�

0
0-
-·

In this major experimental study Peter Robinson casts new light on issues of
::s
VI
central concern to second language acquisition theory and pedagogy. His 0
::s
Fundamental Similarity Hypothesis relates such current issues in SLA theory as
the nature of implicit-explicit learning, the role of 'noticing,' and focus-on-form
versus meaning-only-processing, to recent work in mainstream cognitive
psychology, suggesting a reinterpretation of the familiar acquisition/learning
distinction.

"By applying a research design from experimental psychology to natural

language data, Robinson's important study sheds light on central theoretical
issues concerning the implicit, explicit, incidental, and intentional learning of
second and foreign languages. This book will be read with equal profit by those
in applied linguistics and general learning theory."
·Professor Richard Schmidt
University ofHawai'i at Mano-a/

"This book brings together views from cognitive psychology and SLA theory on
the implicit acquisition of grammar rules. The empirical part of this study is the
first attempt, to my knowledge, to put claims from two prominent figures in
these two disciplines (Reber and Krashen respectively) to the test in a single
design."
Professor Jan H Hulstijn
Free University, Amsterdam

Peter Robinson has held positions as lecturer in applied linguistics and second
language acquisition at the universities of Pittsburgh, Hawaii, and Queensland.
Currently he is Associate Professor of Linguistics and Second Language
Acquisitions at Aoyama Gakuin University, Shibuya, Tokyo. He completed his
M.A. in Language and Literature in Education at the University of London

THEORETICAL STUDIES
Institute of Education and his Ph.D. in Second Language Acquisition at the
University of Hawaii. He has published widely in the areas of SLA theory and

IN SECOND
SL pedagogy.

4 LANGUAGE ACQUISITION
LANG
 Consciousness, Rules,
and Instructed Second Language
Acquisition
 Theoretical Studies in Peter Robinson
Second Language Acquisition

Simon Belasco
General Editor
Consciousness, Rules,
Vol. 7
and Instructed Second Language
Acquisition

PETER LANG PETER LANG

New York• Washington, D.C./Baltimore New York• Washington, D.C./Baltimore
Rem• Frankfurt am Main• Berlin• Vienna• Paris Bern• Frankfurt am Main• Berlin• Vienna• Paris
 Library of Congress Cataloging-in-Publication Data

Robinson, Peter.
Consciousness, rules, and instructed second
language acquisition/ Peter Robinson. Table Of Contents
p. cm. - (Theoretical studies in second language acquisition; vol. 7)
Includes bibliographical references and index.
1. Second language acquisition. 2 . Language and
languages-Study and teaching. I. Title. II. Series. Acknowledgements. . . . . . . . . . . . . . . . . . . . xm
Pll8.2.R58 418--dc20 95-38411

Chapter 1: Consciousness, Implicit Learning

ISBN 0-8204-3040-4
ISSN 10 51-6670

And Second Language Acquisition . 1

1 .0 Introduction . . . . . . . . . 1
Die Deutsche Bibliothek-CIP-Einheitsaufnahme 1 . 1 Focus on form in L2 pedagogy 2
Robinson, Peter :
1 . 1 . 1 Amenability . . 3
Consciousness, rules, and instructed second language 1 . 1 .2 Generalizability 3
acquisition/ Peter Robinson.-New York; Washington, D.C./Baltimore; Bern; 1 . 1 .3 Accessibility . . 4
Frankfurt am Main; Berlin; Vienna; Paris: Lang. 1 . 1 .4 Entailability . . 4
(Theoretical studies in second language acquisition; Vol. 7) 1 . 1 .5 Dependent variables in effect of instruction studies 5
ISBN 0-8204-3040-4
1 .2 Consciousness in L2 learning . . . . . . . . . . . . . . 7
NE:GT
1 .2. 1 The acquisition/learning distinction . . . . . . . . 7
1 .2.2 Consciousness as awareness, intention and explicit
knowledge . . . . . . . . . . . . . . . . . . . 9
1 .3 Renewed interest in consciousness in cognitive psychology 12
1 .4 The generalizability of research using artificial grammars . 14
1 .5 Experimental studies of the effects of structured stimuli on
L l learning . . . . . . . . . . . . . . . . . . . . . 14
1 .6 Experimental studies of SLA under different conditions 15
1 .6. 1 Hulstijn ( 1 990). 16
1 .6.2 Doughty ( 1 988) . . . . . . . . . . . . . . . . 17
1 .6.3 Ellis ( 1 993) . . . . . . . . . . . . . . . . . . 19
The paper in this �ook meets the guidelines for permanence and durability 1 . 7 The need for further studies in a second language context. 19
of the Commtttee on Production Guidelines for Book Longevity 1 .8 The aims of the present study . . . . . . . . . . . . . . 20
of the Council of Library Resources.

Chapter 2: Rules, Complexity And Second

Language Knowledge . . . 23
© 1996 Peter Lang Publishing, Inc., New York
2.0 Introduction . . . . . . . . . . . 23
2 . 1 Pedagogic rules and linguistic rules 23
All rights reserved. 2. 1 . 1 The effectiveness of pedagogic rules 24
Reprint or reproduction, even partially, in all forms such as microfilm' 2. 1 .2 Noticing examples and understanding pedagogic rules 27
xerography, microfiche, microcard, and offset strictly prohibited.
2. 1 .3 Two dimensions of rule complexity . . . . . . . . 28
Printed in the United States of America.
2.2 Relating rule complexity, consciousness and learning . . 29
2.2. 1 Claims based on experimental evidence of implicit
learning studies . . . . . . . . . . . . . . . . . 29
v1 Contents Contents vii

2.2.2 Pedagogic implications of the claims . . . . 30 3 .4.4 Attention, salience and grammatical sensitivity 78
2.2.3 Speculation about rule complexity in L2 pedagogy. . 31 3 .5 Summary . . . . . . . . . . . . . . 79
2.3 Second language rule complexity . 33
2.3 .1 Perceptual salience . . . . . . . 33 Chapter 4: Implicit Learning, Memory And
2.3 .2 Lexical rules versus abstract rules 34 81
2.3.3 Size of context and degree of semantic opacity . 35 Knowledge . . . . . .
2.3.4 Processing load and rule complexity 35 4.0 Introduction . . . . . 81
2.3 .5 The consistency of expert judgments in problem 4. 1 Implicit and explicit learning 82
classification and problem solving 36 4. 1 . 1 Evidence of dissociations using artificial grammars . 84
2.4 Empirically motivating the easy/hard rule distinction . 37 4. 1 . 1 . l Implicit task conditions . 88
4. 1 . 1 .2 Rule-search conditions . . . 88
2.4. 1 Identifying unfamiliar structures . . . 38
42 4. 1 . 1 .3 Effects of providing explicit instruction. . 89
2.4.2 Rating scale surveys o f rule complexity . . .
2.4.3 The simple and complex rules identified for the study 44 4. 1 .2 Arguments against dissociations . . . . . . 89
4. 1 .2 . 1 Fragmented memory or abstract knowledge? 90
4 . 1 .2.2 Verbalizability of knowledge and awareness 91
Chapter 3: Implicit And Explicit Information 91
4 . 1 .2.3 Processing accounts of the evidence .
Processing . . . . . . . . . 51 4.2 Implicit and explicit memory . . . . . . . . . . . . 92
3. 0 Introduction . . . . . . . . . 51 4.2 . 1 Methodologies for studying dissociations . . . 94
3 . 1 Resources, informat ion processing and consciousness 51 4.2. 1 . 1 Recognition and recall . . . . 94
3 .2 Attention . . . . . . 53 4.2. 1 .2 Subliminal exposure and preference rating . 94
3 .2. 1 Filter theories of attention . . 54 4.2 . 1 .3 Priming. . . . . . . 95
3 .2.2 Capacity theories of attention . . 54 4.2. 1 .4 Word completion . . . . . . 95
3.2.3 Noticing and attentional theory . 57 4.2.2 Evidence for dissociations . . . 95
3.2.4 Alertness, orientation and detection . 58 4.2.3 Explanations, dissociations and task demands . 96
3.2.5 Attention, awareness and detection . 59 4.3 Implicit knowledge . . . . . . . . . . . 98
3 .2 .6 Attentional and non-attentional learning versus 4.3 . 1 The rule-based abstractionist position. . . 99
encoding specificity . . . . . . 61 4.3 . 1 . 1 Prototypes and abstraction . . 99
3 .2. 7 Summary of the role of attention 64 4.3 . 1 .2 Connectioni st models of categorizatio n 1 00
3 .3 Memory . . 64 4.3 . 1 .3 Natural language and categorizati on . 101
3 .3 . 1 Short-term and long-term memory 65 4.3 .2 The exemplar-based distributive position . . . 101
3 .3 .2 Automaticity and memory . . 66 4 .3 .2 . 1 Estes' s model of classification . . . . . 101
3.3.3 Access to long-term memory . 68 4.3 .2.2 Nosofsky's exemplar-based model of
3 .3 .3 . 1 Activation views . 69 classification. . . . . . . . . 1 02
3 .3 .3 .2 Processing views . 70 4.3 .2.3 Medin and Murphy's theory model of
3.3.3 .3 Multiple systems views . 71 classification . . . . 1 02
3.3 .4 Multiple systems and multiple explanations . 73 4.3 .3 Rules, instances and task demands . 1 02
3.3.5 Automaticity, retrieval and context instantiation . 74 4.3 .4 The Fundamental Similarity Hypothesis 1 03
3.3.6 Summary of the role of memory . . 75 4.3 .5 Summary . . . . 1 05
3 .4 Individual differences, resources and SLA . 75 4.4 The present study . . . . . . 1 05
3.4. 1 Individual differences and conscious learning . 75 4.4. 1 Summary of the research design. . 1 07
3 .4.2 Individual differences and unconscious acquisition 76 4.4.2 Research hypotheses . . . . . 1 07
3.4.3 Short-term memory capacity and learning . 77 4.4.3 Rationale for the research hypotheses . 1 08
viii Contents Contents ix

