscuous') stranding including R-pronoun

and crosslinguistic stranding (i.e. predominantly R-pronouns

rather than non-R-pronouns)."



Sheppard (1991). At Sea in SLA: Evidence of UG in the
Acquisition of French and English Verbs


In this study 48 native French speakers and 48 native

English speakers were asked to rate the acceptability of a

randomized list of 50 items on a scale from 1 to 3 (1 = total-

ly unacceptable, 2 = uncertainty, 3 = totally acceptable).

Sheppard was essentially interested in three verb types--

believe/croire, promise/promettre, want/vouloir--in three

syntactic environments--fully tensed clausal complements,

Exceptional-Case marking and control structures. The following

examples illustrate the grammatical and ungrammatical exem-

plars used in the study:

(a) *je crois Jean etre. . I believe John to be

(b) *je veux Jean &tre & l'heure I want John to be on time

(c) je crois Jean intelligent *I believe John intelligent

(d) je veux que Jean soit. *I want that John is. .

(e) je promets que je serai. . I promise that I will. .

(Sheppard 1991, p.88)

This study examined four overlapping hypotheses dealing

with markedness, UG and transfer. Some support was found for

the second hypothesis which looked into the "possibility of

an underlying similarity in assumptions about the grammar of







51
the L2" (Sheppard 1991, p.iv). Ambiguous support was found

for the remaining hypotheses which attempted to (1) make

predictions based on markedness differences based on the two

languages, French and English, (2) establish a putative pref-

erence for the French setting of the parameter, and (3) con-

firm an alleged predisposition for L1 transfer.

Sheppard argued that both his groups were relatively suc-

cessful in the acquisition of the proposed parameters noting

that the only exceptions were the failure of the Francophone

group to acquire believe + ECM and the failure of the Anglo-

phone group to successfully acquire croire + PRO. Sheppard

concluded that the Anglophone group transferred 36% of the

items with croire + BCH, and 24% of the items with vouloir +

ECM. In other words, the subjects tested were successful 64%

and 76% each time, respectively.


So from this review of the existing literature, it may be

concluded that the evidence concerning the acquisition of

preposition stranding, Exceptional-case marking, dative alter-

nation and dative passive by English adult learners of French

or Spanish remains contradictory and inconclusive. The only

consensus seems to be that the double object dative structures

are acquired before the prepositional dative structures

(Hawkins 1987, Mazukewich 1984a, Le Compagnon 1984)--but

[NP NP] structures were acquired first when pronominal con-

structions were used instead of full noun constructions.








52

White (1987b) found that English native speakers learning

French as a second language transferred the double object

dative structure into the target language. However, as far as

preposition stranding is concerned, some studies (Bardovi-

Harlig 1988, Birdsong et al. 1984, White 1987b) found that

preposition stranding was not transferred into the second

language (French or English), whereas other studies (Tarallo

and Myhill 1983, Selinker et al. 1975) found evidence of

transfer.

These studies were designed within a framework of marked-

ness or/and transfer. The present study is situated within

two broader perspectives, a learnability perspective and a

parameter setting perspective, which do include the concept

of markedness and transfer, as will be discussed below.



Learnability Perspective

Learnability theory investigates the cognitive principles

that may determine developmental stages and the ultimate suc-

cess attained in language acquisition. In other words, learna-

bility theory addresses the developmental problem and the

logical problem of language acquisition. Cognitive approaches

to learnability in first language acquisition literature have

resulted in many proposals such as the Subset Principle

(Berwick 1985, Berwick and Weinberg 1984), the Uniqueness

Principle (Pinker 1984, 1986), the Principle of Contrast

(Clark 1987), the Competition Model (MacWhinney 1987),







53

the Semantic Bootstrapping Hypothesis (Pinker 1987), and the

Lexical Approach (Roeper, La Pointe, Bing and Tavakolian

1981). Cognitive approaches to learnability are limited to

recent work, e.g. White 1987d, Zobl 1988. With the exception

of Wolfe Quintero (1992) who studies three other learning

principles--cumulative development, continuity and conser-

vatism--the only principles to have been investigated are the

Subset Principle, preemptibility and uniqueness. However, the

Wolfe Quintero study is placed within a learnability perspec-

tive following the O'Grady (1987) framework which differs from

the Universal Grammar framework adopted here.

The learning principle to be investigated here is the

Subset Principle and the predictions that can be made based

on a superset/subset relation. When placed in a situation

where the LI is the superset and the L2 is the subset, it is

predicted that learners will initially continue to use the L1

setting of the parameter until they realize that the L2 actu-

ally only allows a subset of the L1 properties. In other

words, learners may transfer the LI property into the L2 and

create a superset grammar, or they may be able to constrain

their grammar and end up with the proper L2 subset. If second

language learners are guided by a learning principle such as

the Subset Principle, they may proceed based on the positive

evidence available to them which would yield the proper gram-

mar. So the hypothesis to be investigated within a learnabi-

lity perspective is the following:








54
the acquisitional sequence of the syntactic properties

will show that second language learners follow the Subset

Principle.

Let us recall that the hypotheses to be investigated within

the parameter resetting perspective were:

A if the properties do cluster, the acquisition of any

property will trigger the acquisition of the remaining

properties.

B French and Spanish properties will be acquired in the

same way.

Thus the learnability perspective and the parameter resetting

perspective make competing and similar predictions. First,

both approaches predict that the acquisition of the cluster

of syntactic properties will take place based on positive

evidence alone.

a Within a parameter resetting perspective we assume that

it is the positive evidence present in the input that will

trigger the acquisition of the properties subsumed under the

Oblique-Case parameter. The acquisition of any property will

trigger the acquisition of the remaining properties.

Furthermore, once a property has been acquired, the acquisi-

tion of the other properties should follow whether there is

any kind of evidence available, positive or negative.

b Within a learnability perspective, it is claimed that

the learners should entertain the smallest hypothesis compati-

ble with the positive evidence available to them in the input.







55

This would lead the learners to assume that French or Spanish:

allow only prepositional structures for dative

alternation

do not allow the stranding of preposition at the end of

declarative or interrogative sentences

require an animate subject for passive dative sentences

limit instances of Exceptional-Case marking to specific

verbs.

In other words, if the process of acquisition of the L2

learners of French and Spanish is based on positive evidence

alone, these learners should not transfer L1 structures

which are not allowed into the target language.

However, the parameter resetting approach predicts that

the properties will be acquired simultaneously as a cluster,

whereas the learnability perspective breaks up that cluster

by predicting that the properties will be acquired individu-

ally in a particular order, based on the positive evidence

available to them. That particular order of acquisition may

depend on a markedness hierarchy as will be discussed below.

The parameter resetting perspective and the learnability

perspective include the concepts of markedness and transfer

in an interesting way. There are several definitions of

mark-edness. A formal definition of markedness used in work

on language learnability within the framework of generative

gram-mar identifies marked structures as "added on the basis

of direct or indirect evidence, while certain options are







56

selected as unmarked in the absence of evidence to the

contrary by the person acquiring knowledge of the grammar"

(Chomsky, 1981c, p.137). An implicational approach to mark-

edness is based in language typology (Eckman 1977, 1984,

1985; Hyltenstam 1984) which assumes that in a pair of

structures, the marked structure implies the unmarked

structure but the unmarked structure does not imply the

marked one. For instance, preposition stranding which is

marked implies pied piping, its unmarked counterpart, but

the opposite cannot be true. Finally, Kellerman (1983)

argues for psycholinguistic markedness, proposing that adult

speakers know what structures are marked in their native

language, and that they will avoid transferring such marked

structures. A psychological approach to markedness may be

intuitively appealing but it is not defined independently,

whereas the learnability and implicational views of marked-

ness make use of independent criteria.

Let us recall that the parameter resetting and the

learnability perspectives adopted here make a similar

prediction: the acquisition of the cluster of properties

will take place based on positive evidence alone. The

parameter resetting perspective predicts that the acquisi-

tion of any property, marked or unmarked, will trigger the

acquisition of the remaining properties. In contrast, a

learnability perspective predicts that the learners should

acquire the unmarked properties first since they would








57
constitute the smallest hypothesis compatible with the data

(French and Spanish exhibit the unmarked counterpart of the

structures tested here). The role that markedness may play

depends on whether or not one assumes that the second lan-

guage learners are able to ignore their native language to

focus on the target language. Ideally, learners should not

transfer their marked or unmarked structures into the L2 and

should acquire the L2 structures, marked or unmarked, based

on the positive evidence available to them. Within a parame-

ter resetting framework, a markedness scale is irrelevant,

whereas, within a learnability framework, a markedness scale

should determine the acquisitional order.

Chapter 4 restates the hypotheses and predictions made

within both the parameter resetting perspective and the

learnability perspective, and describes the experimental

study which consists in a grammaticality judgment task and a

correction task.













CHAPTER 4
THE EXPERIMlNTAL STUDY


The purpose of this experimental study was to investigate

the acquisition of a cluster of syntactic properties -- prepo-

sition stranding, dative alternation, dative passive and Ex-

ceptional-Case marking -- from a parameter resetting perspec-

tive and a learnability perspective, in an attempt to provide

an insight into the process of second language acquisition.

Let us recall that the hypotheses to be investigated within

these two perspectives make similar and competing predictions.

Both perspectives predict that the acquisition of the cluster

of syntactic properties will take place based on positive evi-

dence alone. However, the parameter resetting approach pre-

dicts that the properties will be acquired simultaneously as

a cluster, whereas the learnability perspective breaks up that

cluster by predicting that the properties will be acquired

individually in a particular order, based on the positive

evidence available to them.

The main question to be answered within a parameter re-

setting perspective is whether the properties do cluster and

if the acquisition of any property will trigger the acquisi-

tion of the remaining properties. In contrast, the question

to be answered within a learnability perspective is whether
58








59
second language learners observe a learning principle such as

the Subset Principle.

Subieots
Forty five learners of French and forty five learners of

Spanish enrolled in language classes at the University of Flo-

rida voluntarily participated in the experiment. The learners

of each language were divided into three proficiency levels

(n=15) based on the class they were enrolled in at the time

of the experiment, and constituted the experimental groups.

Subjects in the first proficiency level (Intermediate I)

were in their second year of study and were enrolled in F2200

or 82200. Intermediate II subjects were in their third year

of study (F3300 or 83300). Subjects in the Advanced group were

in their fourth year of study (F4000 or 84000).

No pre-test screening was conducted, but all participants

were required to fill out a background data sheet to ensure

as much intra-group homogeneity as possible. Subjects who did

not meet the general profile were excluded from the analysis

of the results. This decision was based mostly on their lan-

guage background (See Appendix E).

In addition to the three experimental groups in each lan-

guage, a group of French native speakers (nzl5) and a group

of Spanish native speakers (n=15) participated in the experi-

ment and were used as control groups. All were college educa-

ted.







60

Population attributes
8I: The majority of the population was female (764 of the

non native subjects and 67% of the native subjects).

Aae: The average age for the non native speakers was 20.6

(ranging from 17 to 35), and the average age for the native

speakers was 26.9 (ranging from 18 to 42).

Knowledge of additional languages: Only 21% of the non

native subjects indicated that they knew an additional lan-

guage (besides the language tested in the study) at various

levels of proficiency from beginning to advanced, whereas 87%

of the native speakers reported knowing a third language at

a level of proficiency ranging from intermediate to advanced

(8ee Appendix Z).