6.3 Variance data . . . . . . . . . 1 43

Chapter 5: Methodology For The Study. 1 13 144
6.3. 1 Analyses for Hypothesis 5
5 .0 Introduction . 1 13 1 44
6.3 . 1 . 1 Accuracy . . . .
5 . 1 Subjects : : : :
· 1 13 6.3 . 1 .2 Reaction time . . 145
� :: : : : : : : : :
. . . . . . . . . . . ·

5 .2 Data collection procedur s

. . .
1 14 6.3 . 1 .3 Summary of the evidence for Hypothesis 5 . 1 45
5.2.1 Pret�sting and aptitude testing . . . . . 1 15 1 45
6.4 Aptitude data . . . . . . . . . . . .
5 .2.2 Assignment to experimental conditions . 1 16 6.4. 1 Analyses for Hypothesis 6 . . . . . . . . . . . . . 1 49
5.2.3 The implicit learning condition . . 1 17 6.4. 1 . 1 Accuracy and aptitude . . . . . . . . . . . 1 49
5 .2.4 The incidental condition 1 17 6.4. 1 .2 Reaction time and aptitude . . . . . . . . 151
5.2.5 The explicit rule-search con i i n dt� : 1 18 6.4. 1 .3 Summary of the evidence for Hypothesis 6. 151
5.2.6 The explicit instructed condition . 118 6.5 Summary of research hypotheses and results . 151
5.2.7 The transfer phase . . . . . . . . . . . . 1 19
5 .2.8 The post-experimental debriefing session . 1 19
5 .3 Materials . . . . . . . . . . . . . . 1 20
Chapter 7: Further Analyses . . . 1 53
7 .0 Introduction . . . . . . . . . . . . . . . . 1 53
5 .3 . 1 Computer materials . . . . . 1 20
7 . 1 The effect of grammaticality and sentence type 1 54
5.3.2 Pretest and aptitude measures . 1 20
7. 1 . 1 Simple rule grammatical sentence types . 1 56
5.3.3 Rules . . . . . . . . . 121
7 . 1 . 1 . 1 Accuracy . . . . . . . . . . . 1 56
5 .3 .4 The training set sentences 121
7 . 1 . 1 .2 Reaction time . . . . . . . . 158
5 .3 .5 The transfer set sentences 122
7 . 1 .2 Simple rule ungrammatical sentence types 158
5 .3 .6 Debriefing materials . . . 1 23
7 . 1 .2. 1 Accuracy . . . . . . . . . . . . 1 59
5 .4 Analyses . . . . . . . . . . . 1 23
7 . 1 .2.2 Reaction time . . . . . . . . . . 1 59
7. 1 .3 Complex rule grammatical sentence types. 1 60
Chapter 6: Analyses And Results . 125
7 . 1 .3 . 1 Accuracy . . . . . . . . . . . . 1 60
6.0 Introduction . . - 1 25 7 . 1 .3 .2 Reaction time . . . . . . . . . . 161
6.1 Accuracy and reac i n ti t � �� d t :
aa 125 7 . 1 .4 Complex rule ungrammatical sentence types 1 62
6. 1 . 1 Analyses for Hypothesis 1 1 27 7 . 1 .4. 1 Accuracy . . . . . . . . . . . . . 1 62
6. 1 . 1 . 1 Accuracy . . . . 1 27 7 . 1 .4.2 Reaction time . . . . . . . . . . . 1 63
6. 1 . 1 .2 Reaction time · . 1 29 7 . 1 .5 Summary of the effect of grammaticality and
h � d � � f ; H �� h � :
6. 1 . 1 .3 Summary of t e e i e c o y t e is 1 131 sentence type . . . . . . . . . . . . . . . . 1 64
6. 1 .2 Analyses for Hypothesis 2 . . . . . . 133 1 68
6. 1 .2.1 Complex rule accuracy . . . 7 .2 Simple and ,complex rule pretest structures . . . . . .
13J 1 70
6. 1 .2.2 Simple rule accuracy . . . . 7 .3 Awareness and aptitude . . . . . . . . . . . . . . .
134
7.3 . 1 Measures of awareness and the effects of learning
6. 1 .2.3 Complex rule reaction time . 137 condition . . . . . . . . . . . . . . . . . 1 70
6.1 .2.4 Simple rule reaction time . . . . . . . . . 138 7 .3 . 1 . 1 Noticing across conditions . . . . 1 70
6. 1 .2.5 Summary of the evidence for Hypothesis 2. 139 7.3 . 1 .2 Looking for rules across conditions 171
6. 1 .3 Analyses for Hypothesis 3 . . . . . . . . . . . . . 140 7.3 . 1 .3 Verbalizability across conditions . 1 72
�
6 . 1 .3 . 1 S mple and complex rule accuracy . . . . . 140 7.3 .2 Measures of awareness and aptitude . . . . 1 73
6. 1 .3.2 Simple and complex rule reaction time . . . 1 40 1 73
7.3 .2. 1 Awareness and memory . . . . . .
6. 1 .3 .3 Summary of the evidence for Hypothesis 3 . 1 41 1 74
7.3 .2.2 Awareness and grammatical sensitivity
6.2 Rule awareness data. . . 1 41 1 74
7 .3 .3 Summary of the findings for awareness .
6.2. 1 Analyses for Hyp t esi 4 . �h � . : : :: : : : : : : : 143
7.4 Summary of further analyses and results . . . . . . . .
1 77
6.2. 1 . 1 Summary of the evidence for Hypothesis 4. 143
 Contents XI
x Contents

233
condition fam iliarization phase of training.
· · ·

Chapter 8: The Fundamental Similarity Of s . 239

Appendix G: The training set sentence
. . · · · ·
. . .

Implicit And Explicit L2 Learning . 1 79 s and follow up questions

Appendix H: Training set sentence
8.0 Introduction . . . . . . . . . . . . . . . . . 1 79 for implicit, incidental, rule-search and 24 1
8 . 1 Awareness and task demands . . . . . . . . . . 1 79 instructed conditions . · · · · · · · · · ·

ion
8. 1 . 1 Varieties of awareness under different learning
Appendix I: Computer instructions for the familiarizat 249
conditions . . . . . . . . . . . . . . . . . . 1 80
phase of the transfer task ·

8 . 1 .2 The sensitivity of measures of awareness . . . 1 82 25 1

8.2 Learning at different levels of the dependent variable. 1 84 Appendix J: The transfer set sentences
253
8.2.1 The level of rules . . . . . 1 85 AppendixK: Subject information
8.2.2 The level of grammaticality . . . . . . . . . 1 86 257
8.2.3 The level of type . . . . . . . . . . . . . . . 1 87 Bibliography .
8.3 Explanations of the effects of grammaticality and type on 289
Index .....
learning . . . . . . . . . . . . . . . . . . . . . . 1 88
8.3 . l Learning in the instructed condition. . . . . . . 1 88
8.3 . 1 . 1 Confirmation bias and grammaticality . 191
8.3 .2 Learning in the implicit condition . . . 1 92
8.3.3 Leaming in the incidental condition. . 195
8.3 .4 Learning in the rule-search condition . 1 97
8 .4 Summary . . . . . . . . . . . . . . . . . 1 98

Chapter 9: Summary And Conclusions 1 99

9.0 Introduction . . . . . . . . . . . . . . . . . . . . 1 99
9.1 Implicit and explicit second language learning . . . . . 1 99
9. 1 . 1 Second language learning and rule complexity . 200
9 . 1 .2 Generalizing from Reber' s claims about implicit
learning to Krashen's claims about acquisition . 203
9.2 Limitations ofthe study and directions for future research 205
9 .3 Summary of findings . . . . . . . . . . . . . . . . . . 208

Appendix A: Grammaticality judgment test as input to the

rating scale survey of rule complexity. . 211
Appendix B: Rating scale survey of rule complexity . . 215
Appendix C: Computer instructions for the implicit
condition familiarization phase of training. 223
Appendix D: Computer instructions for the incidental
condition familiarization phase of training. 227
Appendix E: Computer instructions for the rule-search
condition familiarization phase of training. 23 1
Appendix F: Computer instructions for the instructed
 XIV Acknowledgements
the main stud� . included resid ent .students at Tokai University at
.
Honolulu and VlSltors from Tokai Umversity at Tokyo, as well as others
from Kansai Gaidai Junior College and the Japan America Institute of
Management S cience, both in Hawa i ' i Kai, and students in the
University of Haw�i'i at Manoa's NICE program. I am also grateful to
the SEAMEO. Regional Language Centre in Singapore for awarding me
the �ell_?wsh�p that allo�ed me time to prepare this m anuscript for
pubhcat10n. Fmally a special thank you goes to Misook Kim for her help
and support throughout the busy time when this was written.
I gratefull� acknowl �dge the permission of the following to
reproduce copyright m aterial: Language Learning for permission to
reproduce parts of Chapter 3 which originally appeared in 'Attention
memory and the 'noticing' hypothesis', Volume 45, no. 2, pp. 283-3 3 1
_ .
; 1.0 Introduction
Studles m Second Language Acquisition for permission to reproduce
parts of Chapters 2 and 8 which originally appeared in 'Leaming simple
and complex second language rules under implicit, incidental, rule­
search and instructed conditions', Volume 1 8, no. 1 , pp. 27-67; Prentice
Hall Inc., Englewood Cliffs, NJ., for permission to reproduce the figure
'A capacity model of attention' from p. 1 0 of D. Kahneman, ( 1 973)
Attention and effort; Academic Press Inc., New York, NY., for
permission to reproduce the figure 'Multiple resources in attention' from
p. 69 of C. Wickens, ( 1 984) 'Processing resources in attention' in R.
Parasuraman and D. Davies (Eds.), Varieties of attention; and The
American Psychological Association for permission to reproduce the
figure 'A fmite-state grammar' from A. Reber 'Implicit learning and tacit
knowledge' (1989) Journal of Experimental Psychology: General, 118,
2 1 9-235 .
Chapter 1
Consciousness, Implicit Learning And
Second Language Acquisition
Recent debate in second language acquisition (SLA) theory has
focused on the role of consciousness in second language (L2)
development. Schmidt ( 1 990, 1 993a, 1 993b, 1 994a, 1 995 ; Schmidt
& Frota 1 986) has claimed that conscious awareness of the form of
input at the level of 'noticing' is a necessary condition for language
development to occur. A number of other S�� r�searc�ers have
also argued for a role for 'consciousness-raismg (Elhs l 9 9 3 a,
1 993b; Fotos & Ellis 1 99 1 ; Rutherford 1 987; Sharwood S mi�h
1 98 1 ; Yip 1 994), 'input-enhancement' (Doughty � 99 1 ; Elhs
1 994b, 1 995; Sharwood Smith 1 99 1 , 1 993; White, Spada,
Lightbown & Ranta 1 99 1 ), 'focus on form' (Long 1 99 1 , 1 994;
.
Long & Crookes 1 992), 'priming' (Mellow & Cummmg 1 994) �d
'analytic teaching' (Lyster 1 994) activities in L2 classrooms :which
selectively direct learner attention to form as an �id to
internalization of the L2 system, and some have based rat10nales
for form-focused instruction on models of the interface between
conscious and unconscious forms of L2 knowledge (Bialystok 1 97 8 ,
1 98 1 , 1 990; Ellis 1 993a, 1 994b, 1 995; Hulstijn 1 990b; cf. Mahle &
Raupach 1 989; Odlin 1 986; Robinson 1 99 3 a). . .
In part, the current interest in the role of consc10usness m L2
.
learning has been prompted by critical reaction to the Momtor
. _
Model of Krashen which proposes that two distmct proces�es
operate in language development, a conscious process of deduction
based on rule application which results in a learned syst�m, and an
unconscious process of induction which results in an acqmr�d system
(Krashen 1 978, 1 98 1 a, 1 982, 1985, 1 994; Krashen & Sehg�r 1 97 5 ;
Seliger 1 975). The l atter system is claimed to be responsi�le for
most language production since the conditions under which the
learned system can operate successfully seldom arise. In addition,
the development of the learned system is restricted to a relatively
2 Consciousness, implicit learning and SLA
small number of simple rule-governed domains. By implication, the
development of the acquired system is a priority which
methodology based on Krashen's theory, e.g., the Natural Approach
(Krashen & Terrell 1983), seeks to foster by avoiding form-focused
instruction and the provision of target language rules.
In contrast, those who argue for a constitutive role for
consciousness specify that methodology should contrive the
opportunities for learners to focus on form and consciously notice
features of the target language input and some recent studies have
examined the effects of design features of language teaching
materials that attempt to manipulate the extent of noticing
(Alanen 1 995; Doughty 1 988, 1 99 1 ; Ellis 1 995 ; Fotos 1 993 ; F otos
& Ellis 1 99 1 ; Hulstijn 1 990a; Jourdenais, Ota, Stauffer Boyson &
Doughty 1 995; Leeman, Arteagoitia, Fridman & Doughty 1 99 5 ;
Loschky & Bley-Vroman 1 993; Lyster & Ranta 1 995; Rankin
1990; Schmidt 1 995; Shook 1 994; VanPatten & Cadierno 1 993 ;
Watanabe 1 992; Zalewski 1 993). The issues of what forms to focus
on and of how to achieve this focusing of attention are currently
the subject of much debate. The present study addresses both o f
these issues by examining claims arising within cognitive
psychology and SLA research that: (i) complex stimulus domains
are most effectively learned nonconsciously; and that (ii) the
beneficial effects of instruction in rules, and of conscious rule­
search, are limited to cases where the rules describe relatively simple
patterns of structural covariance. This chapter first summarizes
suggestions that have been made within SLA research about which
forms to focus on in L2 pedagogy. Following this, the debate over
the role of consciousness in L2 learning is described and related t o
the investigation o f similar issues i n mainstream experimental
psychology, particularly those studies that have used artificial
grammars as a stimulus domain to be learned under implicit and
explicit conditions. Finally, those studies in the field of SLA that
have examined the implicit learning of natural L2s are reviewed.
1.1 Focus on form in L2 pedagogy
Four concerns could be said to underlie suggestions that have been
made about which forms to focus conscious awareness on in
classroom language learning: (i) amenability, the need to focus on
those forms which are best suited to, or most amenable to explicit
presentation, leaving those less amenable to such presentation t o
incidental o r i mplicit learning processes; (ii) generalizability, the
need to focus on those forms where instruction is maximally
Consciousness, implicit learning and SLA 3
generalizable to other related forms in the interest of increasing rate
of acquisition; (iii) accessibility, the need to focus on those forms
which lead to restructuring or resetting of parameters by making
negative evidence accessible to the learner; and (iv) entailability,
the need to focus on one of those forms which cluster together as a
consequence of a particular parameter setting, and which entail each
other, in the expectation of triggering the acquisition of the whole
cluster of properties.
1.1.1 Amenability
There have been claims that the number of rules that are explicitly
teachable is small and limited to those describing structures that do
not require complex processing operations. Some have argued that
the structures most amenable to explicit instruction are those which
have transparent form-function relationships, in which, for
example, morphological additions or word order permutations have
a clearly identifiable semantic function (see Ellis 1 990:1 67 ; Krashen
1 982:97; Pica 1 983, 1 985, 1 994:66). Some have also argued that
amenability, and processing difficulty, are developmentally
scheduled. For example, where complex permutations in word order
are concerned, it has been claimed that explicit form-focused
instruction cannot break the stage-wise order of acquisition
consequent on the learner shedding grammatical processing
strategies. However, if timed to focus on structures characterizing
the learner's current, or next stage of development then form­
focused instruction can be effective (Ellis 1 989, 1 990, 1 994b,
1 994c; Larsen-Freeman & Long 1 99 1 ; Meisel, Clahsen &
Pienemann 1 98 1 ; Pienemann 1 989, 1 992; Pienemann, Johnston &
Brindley 1 988).
1.1.2 Generalizability
There have also been claims that markedness relations, based o n
implicational universals (see Doughty 199 1 ; Eckman 1 977, 1 98 5 ,
1 99 1 ; Gass 1 979; Hawkins 1 988; Pavesi 1 986; Robinson 1 994b;
Rutherford 1 986, 1 987; Schachter 1 990; Towell & Hawkins 1 994;
White 1 989b; Zobl 1 983, 1 985), hold potential for decisions about
sequencing the items to be focused on (Ellis 1 993a, 1 994c; Larsen­
Freeman & Long 1 99 1 :309-3 1 5; Long 1 988, 1 99 1 , 1 994; Long &
Crookes 1 992) since instruction focused on some marked forms
(e.g., relativization on objects of comparison) has been shown t o
generalize to acquisition of less marked forms (e.g., relativization
on subjects) but not vice versa, i.e., instruction on less marked
4 Consciousness, implicit learning and SLA
forms does not generalize to more marked forms (Eckman, Bell &
Nelson 1988).
1.1.3 Accessibility
Within the framework of Universal Grammar (Chomsky 1981,
1986; cf. Bley-Vroman, Felix & Ioup 1988; Cook 1993; Eubank
1991; Felix 1991; Flynn 1987a, 1987b; Towell & Hawkins 1994;
White 1989a), it has been suggested that candidates for form­
focused instruction should include aspects of the target language
which are inaccessible to the learner from the evidence of positive
examples of grammatical L2 structures alone. Rather, negative
evidence of ungrammaticality needs to be made available to learners
in order to disconfirm inappropriate hypotheses about L2
parameter settings and subset, superset relations in the native and
target language that may arise when based only on positive
evidence. That is, when the L 1 is a superset of the L2 with regard to
the structures allowed by a particular parameter setting, positive
evidence of the L2 alone will not disconfirm a projected similarity
between the L l and L2. In these cases access to negative evidence
of what is not possible in the L2 is necessary to limit the scope of
the projection from the Ll (Trahey & White 1993; White 1992).
Opponents of this view argue negative data cannot initiate
reorganization of the L2 grammar since attending to it requires a
conscious learning process different in kind from the unconscious
process responsible for the acquisition of competence (Krashen
1985; Schwartz 1986, 1992, 1993). They argue that, as in L l
development (Atkinson 1992; Goodluck 1991; O'Grady 1993), all
the L2 learner needs access to is positive evidence of L2 structure.
1.1.4 Entailability
A related motivation for a focus on form ansmg within the
framework of Universal Grammar is that of entailability. While
accessibility, as I have described it above, has to do with the nature
of the evidence of structure that the learner must have access to in
order to fix appropriate parameters in the L2, entailability refers t o