Methodoloay
Test condition

The test was administered by the researcher over a three

week period in a computer language learning center at the Uni-

versity of Florida. Subjects participated in the experiment

on a voluntary basis and whenever it was convenient to them.

They were asked to fill out a confidential background

information sheet and were free to withhold personal data and

to withdraw their consent at any time during the experiment.

None did so. They were given written instructions and the

researcher was available to answer any additional questions

subjects had. The same instructions also appeared on the com-

puter screen to make sure subjects understood and were








61
familiar with the instructions before starting the experiment.

The subjects tested took an average of 20 minutes to fill out

the background data questionnaire (see Appendix D) and

complete both tasks.

Instruments

After subjects had filled out the background data sheet,

they were instructed to perform two tasks on an IBM personal

computer.

Task 1: Grammaticality judgment task

The software was written in Turbo Pascal and was designed
to execute the following: at the C> prompt, the subject typed

'RUN' and the program randomized the stimuli, so that the test

items were presented in a different order for each subject.

'Usercode' would prompt the subjects to enter their last name.

This was done to be able to later establish any eventual cor-

relations between subjects' performance on both tasks and sub-

jects' background data. Confidentiality was ensured by number-

ing the subjects. Then, instructions appeared on the screen

informing the subjects that they were going to be presented

with one sentence at a time and that they would be asked to

rate its acceptability/grammaticality from a scale of 1 to 4

(1 = completely unacceptable, 2 = unacceptable, 3 accepta-

ble, 4 = completely acceptable) by pressing the desired key

on the computer key board. The scale was kept at the bottom

of the screen during the whole task as a reminder. Subjects

were also instructed not to 'think too hard', to rely on their







62
first judgment and to perform the task quickly (see Appendix

H). However, they were not told that they were being timed.

As soon as a key (1 to 4) was pressed, the next sentence would

appear on the screen. Subjects did not have the possibility

to go back and change their answers.

Although there are no theory-based predictions on timing,

timing was included in an attempt to find out if students

would take significantly longer to correctly reject ungramma-

tical sentences or to correctly accept grammatical sentences.

Subjects were told that there were no mistakes of spell-

ing, accent, agreement, word choice or tense on conjugated

verbs. This represented an attempt to prevent subjects from

examining sentences for irrelevant elements.

A scale was used since it has been argued that the scaled

judgments of grammaticality are more informative and reliable

than nominal data (Chaudron 1983). According to Birdsong

(1991) the psychological validity of scalar grammaticality

statements is supported by work by Barsalou (1987) and others

in category theory. The category of well-formedness is suscep-

tible to gradedness effects in experimental performance. Sca-

lar responses would thus provide data that more accurately

reflect subjects' intuitions.

Grammaticality judgments are a way of testing whether the

L2 learners' competence includes knowledge of certain forms.

However, grammaticality judgment tasks present a number of

problems, such as response biases on the part of the subjects,








63
who may also be judging sentences based on a semantic criteri-

on instead of a syntactic criterion for example (Birdsong

1989, Kellerman 1985). A way of verifying that subjects are

judging the intended syntactic phenomena is to ask them to

correct the sentences they judged ungrammatical. This proce-

dure is carried out in Task II.

Other sources of data for grammatical theory include

"observation and recording of spontaneous speech, open-ended

and directed interviews, monitoring of spontaneous responses

to sentences under investigation, and imitation (Eubank, in

press; Flynn 1987b), comprehension (Flynn 1986), compositions

(Jordens 1986, Phinney 1987) and manipulation of test senten-

ces [...] the prevailing data base, however, is judgments of

acceptability or grammaticality" (Birdsong 1989, p.72).

So in spite of the fact that the use of grammaticality

judgments is crippled with problems of validity, reliability

and generalizability (Birdsong 1989, p.80), it is the most

widely used in second language acquisition experimental re-

search, as investigation develops more reliable means of

testing learners' linguistic competence.


Task II: Correction task

The purpose of the correction task was to verify that

subjects had applied the relevant syntactic criterion when

rejecting ungrammatical sentences. As subjects were performing

the first task, the program was saving in a subfile all the








64
sentences rated 1 or 2 (i.e. two levels of unacceptability)

by each subject. Upon completion of the grammaticality judg-

ment task, instructions appeared on the screen informing the

subjects that they were going to be presented with unaccepta-

ble sentences. They were asked to rewrite them the way they

thought they should be (see Appendix H). They were not told

that they were being asked to correct their 'own' sentences,

i.e the sentences they had rejected as unacceptable. Since the

stimuli were randomized for each subject, the order in which

the sentences appeared for the correction task was also

different for each individual.



Stimuli
The stimuli consisted of 50 sentences which were instances

of preposition stranding, Exceptional-Case marking, dative al-

ternation and dative passive (see Appendix 0). As mentioned

above, the sentences were automatically randomized by the

program for each subject. The French and Spanish data included

25 grammatical and 25 ungrammatical sentences. All the equiva-

lent English sentences were grammatical.

a Preposition stranding: 10 grammatical sentences with pied

piping, 5 declaratives and 5 wh-interrogatives, 10 ungrammati-

cal sentences with preposition stranding, 5 declaratives and

5 wh-interrogatives. These are exemplified below.

- Preposition stranding: wh-interrogatives

(1) a. Qui Jean donne t-il le livre a?








65
b. ZQuifn da Juan *e libro a?

'Who does John give the book to?'

(2) a. A qui Jean donne-t-il 1e livre?

b. ZA qui6n lI da Juan el libro?

'To who(m) does John give the book?'