the differential consequences of focusing on individual forms. It has
been suggested that a focus on selected items can not only trigger
the realization that a particular parameter applies in that language
but also lead to the simultaneous acquisition of a cascade of related
features that are entailed by the parameter setting (Hilles 1986;
Lightfoot 1989, 1991; Schwartz 1993; Towell & Hawkins 1994;
White 1989a, 1991, 1992). How much new knowledge is triggered
Consciousness, implicit learning and SLA 5
in this way will depend on the degree of configurational similarity
between the L l and L2 concerned. Entailability and generalizability
are similar to the extent that they are both ways of maximizing the
reorganizational consequences of a focus on form. They differ in
that entailability makes reference to a notion of inheritance of
properties specific to a particular linguistic theory (Chomsky 1981,
1986), while generalizability is based on implicational relations
derived from descriptions of language typology (Greenberg 1966).
1.1.5 Dependent variables in effect ofinstruction studies
A growing body of SLA research addresses the effect of a focus on
form on language acquisition in instructed settings, using structures
motivated by the accessibility and entailability arguments reviewed
above as dependent variables (e.g., Felix & Weigl 1991; Schwartz &
Gubala-Ryzak 1992; Trahey & White 1993; White 1991, 1992;
White et al. 1991). Some research has also studied the effect of
instruction using typological markedness and the notion of
generalizability to motivate the dependent variable (e.g., Doughty
1991; Eckman et al. 1988; Gass 1982; Pavesi 1986; Zobl 1985).
The procedure followed in these studies has usually been to contrast
the performance of one or more instructed groups with a control
group to identify differences in the effectiveness of the
instructional treatments (see Ellis 1994b: eh. 14; Larsen-Freeman &
Long 1991: ch.8; and Long 1983, 1988, and for summaries). While
such studies have demonstrated important effects supporting a role
for form-focused instruction they have, however, largely been
conducted under classroom learning conditions. Consequently, given
the lack of experimental control in such studies, it is unclear what
learning processes underlie the aggregate advantage of instructed
groups. Krashen (1985) has argued that this advantage is due not t o
the content of the lesson, and the formal instructional treatment,
but to the fact that L2 classrooms provide acquisition-rich
environments and a supply of comprehensible input which activates
the unconscious acquisition process, and it is this that is responsible
for the apparent advantages of instruction. Long (1988), arguing
against this position, cites studies that show positive effects for
instruction even in contexts where learners are exposed t o
acquisition-rich environments outside the classroom. Recent studies
appear to support the value of a focus on form in facilitating rate of
learning, and the acquisition of communicatively redundant features
of the L2 which are not learned despite considerable exposure t o
comprehensible input in acquisition-rich immersion environments
 6 Consciousness, implicit learning and SLA
(Day & Shapson 1991; Harley 1989, 1993; Harley & Swain 1984;
Lyster 1994; Spada & Lightbown 1993), but clearer support can
only come from controlled experimental studies showing superior
performance by explicitly form-focused instructed groups relative
to groups that process the dependent variables of interest implicitly
or indirectly.
The debate about the role of consciousness has proceeded,
however, without much empirical evidence from controlled
experimental studies of the differential effects of L2 learning under
implicit, explicit and incidental conditions with regard to any of the
above described motivations for a focus on form. Where
experimentally controlled studies of the effect of explicit
instruction exist (e.g., Robinson & Ha 1993), they often lack
equivalent incidentally or implicitly exposed groups, making it
impossible to draw conclusions from these studies about the
differential effectiveness of explicit conscious attention to form
versus incidental or implicit processing of the forms concerned. An
exception is Doughty (1988, 1991), who studied the acquisition of
relative clauses in the implicational order (the 'Noun Phrase
Accessibility Hierarchy') identified by Keenan and Comrie ( 1977)
as a function of different conditions of exposure. Her study thus
dealt with the generalizability issue of form-focused instruction (see
section 1.6.2).
Similar studies, motivated by the accessibility and entailability
issues and examining the extent of learning under different
conditions of exposure, are lacking in the field of SLA. There are
no experimental studies, to date examining the differential
effectiveness of exposure to positive and indirect negative evidence
of L2 parameter settings under nonconscious, implicit conditions,
versus conscious instructed conditions. There is a similar lack of
experimental studies using performance on supposedly amenable
versus less amenable rules as the dependent variable under different
conditions of exposure. No doubt this is partly because distinctions
between structures of different levels of complexity, which are
supposedly differentially amenable to explicit instruction, have been
difficult to operationalize in an uncontroversial way. Thus, claims
by Krashen (1982, 1985) and Ellis (1990) that simple structures
alone are learnable through explicit instruction (see Chapter 2) have
gone unexamined.
Studies in mainstream psychology, in contrast, have addressed
the issue of the differential effectiveness of learning under explicit
and implicit conditions, and have been claimed to reveal significant
Consciousness, implicit learning and SLA 7
effects for learning the structure of complex stimulus domains
without conscious awareness. However, these studies have
predominantly made use of artificial grammars as the stimulus
domain (e.g., Reber 1967, 1969, 1976, 1989, 1993; Reber & Allen
1978; Vokey & Brooks 1992, and see Chapter 4). The
generalizability of their findings to the domain of natural L2
learning is something which must be demonstrated by SLA research.
Engaging in such research is likely to provide a clearer evidential
base for the speculations of SLA theorists regarding the extent t o
which unconscious learning of L2s is, or is not, possible.
1.2 Consciousness in L2 learning
Recent theoretical arguments supporting a role for consciousness
and focus on form in L2 learning (Bialystok & Sharwood Smith
1985·' Cohen Larsen-Freeman & Tarone 1991; Ellis 1990, .
1993a,
' & Pienem ann
1994b, 1995; Harley & Swain 1984; Lightbown
1993; Long 1991, 1994; Long & Crookes 1992; Rutherfo rd 1987;
Sharwood Smith 1981, 1992, 1993; Schmidt 1990, 1993a, 1993b,
1994a, 1995), have been made, in part, in reaction to claims that
L2 development is a largely unconscious process. Kra�hen � 19 � 1a,
1982, 1985, 1993, 1994), in particular, is associated with this view,
maintaining that two processes operate in language learners. The
first process, which is consciousness dominated, he te�s 'le��� ng ­ :
The second process, which is unconscious, he terms acqms1t10n .
These two processes, he claims, result in the development of
distinct, non-interfaced language systems.
1.2.1 The acquisition/learning distinction
Krashen's distinction though defended by some (e.g., Schwartz
1986, 1993; Zobl 1992, 1995) and assumed by others (e.g., Paradis
1994), has been extensively criticized in the SLA literature (Gre�g
1984 1988· Larsen-Freeman & Long 1991: 245-249; McLaughlm
:
1978 1987, 1990a, 1990b; Munsell & Carr 1981). In parti�ular,
McLaughlin (1978), in an early response to Krashen, has pomted
out that it is impossible to know with certainty whether L2 learners
are operating via consciously accessed 'rule' or via subcons�iou�ly
determined intuitions or 'feel,' in making grammaticality
judgments. Yet Krashen has claimed that this distinction reveals
which system is being drawn on-conscious knowledge of . the. �le
guiding judgments based on the learned system, but feel or mtmt10n,
without conscious access to rules, guiding judgments based on the
acquired system (McLaughlin 1987:21). Further, McLaughlin
8 Consciousness, implicit learning and SLA
(1978), and Gregg (1984) both recount evidence, based on their own
experiences, of something which was initially learned, and
consciously and effortfully produced, eventually being produced
without conscious effort, thus contradicting Krashen's claim that
what is learned cannot become part of the acquired system.
Krashen's counterargument to claims of this type, that the acquired
system had eventually caught up with the learned system· and was in
effect responsible for the effortless production, is, as McLaughlin
(1987) and Gregg (1984) both point out, unfalsifiable.
Another problem with the acquisition/learning distinction, as
formulated by Krashen, is the fact that it is impossible to reproduce
these two conditions experimentally in order to study the extent of
learning that takes place under them since the task conditions which
Krashen asserts must be met for 'acquisition,' or 'learning' to occur
are impossible to manipulate experimentally. The main reason is
that Krashen claims the 'affective filter' must be lowered in the
acquisition condition:
The affective filter is a mental block that prevents acquirers from fully
utilizing the comprehensible input they receive for language acquisition.
When it is 'up' the acquirer may understand what he hears and reads, but
the input will not reach the LAD. This occurs when the acquirer i s
unmotivated, lacking i n self confidence o r anxious (Krashen 1985:3).
While it is possible to manipulate variables like intention or
attention through task instructions, it is difficult both ethically and
within an experimental setting to manipulate anxiety and other
components of the 'affect' construct, even if the subtle differences
Krashen assumes are causal, i.e., between 'low' and 'moderate'
anxiety (1985:25), could be quantified. It is true that there are
constructs for measuring anxiety, for example the Taylor Manifest
Anxiety Scale (see Krashen 1981a: 30), but although Krashen
reports that moderate anxiety, as measured by this scale, was
positively correlated with achievement in Spanish as a second
language (in a study by Chastain, 1975), no particular point on the
scale is identified by him as distinguishing low from moderate
anxiety. Also, as Larsen-Freeman and Long (1991:243) point out,
it isn't clear to what extent other factors supposedly regulating the
lowering and raising of the affective filter, i.e., high motivation and
high esteem, can offset the problem of excessive anxiety, and vice
versa.
Consciousness, implicit learning and SLA 9
The pedagogic consequences of Krashen's theory have appeared
to be a reluctance to engage in formal grammar teaching or a focus
on form in isolation from meaning-focused activities. The aim of
the Natural Approach (Krashen & Terrell 1983) is to create the
conditions for facilitating unconscious 'acquisition' and minimize
the time spent on conscious learning of formal properties of the
language. Pedagogic priority is given to ensuring the provision of
comprehensible input to learners, often in the form of reading
(Elley 1991; Krashen 1993), and engaging them in activities which
lead them to process the input. It is argued that vocabulary is
acquired incidentally through engaging with comprehensible input
during such activities (Dupuy & Krashen 1993; Elley 1991; Krashen
1989, 1993, 1994; Pitts, White & Krashen 1989). This process of
incidental acquisition is argued to be preferable to, and distinct from,
that induced by conscious efforts to learn, following direct
vocabulary instruction (Dupuy & Krashen 1993). It is further argued
that the incidental acquisition of vocabulary via meaning-focused
activities takes place as a result of focusing on key vocabulary items
that are known, and inferring the meanings of unknown words on
the basis of this knowledge together with knowledge of the global
context (Krashen, Terrell, Ehrman & Herzog 1984). This
inferential process will lead not only to vocabulary development,
but will also trigger the acquisition of the formal properties of
syntax and grammar in the absence of conscious awareness of such
properties. The influence of Krashen's theory on the thinking of
Prabhu (1987) and the conceptualization which underlay the
development of the Bangalore 'Procedural Syllabus,' particularly
the acquisition/learning distinction and the characterization of
acquisition as a "subconscious" process, is obvious:
Task-based teaching operates with the concept that, while the conscious
mind is working out some o f the meaning-content, a subconscious part of
the mind perceives, or acquires (or recreates as a cognitive structure) some
of the linguistic structuring embodied in those entities, as a step in the
development of an internal system o f rules (Prabhu 1987 :69�70).
1.2.2 Consciousness as awareness, intention and explicit knowledge
Schmidt (1990, 1993a, 1993b, 1994a, 1995), in contrast, questions
the extent to which unconscious processes can result in language
learning and distinguishes between three different senses of the term
'conscious' as it is used in cognitive psychology and SLA theory.
The first of these identifies consciousness with the contents of
10 Consciousness, implicit learning and SLA
awareness, which can be at the levels of perception, noticing and
understanding. Consciousness in the sense of awareness at the level
of perception, in which attention is allocated to signal detection
(see Tomlin & Villa 1994 and Chapter 3) is necessary for all
learning. Perception is also necessary for noticing to occur, and
noticing, which involves rehearsal in short-term memory (see
section 3.2.5) Schmidt argues, is a necessary condition for language
learning. Consciousness, in the sense of awareness at the level of
rule understanding, he argues, based in part on his own experience of
learning Portuguese (Schmidt & Frota 1986), usually accompanies,
and is strongly facilitative of, attempts to learn. The second sense
of consciousness that Schmidt identifies equates consciousness with
intention. In this sense consciousness is not necessary to language
learning since incidental learning is possible. Finally Schmidt notes
that consciousness is also often equated with the contents of
explicit knowledge. However, this term is itself ambiguous (see
Mohle & Raupach 1989; Odlin 1986; Robinson 1993a, 1994a,
1995d) and its senses are perhaps best illustrated by examining the
meanings of the counterpart term, implicit knowledge. There are
three senses in which the term 'implicit' is used to describe
knowledge, and so distinguish it from an explicit counterpart.
(i) Implicit knowledge as innate. Implicit knowledge can refer t o
the innate constraints on a possible language which, some have
proposed, learners are alert to in learning their first (e.g., Chomsky
1986; Lightfoot 1989, 1991; Pinker 1984, 1989, 1994), and
possibly second languages (White 1989a, 1990), though the form of
such constraints is claimed to be unavailable to consciousness and
cannot, therefore, be articulated by language learners (Chomsky
1990).
(ii) Implicit knowledge as automatized. Implicit knowledge can
also be used as a term to refer to automatized knowledge, meaning
knowledge which is accessed with relatively little attentional effort
either as a result of practice (McLaughlin 1987; Shiffrin &
Schneider 1977; Strayer & Kramer 1990), or as a result of the
strength of individual representations in memory (Logan l 988a;
Logan & Stadler 1991; Robinson & Ha 1993; Rothkopf & Dashen
1995). In this sense procedural, in contrast to declarative knowledge
(Anderson 1983), could also be said to be implicit in that it is
accessed with little attentional effort and lies beyond conscious
control.
Consciousness, implicit learning and SLA 11
(iii) Implicit knowledge as the result of implicit learning. In yet
another sense implicit knowledge refers to the knowledge base
established by learning under implicit conditions, as in Reber (1989).
In this sense implicit knowledge of a domain is the consequence of
processing information without conscious awareness of the rule­
governedness, or structure of the domain, and without an intention
to learn the rules regulating the structure. This sense of implicit
knowledge is different from the first sense above in that the
knowledge that results from implicit learning need not invoke an
innately prespecified component to explain its nature or form,
"...there is, at this juncture, no reason to place any priority on
particular biological determinants of a specific kind. All forms of
implicit knowledge are taken as essentially similar at their deepest
levels" (Reber 1989:219). This sense of implicit knowledge is also
claimed to be different from the second sense in that the knowledge
which results from implicit learning generalizes to contexts beyond
those represented by the specific stimuli experienced during
learning. In contrast, automatized knowledge develops as a function
of practice in constant environments (Logan 1988a, 1988b;
Shiffrin & Schneider 1977; Strayer & Kramer 1990). Similarly,
Anderson's distinction between procedural and declarative
knowledge concerns skill acquisition, about which "implicit learning
says little or nothing" (Reber 1993:15). Nonetheless, despite
making a distinction between implicit knowledge as the result of
implicit learning, and implicit knowledge as automatized, Reber
concedes that the two display similar properties: "Automatic
processes are classic examples of the actions of implicit systems, "
occurring outside of consciousness and conscious control, and
requiring minimal attentional effort (see Reber 1993:15).
Schmidt's position regarding the constitutive role played by
conscious awareness, or noticing, of target language features during
L2 learning is supported by the introspective evidence of his own
experiences (Schmidt & Frota 1986) and by claims of cognitive
psychologists in the area of general learning theory (Brewer 1974;
Ericsson & Simon 1980) that learning without awareness is not
possible. Several other L2 researchers have argued that conscious
learning is an important aspect of L2 development, albeit to a
considerably lesser extent than Schmidt. In Bialystok's case (1990)
implicit, unanalyzed knowledge is eventually transformed into
analyzed knowledge via a process that takes place below the level of
awareness. This analyzed knowledge is subsequently available to be
drawn on in developing explicit conscious metalinguistic
12 Consciousness, implicit learning and SLA
representations. Such metalinguistic awareness is only useful,
however, for the kind of language tasks that are common in formal
education, like grammar tests and academic writing. Unlike
Bialystok, who does not concede the possibility that conscious
metalinguistic knowledge can influence the development of
analyzed representations, Ellis (1993a, 1994b, 1994c, 1995) adopts
a "weak interface position" in which explicit knowledge, at
developmentally sensitive points, is claimed to be capable of feeding
into the implicit system causing changes in its representation. The
information processing model of SLA proposed by McLaughlin,
Rossman and McLeod (1983), based on the model of automaticity
developed by Shiffrin and Schneider (1977), views language
development as a shift from controlled to automatic processing.
McLaughlin (1990a, 1990b) prefers to discuss this transition in
terms of the concepts of attention and control rather than in terms
of consciousness, since he sees many definitional problems with the
use of the term. As Hulstijn ( l 990b) points out, in McLaughlin et
al.'s information processing framework the relationship of explicit
to implicit knowledge during development is not viewed as
consecutive, in either direction, since both knowledge sources can be
drawn on in the development of automaticity, depending on "the
subject matter concerned, ...environmental conditions, and ...
individual capacities" (Hulstijn 1990b:34).
1.3 Renewed interest in consciousness in cognitive psychology
The claims of such SLA theorists need to be placed in the context
of the growing literature in mainstream cognitive psychology that
examines the role of consciousness in learning and memorial
processes (see Schmidt 1990, 1993a for a review of this literature
with special reference to SLA issues). Throughout the fifties and
early sixties a number of social scientists had argued for the
importance of 'tacit' knowledge (e.g., Polanyi 1958; von Hayek
1962), but the issue of the nature, and possible limits, of the
influence of conscious awareness on learning was a neglected topic
for psychological research and theoretical speculation. This was no
doubt a consequence of the influence of behaviorism on
psychological research which "removed consciousness (not t o
mention the unconscious) from the vocabulary of the science"
(Kihlstrom 1984:159). However, as a consequence of the demise of
behaviorism and the 'cognitive revolution' in psychology, which
was motivated in part by criticisms of Skinnerian learning theory by
Chomsky (1959), by the growth of information processing theory
Consciousness, implicit learning and SLA 13
(see e.g., Broadbent 1954; Shannon 1948), and by progress in the
field of neurocomputing (see e.g., Hebb 1949; Rosenblatt 1958), the
topic of consciousness and its relationship to models of information
processing became respectable again (see e.g., Baars 1988;
Bickerton 1991; Bowers & Meichenbaum 1984; Crick 1995;
Dennett 1991; Jackendoff 1987; Pekala 1991; Searle 1990, 1992;
Velmans 1991). In Kihlstrom's words, "consciousness, in the form
of attention, perception, memory, imagery and thought, is once
again at the center of things-with the difference that mental
contents and processes are approached with the same commitment
to publicly verifiable, quantifiable observation that characterized the
behaviorist paradigm of Watson and Skinner" (1984:150).
Along with the interest in specifying the role of consciousness
during information processing and learning there was, necessarily,
renewed interest in evidence regarding the extent to which
information could be processed and learned in its absence. In a
historical overview of the "rediscovery" of the unconscious in
psychology Reber observes that:
... during the 1970's it was becoming increasingly apparent that people d o
not always solve problems, make decisions, or reach conclusions using
the kinds of standard, conscious and rational processes that they were
more or less assumed to be using. Moreover, importantly, they often do not
seem to know what they knew nor what information it was that they had
based their problem-solving or decision-making on (Reber 1992:98).