- Preposition stranding: declarative

(3) a. C'est la raquette que je joue avec

b. Es la raqueta que juego con

'That's the racket (that) I play with'

(4) a. C'est le ballon avec lequel jj oue

b. Es el bal6n con el cual juego

'That's the ball with which I play'

b ECM : 5 grammatical 5 ungramnatical.

(1) a. Je crois Jean 6tre en retard

b. Creo Juan estar atrasado

'I believe John to be late'

(2) a. Je crois que le president a raison

b. Creo que el president tiene raz6n

'I believe that the president is right'

c Dative alternation: 5 [NP PP], 5 [NP NP]

(1) a. Ils donnent des jouets aux enfants

b. Les dan juguetes a los niftos

'They give toys to the children'
(2) a. Ils donnent les enfants des jouets

c. Dan los niftos juguetes

'They give the children toys'








66
d Dative passive: 5 ungrammatical, 5 grammatical.

(1) a. L'enfant a 6t6 donn6 un jouet

b. El nifo fue dado un juguete

'The child was given a toy'

(2) a. Un message a 6t6 envoy & Jean

b. Un mensaje fue mandado a Juan

'A message was sent to John'

The items were controlled for length, gender of lexical items,

number of syllables, and simplicity of vocabulary. No distrac-

tors or dummy items were introduced because it was believed

that the syntactic properties tested ensured sufficient

variety.

Pilot study

A pilot study was conducted to detect any possible flaws

in the design and/or the stimuli.

Subiects

Nineteen adult speakers of English enrolled in a proficien-

cy-oriented program geared toward all four skills at the Uni-

versity of Florida served as subjects.

Group 1 : Beginning French 5 8s having completed their

second year of study participated.

Group 2 : Advanced French 5 as having completed their third

year (1 8) or fourth year of study (4 Ss) participated.

Group 3 : Beginning Spanish 4 Ss currently enrolled in their

second year of study participated.







67

Group'4 : Advanced Spanish 5 8s having completed their third

or fourth year of study participated.

Method

The first task was a grammaticality judgment task : Ss were

instructed to read the sentences one at a time and to rate

them on a scale from 1 to 5--1 and 2 (completely unaccepta-

ble), 3 (unsure), 4 and 5 (completely acceptable)--without

going back to previous sentences in order to compare them.

They were asked to proceed quickly and to indicate their

ratings on an answer sheet. The second task was a correction

task: the experimenter circled the sentences rejected by the

subject (1 or 2) and asked the 8s to rewrite them the way they

thought they should be without indicating whether the circled

sentences were acceptable or not. The two tasks were adminis-

tered on an individual basis.

Results

The 50 sentences were analysed by category: preposition

stranding, Exceptional-Case marking, dative alternation and

dative passive. The sentences were divided into three types:

sentences which were correctly rejected (CR), sentences which

were incorrectly rejected (UR) and sentences which as failed

to reject (PR).

A Grammaticality judgment task

The results indicated that most of the properties were suc-

cessfully acquired by the learners. No significant difference

was found between beginner and advanced learners.









B Correction task

Proper corrections include items where the relevant correc-

tion was made whether the resulting sentence was correct or

not. For example, the following sentence was counted as

correct because the subject used pied piping even though the

relative pronoun is wrong:

1 c'est la raquette que je joue avec

c'est la raquette avec que je joue

The corrections provided for incorrectly rejected sentences

show three patterns:

in some cases, only minor changes irrelevant to the

structure in question were made such as verb tense or pronoun.

some corrections resulted in ungramaatical sentences,

suggesting that the structure was not understood or simply

that the subjects were unable to provide the proper correc-

tion.

the structure was significantly altered, e.g. a passive

sentence was turned into an active.one, suggesting again that

the structure might not have been understood or that the sub-

jects had detected the ungranmaticality but lacked the meta-

linguistic sophistication to provide the relevant correction.

Or it may simply be that they were more familiar with, and

therefore favored, a specific structure over another structure

such as an active structure versus a passive structure. The

results of the correction task seem to indicate that advanced







69

learners were more successful at providing the appropriate

correction.

The following modifications were made based on the results

of the pilot study:

the scale from 1 to 5 was changed to a scale from 1 to

4, eliminating 3=unsure, which had to be thrown out in the

final analysis because "unsure" cannot reasonably be consid-

ered a mid point between ungrammaticality and grammaticality

both tasks were computerized allowing to randomise the

stimuli for each subject, test up to fifteen subjects at a

time, allow subjects to proceed faster and eliminate the

possibility of human error in saving the data

the instructions were finessed by adding that there were

no errors in spelling, agreement, word choice or tense on

conjugated verbs

a few vocabulary items were changed because some group 1

subjects indicated that they were not familiar with them.


Chapter 5 will restate the hypotheses and predictions made

within the parameter resetting perspective and the learnabili-

ty perspective, and report the results of the grammaticality

judgment task and the correction task.













CHAPTER 5
RESULTS AND INTERPRETATIONS

Let us recall that the hypotheses to be investigated within

the parameter resetting perspective were :

A If the properties do cluster, the acquisition of

any property will trigger the acquisition of the

remaining properties

B French and Spanish properties will be acquired in

the same way

Two peripheral issues related to the parameter resetting ap-

proach will also be treated: the learners'level of proficiency

and the timing of the grammaticality judgment task. First, al-

though the literature offers no principled account, it is of-

ten implied that the performance of the subjects is expected

to improve with their level of proficiency. Second, in spite

of the fact that there are no theory-based predictions on

timing, timing of the grammaticality judgment task was in-

cluded in an attempt to establish a correlation between the

amount of time taken by the subjects to perform the various

tasks, and their performance on these tasks.

And the hypothesis to be investigated within a learnability

perspective was the following:

the acquisitional sequence of the syntactic








71
properties will show that second language learners

follow the Subset Principle.