Research by Kahneman and Tversky (see Kahneman, Slovic &
Tversky 1982), Nisbett and Wilson (1977) and Langer (1978),
amongst others, on the relationship between complex decision
making and awareness is representative of the renewed interest in
the influence of the unconscious. The strong claim of Reber, and
other psychologists who argue for the primacy of implicit,
nonconscious information processing in regulating our response t o
our physical environment, can be stated as follows: the acquisition
of knowledge of complex regularities, either of our first language
(Ll ), or of the world around us, proceeds by a process of induction
which is largely independent of conscious explicit efforts to learn.
The phenomenon of acquiring knowledge in the absence of
conscious explicit attempts to learn has been termed implicit
learning, and it has been studied in a variety of domains e.g.,
knowledge of facial characteristics and typical accompanying
personality traits (Lewicki 1986), knowledge of the working of
miniature economic systems (Broadbent, Fitzgerald & Broadbent
14 Consciousness, implicit learning and SLA
1986), and knowledge of the rules underlying the distribution of
sequences of lights at different points on a computer screen
(Howard, Mutter & Howard 1992). Most work has been done,
however, regarding the acquisition of knowledge of rules of artificial
grammars (e.g., Dulany, Carlson & Dewey 1984; McAndrews &
Moskovitch 1985; Reber 1967, 1969, 1976, 1989, 1993; Turner &
Fischler 1993; Vokey & Brooks 1992).
Reber (1989, 1992, 1993) has claimed that implicit and explicit
learning are completely dissociable. Both involve the allocation of
attentional resources to input and result in memorial
representations of the input. Critics of Reber (e.g., Dulany et al.
1984; Perruchet, Gallego & Savy 1990; Shanks & St. John 1994;
Vokey & Brooks 1992) argue over his interpretation of the
representation that results from learning under these two
conditions, particularly over his claim that the knowledge base
established by learning under implicit conditions is abstract, and
independent of the influence of memory for specific instances
encountered during learning. In Chapter 4 the representational
issues involved in this debate are considered in some detail. In
particular the relationship of attention to memory is examined in
the light of criticisms of Reber's ' abstractionist' claims about the
representation that results from implicit learning. The specific
issues involved in measuring and interpreting the measurement of
implicit learning, as well as the issues involved in measuring and
interpreting the measurement of memory are also examined.
1.4 The generalizability of research using artificial grammars
The tradition of research into implicit learning that has arisen out
of the claims and counterclaims referred to above has been
uninformed by information about the course or conditions of first
or second language acquisition. The selection of language, and an
artificial grammar at that, as the object to be learned is made simply
because it represents a stimulus domain of a sufficient level of
complexity to study the general phenomenon of implicit learning.
Consequently the generalizability of claims made about the effects
of learning artificial grammars under different conditions needs t o
be demonstrated by both first and second language researchers (Carr
& Curran 1994; Schmidt 1994b). This will involve using natural
language as the stimulus domain. In the sections below I review the
use of natural language in experimental studies of L1 and L2
acquisition.
Consciousness, implicit learning and SLA 15
1.5 Studies ofthe effects ofstructured stimuli on Ll learning
Work in a research tradition , related to that established by Reber
(1967) has also made use of artificial grammars. In contrast, this
work has been motivated by information about the course and
conditions of language acquisition. Most of these studies address
research questions arising from hypotheses about the nature of L 1
development, in particular the question of whether features
characteristic of motherese e.g., prosodic and morphological cues t o
phrasal and constituent structure, help to make salient features in
the input to learners and so aid induction of the underlying grammar
of the child's developing system (see Snow 1977, and Bohannon &
Stanowitz 1988 for arguments supporting the claim that motherese
is instrumental in facilitating L l development, and Gordon 1990 for
a critique of their position). However, these studies (e.g., Braine,
Brody, Brooks, Sudhalter, Ross, Catalano and Fisch 1990; Morgan,
Meier & Newport 1987, 1989; Morgan & Newport 1981), never
use children of the appropriate age as subjects. Braine et al. (1990)
is the single study using children, and they used children aged 7-10
years, which is older by at least one year than the age at which
maturational constraints have been claimed to affect the language
learning mechanisms (see Long 1990, for a review of the evidence
for maturational constraints which concludes that such constraints
become operative at around the age of 6 years).
Most of these studies, then, extrapolate from adult performance
to child performance, which is a dubious procedure since adults and
children differ cognitively in fundamental areas of likely influence
on the course of language acquisition (see e.g., Keil 1989, and
Wellman 1990 on conceptual development, and Long 1990;
Newport 1989; Patkowski 1980; Rosansky 1975 and Scovel 1988,
on critical period effects in language learning ability). These L 1
studies are also much less concerned with the effect of the
conditions of learning manipulated by the researchers on implicit
learning cited above.
1. 6 Experimental studies ofSLA under different conditions
A number of experimental studies have also been done in the area of
SLA. In contrast to the L l motivated studies, these have used the
population of interest, i.e., L2 learners, as subjects during
experimentation. These studies have used either artificial (DeGraaff
1995; DeKeyser 1994, 1995; McLaughlin 1980; Nation &
McLaughlin 1986), semi-artificial (Hulstij n 1989, 1990a) or natural
language as input to the learning activity (Doughty 1991; N. Ellis
16 Consciousness, implicit learning and SLA
1993; Hulstijn 1990a). Of interest to the present study are results
fr ?m studies reviewed below that used natural language as the
. and-form group performing most successfully. There �ere no
stimulus domam. However, as Schmidt (1994b) notes, these studies
have demonstrated few, if any, effects for learning under implicit
conditions.
1. 6.J Hulstijn (1990)
Hulstijn (1990a) studied the effects of two primary conditions,
focus on form versus focus on meaning, on measures of recall and
retention of a structure of Dutch. This structure was a sentence
containing a subordinate clause and a passive modal auxiliary. This
structure was chosen because "intermediate learners of Dutch were
unlikely to be familiar with all its grammatical properties"
(1990a:55). The research questions were whether attention t o
meaning alone would incidentally facilitate the acquisition of formal
knowledge, and whether the attention to meaning group would
outperform the attention to form group on measures of such formal
knowledge as well as measures of knowledge of the content of target
sentences. Two sentence-copying pretests of knowledge of the
target structur.e were given during which subjects were briefly shown
sentences which they had subsequently to copy. Since exposure t o
the sentences was brief, accurate copying was taken t o indicate the
use of prior knowledge of the structure. The treatment for the form
gro�p was a series of sentence fragment ordering tasks, in which
sub3ects had to match eight sentence fragments with the order
illustrated by a sentence appearing on a computer screen. In
contrast, the meaning-oriented group were asked to read the
sente�ces appearing on the computer screen and answer open-ended
q�esttons �b?ut the . content. These questions required subjects t o
�Ive an opm10n which they then had to write down. The posttest
mvolved a cued recall task in which subjects were required t o
remember all the target sentences presented during training. This
was scored for accuracy of form and content. A second posttest
measure followed this. This was a retention test, identical in form to
the pretest sentence copying test. Due to unequal periods of
exposure required during the form and the meaning treatments (the
anagram task required 30 seconds of exposure to the target while
the opinion qu�stion needed only 10 seconds of exposure), both a
form-plus-meamng and a meaning-only group with much shorter
exposure time were added to the two primary conditions as controls.
Subjects were adult learners of Dutch from a variety of non-Indo­
Germanic native language backgrounds. Results showed that recall of
Consciousness, implicit learning and SLA 17
meaning was better for meaning-oriented groups, an� recall of f? rm
was better for form-oriented groups, with the attent10n-to-meanmg­
differences between groups on the second measure of learnmg, that
of retention.
There are a number of design problems with this study. Firstly
there is the issue of the target structure selected. Hulstijn reports no
evidence to support the claim that it was unlikely to be known by
intermediate learners other than the assertion that it is "generally
found difficult" (1990a:55). Despite the length of the pretest it is
unclear whether this was an adequate measure of prior knowledge
since it involved a confound between knowledge of the target
structure's grammaticality and the subjects' skill at writing it down
quickly. Secondly, eleven out of the total of twenty sentences used
in the pretest sentence-copying task contained the target structure,
and one of the pretests actually used two stimulus sentences
presented during the treatment. This is likely to have alerted
subjects in the incidental, meaning-oriented condition to the formal
characteristics of such sentences. Consequently there 1s . a strong
possibility that these subjects were aware of, and consciously . trying
to understand, the formal characteristics of the sentences while also
answering the meaning questions, thereby invalidating the
distinction between treatments. No debriefing measures were taken
to identify what subjects were thinking about during the treatments,
which could have been used to help confirm that separate
treatments were actually delivered. Thirdly, there is the problem of
unequal length of necessary exposure which Hulstijn attempted to
correct for by having control groups. Rather than balance the
results of the thirty-second incidental exposure group with a ten­
second incidental group it would have been better to devise a form­
oriented task that required the same amount of exposure time as the
original incidental exposure group. Finally �here i� the � robl�m of
possible learning effects in the posttests. It 1s poss1bl� , given (1)_ � he
length and nature of the cued recall task (remembermg and wntmg
down the nine target stimuli), and (ii) the fact that the format of
the sentence-copying posttest was repeated from the pretes� phase,
that learning could have occurred here and not dunng the
treatments.
1. 6.2 Doughty (1988)
s at
Doughty (1988, 1991) studied the acquisition of relative clause
Accessibility Hierar chy
different points on the Noun Phrase
18 Consciousness, implicit learning and SLA Consciousness, implicit learning and SLA 19