Analyses

The analyses were performed on an IBM computer using BAS

(Statistical Analysis Software). Various analyses of variance

(ANOVA) were conducted with the derived data and the scalar

data obtained from the grammaticality judgment (OJ) task. The

derived data were obtained by sorting the raw data into gram-

matical sentences and ungrammatical sentences. Sentences were

assigned 1 if they were correctly rated (i.e. rated 3 or 4)

and 0 if they were incorrectly rated (i.e. rated 1 or 2). This

value, 1 or 0, was used as the dependent variable. The scalar

data are the raw data of the ratings (from 1, completely

unacceptable to 4, completely acceptable) subjects assigned

to the stimuli. The percentages for the derived data and the

means for the scalar data are given in terms of correctly

accepted and correctly rejected sentences. Derived data will

be abbreviated to DD and scalar data to SD. The properties

tested are referred to by the following acronyms:

DA = dative alternation

DP = dative passive

ECM = Exceptional-Case marking

PSI = preposition stranding, interrogative structures

PSD = preposition stranding, declarative structures








72

Results of the Grammaticality Judgment Task

Parameter resetting perspective

A If the properties do cluster, the acquisition of

any property will trigger the acquisition of the

remaining properties



Overall results: French and Spanish groups

From a parameter resetting perspective the overall derived

and scalar results of the grammaticality judgment task indi-

cate that the hypothesis A cannot be accepted: the acquisition

of a given property did not trigger the acquisition of the

remaining properties since the five syntactic properties

tested do not cluster. They fall into three highly significant

REGWF (Ryan-Einot-Gabriel-Welsch Multiple F Test) groupings

for correctly accepting grammatical sentences, and into four

highly significant groupings for correctly rejecting ungramma-

tical sentences. The means with different letters (A, B, C,

etc.) are significantly different. The overall percentages

for correctly accepting grammatical sentences are:

DD REGWF grouping Properties % Correctly ACC
A DA 95%
B DP 88%
B PSI 86%
C PSD 76%
C ECM 75%


The overall means of raw data for correctly correctly

accepting grammatical sentences are:










8D RIGNF Grouping Mean N Properties
A 3.7870 568 DA
B 3.7129 519 PSI
C 3.6025 478 DP
C 3.5987 451 BCM
C 3.5646 457 PSD


The fact that, for example DA is placed in one grouping and

DP in a different grouping, indicates that not only subjects

performed differently for each of these properties but also

that their performance for each group, A, B and C was signifi-

cantly different. They performed significantly better on DA

than on the remaining properties and their performance for DP

and PSI was significantly better than for PaD and CMN. If the

acquisition of, for example, preposition stranding, had trig-

gered the acquisition of dative alternation, dative passive

and Exceptional-Case marking, the results would have shown a

single grouping, indicating that no significant difference was

found between the syntactic properties. In that case it would

have been concluded that subjects performed in a similar

manner for all properties. That performance would have allowed

us to claim that the subjects tested were at the same level

of competence for all the properties tested. But the fact that

significant differences were found between the properties in-

dicates that the subjects are acquiring these properties one

by one as opposed to as a cluster. These percentages indicate

that all five properties were successfully acquired by the

subjects tested since all percentages are above the decisive







74

68% (p < .05. See Appendix I) indicating that subjects per-

formed better than chance.

Following the general trend in second language acquisition

literature, we assume that a better than chance performance

may be equated with the acquisition of the properties tested.

Different criteria are used throughout the literature to

establish a better than chance level. Wolfe Quintero (1992,

p.57) states that "there were three test sentences for each

type of extraction (two from Task 1; one from Task 2), and if

subjects were able to produce the target response type at

least two out of three times (67%), they were considered to

have acquired that type. This criterion seems reasonable

considering that subjects had only three chances to produce

each type". Bley-Vroman, Felix and loup (1988) reporting that

chance performance would have been only 25%, conclude that

"nonnative performance was clearly not simply based on

guessing" (p.32). Other studies, such as Broselow and Finer

(1991) simply mention that the subjects performed at a "far

greater than chance frequency" without specifying how such a

conclusion was reached.

The overall percentages for correctly rejecting ungrammati-

cal sentences are:

DD REGNF Grouping Properties % Correctly RUJ
A PSI 96%
A PSD 91%
B ECM 80%
C DA 64%
D DP 49%







75

The overall means for correctly rejecting ungrammatical

sentences are:

SD REGHF Grouping Mean N Properties

A 1.4583 384 DA
A 1.4108 297 DP
A 1.4017 483 ECM
B 1.2344 546 PSD
B 1.2336 578 PSI


We may conclude that there is only weak evidence for cluster-

ing and more so for correctly accepting than correctly reject-

ing sentences. Subjects performed better than chance for only

three properties, obtaining only 64% for DA and 49% for DP,

which is well below the percentage for better than chance. In

addition, it is interesting to notice that, overall, subjects

perform quite differently on the two instances of preposition

stranding, declaratives (PSD) and interrogatives (PSI). A

significant difference was found between PSI and PSD in cor-

rectly accepting (86% vs 76%), but no significant difference

was found for correctly rejecting (96% vs 91%).

The scalar data indicate the relative degree of grammati-

cality the subjects attributed to the various sentences. For

ungrammatical sentences, the more they perceived the sentences

to be ungrammatical, the closer to 1 the means should be. For

grammatical sentences, the closer to 4, the more the subjects

perceived the sentences to be grammatical. It is interesting

to notice that most of the properties tested fall within the

same groupings for the scalar data as they do for the derived








76

data. For example, the derived data indicate that the subjects

correctly rejected 96% of the ungrammatical PSI sentences, and

the scalar data show they attributed a high degree of ungram-

maticality to these sentences since PSI has the mean closest

to 1 (1.23).