(Keenan & Comrie 1977) by adult learners of English as a second 1. 6.3 Ellis (1993) . .
co�dit10ns on th e
language (ESL) as a function of learning under three conditions. In N. Ellis (1993) studied the effects of three
native speakers of
the first condition, a rule-oriented group were given rules for learning of rules of Welsh morphology by
(i) a r a dom gr��p, who �ere
relativization at different points along the hierarchy and examples. English . The three groups were: . . �mg-Elhs s oper a�� on­
In the second condition a meaning-oriented group were required t o exposed to instances alone durm g tra ��
answer comprehension questions on stimuli containing examples of alizat ion of an 'impl icit' learn ing cond ition, (1993:302); (11) a
relativization at different points on the hierarchy. These stimuli grammar group, who were taught the rules to a criterion of success
had been highlighted at structurally salient points. In the third before presentation of those instances seen by the random group;
condition subjects were simply exposed to examples of structures and (iii) a structured group, who were taugh t by a blend of rules and
.
mstan ces as the random
containing relative clauses without instructions to manipulate them examples before being exposed to the same
te ext nsi"'.e amou ts of
structurally or to attend to meaning. and grammar groups. However, despi � . lea �mg � y
nstra te impli cit
The meaning-oriented group outperformed both the others in training, Ellis' research failed to demo �
transfer comprehension tests. However, it is impossible to say what rando m group of the rule for soft muta tion in Welsh. Subjects m
the
.
caused this since in this condition there was a confound between two the structured group performed best in the test phase
(1993 ) study are the fact
possible causal variables: the presence of both attention-drawing Two things to be noted about Ellis's
erien ced langu age leam�rs,
devices and meaning-oriented instructions to answer comprehension that: (i) the subjects were relatively inexp
tiz ment s place d on notic e
questions on the stimuli. Performance resulting from this treatment, having simply been recruited via adver �
tion to be l�arn ed was
that is, may either have been the result of simply attending to the boards; and (ii) the rule of soft muta . iar to the
to be famil
meaning of the presented sentences containing relative clauses, or linguistically sophisticated and likely � .
the result of a conscious search for rules prompted by the saliency cts, both conc eptua lly and in term � of their expene�ce of
subje
age learmng, of
inducing highlighting of structural points in the examples. This solving problems in the context of foreign lan� .
confound is clear from the instructions to the meaning-oriented which they appear to have had very little . Studi es hke _those .of
group which require them to read sentences in order to understand Hulstijn (1990a) and Doughty (1991) using subje ct I? opula twns w:ith
t L2 and usmg rn · es relat mg
the meaning and the grammar (1988:85). greater experience in learning the targe synta x,
e.g., word order and
It is also likely that there was some contamination of treatment to a more familiar area of L2 structure,
groups resulting from informal contact between subjects over the seem more likely to produce clear er evide nce of an e � fect for
v�h. �ity, smce
. .
1cal
ten-day period of the treatment, when subjects in meaning-oriented implicit learning and also to have greater e_colog
maJo nty of L2
and rule-oriented treatments could easily have conferred with each this situation is more typical of those facmg the
other about the differences in the treatments. No debriefing sessions learners.
were used to identify the differential extent to which subjects were
searching for or aware of rules during the meaning-oriented 1. 7 The needfor further studies in a second language context
treatment. While the three studies reviewed above all demonst:ate the
The third condition, which consisted simply of exposure t o superiority of conditions that require attention to meamng over
examples, appears closest to the implicit condition manipulated by conditions that require attention to form alone, none � f t�e results
Reber, but this condition was the one that demonstrated least for conditions approximating Reber's implicit memonzat10n tas_ks
success on the transfer tasks. Consequently, as Schmidt (1994b:32) show significant evidence of learning. Further, � ach of these studies
comments, it is unclear whether any implicit learning occurred in contained design flaws that obscu: e th� potentl�l relevanc� of the
.
this experiment, since in both the unsuccessful exposure-only results. It is simply not clear that 1mphc1t learnmg, as described and
condition, and the successful meaning-oriented condition learners facilitated by Reber's experiments, actually occu: red.
could have actively been pursuing hypotheses about the formal Yet, despite this lack of research evidence, some L2
characteristics of stimuli sentences. methodologists, for a number of years now (Asher 1979; Krashe!l
198l a, 1985, 1993; Prabhu 1987; Terrell 1982), have based their
20 Consciousness, implicit learning and SLA
prescriptions on assumptions that implicit learning is not only a
factor, but the most significant factor in L2 development.
Consequently, the Natural Approach (Krashen & Terrell 1983)
Total Physical Response (Asher 1979) and the Bangalore
Procedural Syllabus (Prabhu 1987; cf. Beretta 1989) advocate a set
of methodological procedures that avoid an explicit focus on formal
rules of grammar during instruction. Such prescriptions are worth
little without research demonstrating in detail the extent of the
phenomenon in the domain of L2 learning and the extent to which
attention to formal features of the input to learners can be
manipulated so as to aid induction, independently of accompanying
grammar explanations (Long 1991; Schmidt 1990, 1993a).
It is also premature to draw firm conclusions for syllabus design
from the few studies of implicit L2 learning that have so far been
done (Ellis 1993a; Robinson 1993a, 1994a). Motivating decisions
about the units and sequence of language instruction is necessarily
speculative, and such work must integrate established research
findings from a variety of relevant areas (see e.g., the approach
taken by Long and Crookes 1992 to motivating a task-based
approach to syllabus design based on the evidence of SLA research).
If research in the area of implicit learning is to have any
consequence for the design of structural syllabuses, as Ellis (l 993a)
claims, the important question of exactly what areas of syntax are
learnable implicitly and explicitly needs to be further addressed.
Both Reber (1989, 1993) and Krashen (198l a, 1982, 1985, 1994)
make similar claims about the differential effectiveness of
nonconscious, implicit conditions and conscious, explicit conditions
in learning rules of different orders of complexity. Both claim that
complex rules are best left to implicit processes, whereas simple
rules can be learned explicitly. If this could be shown to be so for
SLA, then it would have important implications for L2 syllabus
design and methodology. However, it is not clear whether the results
from Reber's experimental studies of artificial grammar learning,
which have been used to support this conclusion, can be used t o
support similar arguments made by Krashen and others who claim
that acquisition of the formal characteristics of a second language is
necessarily a largely unconscious process.
1.8 The aims of the present study
There is thus a need for further studies that address the effects of
different conditions of learning on the acquisition of a natural
language by a representative sample of its potential acquirers, adult
Consciousness, implicit learning and SLA 21
speakers of another language. In particular, the present study
.
examines the issue of the relationship between the complexity of
the L2 rule to be learned, and the conditions under which learning
takes place. The generalizability of claims made by �eb�r (19 � 9,
1993) concerning the acquisition of the rules of an artificial fimte­
state grammar to SLA is examined by contrasting performance by
non-native speakers of English on learning simple and complex
rules of English under four different conditions of exposure. T�� of
these conditions, an implicit and a rule-search � onditl? n,
.
operationalize the distinction between implicit and explicit learnmg
using experimental tasks familiar from the work of Reber (1969,
1976, 1989, 1993). Two other conditions, an inc� dental and an
. .
instructed condition, operationalize the distmct10n usmg
experimental tasks familiar from the discussions of the difference
between acquisition and learning in SLA research and pedagogy (e.g.,
Elley 1991; Ellis 1992, 1994b; Gregg 1984; Krashen 198l a, 1982,
1985 1994· Krashen & Terrell 1983; McLaughlin 1987; Terrell
; '
1982 Zobl 1992, 1995). The operationalization of the distinction
between a simple and complex rule is described in Chapter 2. The
four conditions are motivated by the discussion in Chapters 3 and 4
and are described in Chapter 5. The aim of the research is t o
establish the degree of generalizability from the paradigm for
examining the extent of implicit learning developed by Reber
(1969, 1976, 1989, 1993) to SLA research, and to address the issue
of the extent to which structural knowledge of a second language
can be acquired without consciousness (Schmidt 1990) in the sense
of an awareness of rules, or a deliberate intention to learn the
structure of the language.
 Chapter 2