To sum up, the acquisition of any property did not trigger

the acquisition of the remaining properties. So it may be

concluded that these overall results do not support parameter

resetting in adult second language acquisition for the pro-

posed instantiations of preposition stranding, Exceptional-

Case marking, dative alternation and dative passive.

Results by groups

By taking a closer look at the results by groups we find

somewhat stronger evidence in favor of the hypothesis within

a parameter resetting perspective: the results of the ANOVA

indicate only two groupings for correctly rejected sentences

for F2200, F4000, S2200, S3300, S4000 and surprisingly also

for SNS. FNS show only one grouping, as expected.

F2200 Internal ANOVA

DD REGWF Grouping % CORR ACC Mean Property
A 100% DA
B A 87% DP
B A 85% ECM
B 79% PSI
B 72% PSD

SD REGWF Grouping Mean N Property
A 3.733 75 DA
B A 3.625 64 ECM
B A 3.525 59 PSI
B 3.463 54 PSD
B 3.430 65 DP








77

Properties fall in two significantly different groupings for

both the derived data and the scalar data. The scalar data

show that the subjects were relatively unsure of the grammati-

cality of the sentences illustrating DP. They correctly ac-

cepted 87% of the agrammatical sentences but the mean rating

is the lowest of all properties.



DD REGWF Grouping % CORR RKJ Mean Property
A 95% PSI
A 95% PSD
A 83% 3CM
B 59% DA
C 33% DP

8D REOWF Grouping Mean N Property
A 1.5682 44 DA
B A 1.4000 25 DP
B 1.3065 62 3Cn
B 1.2394 71 PSI
B 1.2254 71 PSD

The derived data indicate that the properties fall in three

significantly groupings for correctly rejecting sentences,

whereas the scalar data show only two groupings. They per-

formed significantly differently on DA and DP, but they did

not rate them as significantly different.


F3300 Internal ANOVA

DD RIOWF Grouping % CORR ACC Mean Property
A 95% DA
A 93% DP
A 87% PSD
A 83% ECM
A 79% PSI








78

8D REGWF Grouping Mean N Property
A 3.8310 71 DA
B 3.5846 65 PSD
B 3.5763 59 PSI
B 3.5429 70 DP
B 3.5323 62 ECM

No significant difference was found for correctly accepting

the properties.

DD REGWF Grouping % CORR REJ Mean Property
A 95% PSI
A 92% PSD
B A 73% EC3
B C 56% DA
C 40% DP

SD REGWF Grouping Mean N Properties
A 1.5333 30 DP
A 1.4762 42 DA
A 1.4364 55 ECM
B 1.2029 69 PSD
B 1.1690 71 PSI

The properties fall in three significantly groupings for the

first ANOVA and in two groupings for the second ANOVA.


F4000 Internal ANOVA

DD REGWF Grouping % CORR ACC Mean Property
A 97% DA
B A 97% DP
B A 89% ECM
B A 85% PSI
B 85% PSD


SD ROEGR Grouping Mean N Property
A 3.8630 73 DA
B A 3.6875 64 PSI
B 3.6406 64 PSD
B 3.5616 73 DP
B 3.5522 67 EC(

There is a discrepancy between the subjects' performance and

their perception of grammaticality. They successfully accepted







79

97% of the DP but the mean rating is relatively low and closer

to 3 than to 4.


DD REGWF Grouping % CORR REJ Mean Property
A 96% PSI
A 89% PSD
B A 81% ECM
B 64% DA
C 41% DP


SC REGWF Grouping Mean N Property
A 1.5246 61 ECM
A 1.5161 31 DP
A 1.4167 48 DA
B 1.1642 67 PSD
B 1.1389 72 PSI


FNS Internal ANOVA

DD REGWF Grouping % CORR ACC Mean Property
A 100% DA
A 100% PSI
A 99% DP
A 97% PSD
A 95% ECN

SD REGWF Grouping Mean N Property
A 3.9600 75 DA
A 3.9333 75 PSI
B A 3.8904 73 PSD
B A 3.8732 71 ECn
B 3.7838 74 DP

No significant difference was found for correctly accepting

the properties, but the scalar results, which show two sig-

nificantly different groupings, indicate that the French

native speakers attributed a varying degree of grammaticality

to some properties.










DD RBGWF Grouping
A
A
A
A
A

SD REOWF Grouping
A
A
A
A
A


80

% CORR RZJ Mean
99%
97%
96%
92%
89%

Mean
1.1250
1.1216
1.1194
1.1096
1.1014


Property
PSI
DA
ECM
DP
P8D

Property
ECM
PSI
PSD
DA
DP


82200 Internal ANOVA

DD REGWP Grouping % CORR ACC Mean Property
A 92% DA
B A 77% PSI
B A 67% DP
B A 67% PSD
B 61% ECM

SD REGWP Grouping Mean N Property
A 3.6957 69 DA
A 3.6250 40 DP
A 3.6207 58 PSI
B 3.3600 50 PSD
B 3.3043 46 EC1

Two significantly different groupings were found for correctly

rejecting ungrammatical sentences, and the subjects' perceived

degree of grammaticality seems to be a good predictor of their

performance.

DD REGWF Grouping % CORR REJ Mean Property
A 96% PSI
B A 80% PSD
B A C 77% ECK
B C 60% DA


SD REGWP Grouping
A
B A
B C
B C
C


: 52%

Mean
1.7556
1.5641
1.5000
1.3750
1.3000


DP

Property
DA
DP
ECM
PSI
PSD








81

The three groupings indicate that the subjects' performance

was significantly different depending on the property tested.

Once again, the subjects' perceived degree of grammaticality

seems to be a relatively good predictor of their performance.


83300 Internal ANOVA

DD REGWF Grouping % CORR ACC Mean Property
A 93% DA
A 88% DP
A 87% PSI
B 55% PSD
B 52% ECM

SD REGWF Grouping Mean N Property
A 3.6286 70 DA
A 3.5849 53 DP
B A 3.5077 65 PSI
B C 3.3077 39 ECM
C 3.2195 41 PSD

The properties fall in two significantly different groupings

for the derived data and in three significantly different

groupings for the scalar data. Once again, we may say that

the subjects' perceived degree of gramaticality is reflected

in their performance. The higher the percentage is to 100%,

the closer is the mean rating to 4.

DD REGNP Grouping % CORR REJ Mean Property
A 96% PSD
A 96% PSI
B 71% ECM
C 33% DP
C 33% DA

SD REGWF Grouping Mean N Property
A 1.840 25 DA
B A 1.642 53 ECM
B A C 1.640 25 DP
B C 1.444 72 PSD
C 1.389 72 PSI










84000 Internal ANOVA

DD -REGNP Grouping % CORR ACC Mean Property
A 88% DA
A 87% PSI
A 82% DP
B A 67% PSD
B 51% ECM

SD REGWF Grouping Mean N Property
A 3.7692 65 PSI
B A 3.6122 49 DP
B A 3.6061 66 DA
B A 3.5526 38 KCM
B 3.4600 50 PSD

Two groupings were found for both ANOVAs. However, there is

a discrepancy between the subjects' performance and their

rating of the sentences. For example, only 51% of the ECM

instances were correctly accepted, but the mean rating is

relatively high, being closer to 4 than to 3.


DD REGWF Grouping % CORR REJ Mean Property
A 95% PSD
A 95% PSI
B 72% ECM
C 48% DA
C 36% DP

SD REGWF Grouping Mean N Property
A 1.7778 36 DA
A 1.7407 27 DP
B 1.4815 54 ECN
C B 1.2676 71 PSI
C 1.2535 71 PSD

The properties fall into the same number of groupings for both

ANOVAs, and the mean percentages for correctly rejected sen-

tences seem to correspond to the degree of grammaticality

perceived by the subjects.










888 Internal AMOVA

DD REGWF Grouping % CORR ACC Mean Property
A 99% PSI
B A 92% DA
B A 90% DP
B A 85% ECM
B 80% PSD

SD REGWF Grouping Mean N Property
A 3.9730 74 PSI
A 3.9565 69 DA
B 3.7969 64 ECM
B 3.6852 54 DP
B 3.6500 60 P8D

DD REGWF Grouping % CORR REJ Mean Property

A 100% PSI
A 95% DA
A 92% PSD
A 91% ECM
B 68% DP

SD REGkN Grouping Mean N Property
A 1.3088 68 ECM
A 1.2941 51 DP
A 1.2817 71 DA
A 1.1733 75 PSI
A 1.1594 69 PSD

Surprisingly, the properties fall into two significantly

different groupings for correctly rejecting sentences.

To sum up, the best result is obtained by the P3300 group

for which no significant difference was found among all five

properties in terms of correctly accepting sentences (F (4,

70) = 2.35, p < .06); consequently, only one grouping was

found. Not only did that group of subjects seem to have ac-

quired all the properties tested at the time of the experi-

ment, but the mean (87%) indicates that subjects performed

quite well and better than chance for all properties.








84

However, some caution is in order. The mean for the percentage

of correctly rejected ungrammatical sentences is significantly

lower than for correctly accepted grammatical sentences (71%

vs 87%) where three groupings were found for all non native

groups including the F3300 group with highly significant

differences (F (4, 70) = 10.72, p < .0001). Once again, the

results of the SNS group, which does show two significantly

different groupings for correctly rejected sentences, are

surprisingly different from the FNS group.



Rejecting versus accepting

Overall, subjects were much better at correctly accepting

grammatical sentences (mean = 84%) than at correctly rejecting

ungrammatical sentences (mean = 76%). The non native French

groups performed better than chance (68% or better) on accept-

ing sentences for all properties, whereas all three groups,

F2200, F3300 and F4000 failed to perform better than chance

on DA and DP. As far as accepting, the S2200 group failed to

perform better than chance for DP, PSD and ECM, the S3300 and

S4000 groups failed to perform better than chance for PSD and

ECM. As far as correctly rejecting ungrammatical sentences,

all three Spanish groups, just like the French groups, failed

to perform better than chance for DA and DP. In other words,

a comparison between properties for correctly accepting and

correctly rejecting, indicates that even if subjects accepted

a high percentage of DA and DP, they were also unable to








85
correctly reject a great number of the ungrammatical counter-

parts of these properties: 95% vs 56% and 93% vs 40% respec-

tively. This failure to reject ungrammatical sentences

suggests that some subjects may not have been aware, at the

time of testing, that French or Spanish do not allow double

NP structures or passive dative alternation. In other words,

these subjects appear to assume that French or Spanish allowed

the same structures as the LI. That assumption was expected,

especially during the beginning stages of second language

acquisition. So the results for accepting sentences provide

stronger evidence in favor of the parameter resetting hypoth-

esis than do the results for rejecting sentences. However, as

mentioned above, no definite conclusion may be drawn without

a closer examination of accepting versus rejecting and a

careful interpretation of these differences.

Correctly ACC VS Correctly REJ
F2200 DA 100% 59%
DP 87% 33%
ECM 85% 83%
PSI 79% 95%
PSD 72% 95%

F3300 DA 95% 56%
DP 93% 40%
PSD 87% 92%
ECM 83% 73%
PSI 79% 95%

F4000 DA 97% 64%
DP 97% 41%
ECM 89% 81%
PSI 85% 96%
PSD 85% 89%









PNS DA 100% 97%
PSI 100% 99%
DP 99% 92%
PSD 97% 89%
ECM 95% 96%

As stated above, the three French experimental groups were

worse at correctly rejecting ungrammatical sentences for all

properties except for PSI and PSD. The most striking contrast

is for DA (mean correctly ACC = 97% vs mean correctly REJ =

60%) and DP (mean correctly ACC = 92% vs mean correctly REJ

= 38%); so there is a common trend with all three French ex-

perimental groups. Surprisingly, the French native speakers

follow the same trend in that the means for correctly re-

jecting sentences are slightly lower than the means for cor-

rectly accepting sentences.

The Spanish experimental groups' performance is similar to

the French experimental groups' results in that they too per-

formed worse at correctly rejecting than accepting sentences

for DA and DP. However, they performed better at correctly

accepting than correctly rejecting PSI, PSD and ECM (mean CORR

ACC = 55% vs mean CORR RlJ = 73%). They performed quite poorly

with DA and DP. The Spanish native speakers differ from the

French native speakers in that they performed better, not

worse, at rejecting than accepting, except for DP whose mean

of 68% is surprisingly low.








87

Correctly ACC vs Correctly REJ

S2200 DA 92% 60%
PSI 77% 96%
DP 67% 52%
PSD 67% 80%
ECM 61% 77%

S3300 DA 93% 33%
DP 88% 33%
PSI 87% 96%
PSD 55% 96%
ECM 52% 71%

S4000 DA 88% 48%
PSI 87% 95%
DP 82% 36%
PSD 67% 95%
ECM 51% 72%

SNS PSI 99% 100%
DA 92% 95%
DP 90% 68%
ECM 85% 91%
PSD 80% 92%

These facts will be further examined below.


B French and Spanish properties will be acquired in

the same way.

Overall results

First of all, the results of the grammaticality judgment

task show that overall French subjects performed signifi-

cantly better than the Spanish subjects at correctly accepting

grammatical sentences (F (9, 590) = 17.56, p < .0001):

DD REGWF Grouping NNSs and NSs % CORR ACC
A French 90%
B Spanish 78%


SD REGWF Grouping
A
B


NNSs and NSs
French
Spanish


N
1353
1120


Mean
3.6903
3.6241








88
Furthermore, a strong interaction was found between levels of

proficiency and properties (p=2.97, df=28, p < .0001), indi-

cating that the subjects at the various levels of proficiency

performed very differently depending on the syntactic proper-

ties. The same interaction was found for rejecting sentences

even if no significant difference was found between the two

language groups as far as correctly rejecting ungrammatical

sentences (P (9, 590) = 27.77, p < .0001):

REGNF Grouping NNBs and NSs % CORR RUJ

A French 78%
A Spanish 74%

Thus the French groups performed significantly better than

the Spanish groups. The French groups' mean rating for

correctly rejecting ungrammatical sentences is closer to 1,

and the mean rating for correctly accepting grammatical

sentences is closer to 4.

Based on the means for percentage of correctly accepted and

percentage for correctly rejected for each language groups,

and the various groupings for the proposed properties, the hy-

pothesis that French and Spanish subjects would acquire the

proposed cluster of properties in a similar way cannot be ac-

cepted. This is further evidence against the parameter

resetting approach. Not only was little evidence for cluster-

ing found overall, but these results show that the same prop-

erties subsumed under a parameter are not acquired in the same

way in French and Spanish. If the parametric approach were








89

tenable, the acquisition of French and Spanish--languages

alike in the relevant syntactic dimensions--should be similar.

We have seen that this prediction was not verified.

Analysis by sentence sets

a French

1 Dative alternation

SID 38 Marc a envoy un paquet a son cousin

33 Marc a envoy son cousin un paquet

% NNSs F2200 F3300 F4000 FNS

CA 100% 15 100% 15 100% 15 100% 15 100%

CR 58% 9 60% 9 60% 8 53% 15 100%

The sentence (38) was correctly accepted by all the nonnative

subjects but 42% failed to correctly reject its ungrammatical

equivalent, with F4000 doing the worst of all groups since 47%

of the subjects failed to correctly reject (33). These results

seem to imply that 42% of the F4000 groups assume that French

is similar to English in allowing [NP NP] and [NP PP] struc-

tures. They may be overgeneralizing by transferring the double

NP structure into the L2.

SID 36 Ils donnent un jouet a l'enfant

31 Ils ont donn6 l'enfant un jouet

% NNSs F2200 F3300 F4000 FNS

CA 98% 15 100% 14 93% 15 100% 15 100%

CR 53% 6 40% 7 47% 11 73% 13 87%

47% of the subjects overall failed to correctly reject (31)

and did not do quite as well for correctly accepting (36).








90
Even 2 native speakers failed to reject the ungramatical sen-

tence. The worst performance is from the P2200 groups where

only 40% of the subjects correctly rejected the double NP

structure. Only 1 subject from the P3000 group rejected (36)

and changed it to correct in the correction task.


8ID 37 J'ai dit la v6rit6 A Marc

32 J'ai dit Marc la v6rit6

% NNSs P2200 F3300 74000 FN8

CA 98% 15 100% 14 93% 15 100% 15 100%

CR 78% 11 73% 11 73% 13 87% 15 100%

All subjects performed better with this set of dative alterna-

tion. The worst performance comes from P2200 and P3300 where

27% of the subjects failed to reject (32). Once again it is

a subject from the F3000 group who rejected (37) and later

changed it to correct.


SID 40 Jean a offert un vl6o a Pierre

35 Jean a offert Pierre un cadeau

% NNSs P2200 73300 74000 PF8

CA 98% 15 100% 14 93% 15 100% 15 100%

CR 40% 6 40% 6 40% 6 40% 15 100%

All experimental groups performed equally poorly at correctly

rejecting the double NP structure. 60% failed to do so.




University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - Version 2.9.7 - mvs