Rules, Complexity An d Second

Language Kn owledge

2. 0 Introduction
This chapter is concerned with motivating a distinction between the
simple and complex rules of English which serve as the dependent
variables for the present study. After a discussion of the distinction
between pedagogic rules and linguistic rules, a description is made of
claims arising from experimental studies of implicit learning (Reber
1 967, 1 969, 1 976, 1 989, 1 993), and within the field of SLA theory
(Krashen 1 982, 1 985, 1 994), that implicit and explicit learning
conditions are differentially sensitive to the complexity of the
domain to be learned. Next, a comparison is made of proposals by a
number of SLA researchers for distinguishing between the order of
difficulty of L2 rules. Finally the procedure adopted in the present
study for identifying simple and complex L2 rules is explained, and
the rules identified are related to the preceding discussion.

2.1 Pedagogic rules and linguistic rules

Second language researchers frequently invoke the notion of rule t o
describe what is learned during the process of instructed ci r
naturalistic SLA . A s in L1 acquisition, L 2 learners engage in
processes of regularization and generalization (see Fodor & Crain
1 987; Larsen-Freeman & Long 1 99 1 ; Marcus, Pinker, Ullman,
Hollander, Rosen & Xu 1 992), and such observable behaviour is
often the basis on which cognitive correlates of rules are attributed
to learners (see Reese 1 989; Slobin 1 979) . However, there are two
senses of the term rule which need to be distinguished. In the first
sense, the use of the term rule implies a claim about the form in
which knowledge of language is represented in the mind of the
learner. However, linguistic theories differ with respect not only t o
the formalisms, but also to the nature o f the rules and cognitive
mechanisms they propose. In fact, the need to posit rules of the
symbol-processing type traditionally invoked by linguists as part o f
 24 Second language rule complexity
language representation is questioned by physicalist philosophers o f
mind (e.g., Crick 1 995; Searle 1992) and by production systems and
connectionist models of cognition, see Holland, Holyoak, Nisbett &
Thagard 1 986; McClelland & Elman 1 986). For example the debate
between White (1 992) and Schwartz and Gubala-Ryzak ( 1 992)
regardin g the interpretation of data supposedly demonstrating the
acquisition, or resetting, of the verb-movement parameter for
English by French native-speaker learners invokes this linguistic
sense of rule. White argues that the data demonstrate that learners
have reset the parameter, or learned the rule. Schwartz and Gubala­
Ryzak argue against this interpretation, since the theory o f
Universal Grammar predicts that resettin g the parameter entails the
acquisition of a cluster of properties which the learners show n o
evidence o f having acquired. Pedagogic rules, in contrast, are
traditionally presented as simplified versions of such linguistic rules
which necessarily fall short of exhaustive treatment and avoid the
abstract theoretical characterization typical of linguistic rules
(Corder 1988; Faerch 1 986; Sharwood Smith 1 994; Westney 1 994).
Given their limited scope and level of detail how is it possible fo r
pedagogic rules to be used to develop L 2 competence?
2.1.J The effectiveness ofpedagogic rules
There has, to date, been little empirical investigation of the
effectiveness of different pedagogic rule formats on the
development of L2 knowledge. In contrast, a large empirically based
literature exists on the effectiveness of pedagogic rule formats, rule
explanations and their use during problem-solving in other areas o f
the curriculum. This includes studies o f the effectiveness o f physics
and mathematics explanations imposing differential 'cognitive load'
on the learner (Swelle r 1 988, 1 990); studies of the effectiveness o f
reminding learners o f earlier presented, analogous examples during
general problem-solving (Ross 1989; Ross & Kennedy 1 990) and
studies of problem classification systems used by expert and novice
learners (Chi, Glaser & Rees 1 982; Voss, Perkins & Segal 1 99 1 ) .
Discussions of L2 pedagogic grammar have failed to draw on this
literature, and have revolved instead around the general question o f
whether rules (of any description) should be presented to learners.
The theoretical claims regarding the effectiveness of pedagogic
rules can be grouped as follows: (i) the non-isomorphy, non­
interface position (implicit knowledge, or competence, and explicit
knowledge of rules are different in kind and non-interfaced:
Bialystok 1 978, 1 988; Krashen 1 979, 1 9 8 1 a, 1 982, 1 985, 1 994;
Second language rule complexity 25
Prabhu 1 987); (ii) the attention-focusing position (rules are useful
as devices for focusing attention on selected aspects of the
structures to be learned, which are then . learned as a consequenc� o f
.
inductive processes that are not accessible to co�� c10usness: Se iger �
1 979; Sharwood Smith 1 99 1 , 1 993); a�d (111) the consc10us
.
understanding position (learning and applymg pedagogi� . rules can
lead to a conscious understanding of the structural regulanties of the
.
domain the rule applies to: Schmidt 1 990; S�hmidt & Frota 1 986).
Examples of these theoretical positions are given below.
(i) The non-isomorphy, non-interface position. Bialystok ( 1 98 8 )
claims that the form o f pedagogic rules and the form of our
linguistic competence, or implicit knowl� dge of language, are
�
different in kind, and consequently are not mterfac d, or mutually
accessible. Bialystok distinguishes between 'analyzed knowle ge and �
'articulated' knowledge. Language development is charactenzed y �
her as the accumulation of unanalyzed knowledge of language. This
accumulation takes place independently of awareness of the
structure of the knowledge, though increases in the amount o f
knowledge are accompanied by an increase in the exten . o f i s � �
analysis. Analyzed knowledge "is represented as a proposi ion m !
which the formal structure and the relationship to meanmg are
apparent. Non-analyzed knowledge is assigned a mental
.
representation in which the underlying formal constituents are n o t
necessarily identifiable... I n these terms, analyzed knowledge can be
described as access to the propositional structure of � on-analyzed
�
knowledge" ( 1 988:33). However, the process of analys! s ta es place
below the level of awareness and is thus different m kmd from
'verbal ' articulated knowledge: "Thus it is erroneous to equate
�
analyz d knowledge with articulated knowledge, or knowl�dge o f
rules" ( 1 988:40). Consequently for Bialystok it is not possible for
pedagogic rule-focused activity to contribute to the development o f
the implicit system.
�
Prabhu ( 1 987) also argues that pedagogic rules cann_o be used t o
initiate language development. He claims that explicitly learn.ed
grammar rules only " . . . 'make sense' when. they acco d with \
language samples arising from or conformmg . to �ne s own
!
competence, and there is often a sense of satisfac 10n or . o f
discovery when that happens: what one has 'known' with�mt b�mg
�
aware of it is now confirmed as being right (hence the sa isfact10n ?,
and is also seen to be rule-governed (hence the sense of dis� overy)
( 1 987:77). For Prabhu the sense of satisfaction that can anse from
26 Second language rule complexity
rule-focused a�tivity is only possible when "one has already
developed an mternal system capable of yielding samples which
conform to the .rule. When that is not the case, rules are just so
much complex mform�tion" ( 1 987:78). In other words, explicit
rules can sei:e a� ?evices for developing awareness of what has
already been implicitly learned. Contrary to Schmidt ( 1 990), Prabhu
asserts that development of the representational system proceeds
. was sometimes accompanied by recollection of an explicitly taught
without awareness. Like Bialystok ( 1 9 8 1 ), he claims explicit rules
make sense �nly after the implicitly learned system has reached a
. development that understanding a pedagogic rule can make, does
particular pomt of development.
(ii) The attention-focusing position. Seliger ( 1 979) observes that
teachers �resenting pedagogic rules to learners often "have as their
goal causmg someone to P.roduce a language form, that is, getting a
learner to perform consistently with regard to some aspect o f
language behavior" (Seliger 1 979:360). However, as both Green and
Hecht ( 1 992) an � � �liger ( 1 ?'. 9) have shown, learners can perform
form-focuse � activities requmng article production (Seliger 1 979) or
error correction (Green & Hecht 1 992) that suggests that conscious
knowledge of such rules, as indicated by the extent to which a
learner is able to ver�alize them. after the task, is disjoint with the
knowledge .drawn on m pe�formmg the tasks. Seliger's conclusion,
therefore, �s that pedagogic rules are neither isomorphic with the
represe �tat10n of languag� knowledge in the learner' s internally
developm� system, no: devices drawn on in monitoring production,
but r�ther ,�re useful m language teaching in order to get learners t 0
do thmgs with language in an efficient manner, to focus on those
asp:cts . of th� language phenomenon that must be acquired, and t o
avoid mefficient testing of hypotheses" ( 1 979:3 60). Pedagogic
rules, h � concludes, are not used deductively to ' control output, '
ex�ept m an e.xtremely small number of cases (Seliger 1 979:3 67).
Seh�er !hus differ� . from Krashen ( 1 982: 1 2 1 ), who attributes a
mom!ormg and edi�mg function to pedagogic rules. For Seliger the
function of pedagogic rules is to aid induction:
!
. . .the� m ?st like y serve as a mechanism to facilitate the learner's focusing
�
on cr�tenal at nbutes of the real language concept that must be induced.
That is, consc10us or pedagogical rules make the inductive hypoth e s i s
.
testmg process more efficient (Seliger 1 979:368).
(iii) The conscious understanding position. The positions
regarding the use of pedagogic rules reviewed above conflict with
Second language rule complexity 27
Schmidt's account of his own learning of Portuguese (Schmidt &
Frota 1 986). Seliger ( 1 979), Bialystok ( 1 988) and Prabhu ( 1 9 8 7 )
consider awareness at the level of understanding, which may be
made possible by exposure to a pedagogic rule, to be irrelevant t o
L2 development. Schmidt, in contrast, reports his own experience
that "improvement in second language performance follow(ed) o n
the heels o f understanding" ( 1 990: 1 47 ) and that this understanding
rule. Seliger's position, while denying the potential contribution t o
consider such rules facilitative to the extent that they focus
attention on aspects of the language to be learned, leading t o
'noticing' in Schmidt's terms ( 1 990). Schmidt notes that what
studies there are of the effectiveness of providing explicit rules t o
learners are inconclusive about whether "such effects are due t o
increased learner understanding as a result o f instruction, or
increased salience of forms leading to awareness only at the level o f
noticing" ( 1 990: 1 46).
2.1.2 Noticing examples and understanding pedagogic rules
There are thus two ways in which explicitly presented pedagogic
rules have been claimed to lead to language learning, and these are
associated with developing awareness of the language system at two
of the different levels identified by Schmidt ( 1 990), noticing and
understanding. Attending to the rule may: (i) simply cause learners
to notice the structures which it is explaining, or (ii) attending t o
the rule may also lead to an understanding of the structural
regularities upon which it is based.
Complexity is likely to affect structural noticing, and rule
understanding in the same way. Simply noticing the structures
presented as examples is unlikely to be facilitative of learning if the
structures themselves are too complex and the salient features of
the structures that the rule regulates are consequently not obvious.
Similarly, the more complex the explanation of a rule is, the less
likely it is to lead to understanding, and the more likely it is to be
ignored in favor of simpler rules of thumb, as in the case o f
Schmidt's choice o f rules for learning the aspectual requirements of
past tense Portuguese verbs: "I believed that I used two rules of
thumb for aspectual choice-use imperfect for used to and for
making excuses-I was told some more complex rules for aspectual
choice in class, but I ignored them" ( 1 990: 1 4 7).
 28 Second language rule complexity
2.1.3 Two dimensions ofpedagogic rule complexity
These two components of pedagogic rules must be well matched if
the possibilities of noticing and understanding are to be maximized
for the learner. The complexity of the structural regularities
captured by exemp lars must match the complexity of the
accorr,ipanying explanation of them. Simple patterns of structural
_ examines claims that have been made within the field of cognitive
covar�ance �h1ch are present in examples may, in one scenario, be
explame� m an overly complicated fashion. Conversely, the
�omplex1ty of the . structural covariance in the exemplars may be
madequately descnbed, because oversimplified. Estimates of the
relative complexity of pedagogic rules have, therefore, to take both
of these factors into consideration. The relationship between these
two dimensions of pedagogic rules is illustrated in Figure 1 .
Structure
+simple
+.
1 2
Explanation + brief + detailed
3 4
+ complex
Figure 1. Elements Of Pedagogic Rules
Effective pedagogic rules, in which the level of detail of the
explanation matches the extent of the covariance of the structures
regulated by the rule, would occupy quadrants 1 and 4. These
correspond to effective simple and complex pedagogic rules
respectively. Less effective rules would occupy quadrants 2 and 3 .
These correspond to rules which explain a relatively simple pattern
of structural covariance in a complicated, overly detailed way
(quadrant 2), �nd rules which describe a complex pattern o f
structural covariance in a brief oversimplified form (quadrant 3).
Second language rule complexity 29
The procedure for identifying the simple and complex structures,
and the accompanying brief and detailed explanations of them,
which together make up the simple and complex rules presented t o
learners in the instructed condition o f the present study are
explained in section 2.4 of this chapter. The following section
psychology and L2 pedagogy regarding the relation between
consciousness, learning condition, and the complexity of the
stimulus domain to be learned.
2.2 Relating rule complexity, consciousness and learning
Reber ( 1 989), approaches the issue of the relation between rule
complexity, consciousness and learning from a tradition of using
artificial grammars to study the effects of different degrees of rule
awareness on learning. Reber claims that where the stimulus domain
is non-complex, and where the rules underlying it are simple, then
implicit processes will not be displayed. Implicit processes will,
however, be displayed where the stimulus domain is complex, and
will result in the unconscious abstraction of the corresponding
complex rules. In contrast, explicit efforts to learn complex rules
are much less likely to be successful. In the field of SLA theory
Krashen ( 1 98 1 a, 1 982, 1 985) distinguishes between 'easy' and
'hard' rules of English for L2 learners, often using the terms
'simple' and 'easy' interchangeably to refer to the same rule (see
section 2.2.3). Like Reber, Krashen asserts that only easy rules are
consciously leamable; hard rules must be acquired via the
unconscious processes that develop the acquired system.
2.2.1 Claims based on experimental evidence of implicit learning
studies
Based on the research using artificial grammars referred to above,
Reber's claim is that the knowledge base that results from learning
under implicit conditions is abstract. Reber ( 1 976, 1 989, 1 992,
1 993) consistently refers to the notion of 'rules' to describe what is
learned during exposure under implicit conditions: " . . . the basic
procedure is to have an acquisition phase, during which subjects
acquire knowledge of the rules of the grammar" ( 1 989:220).
Further, Reber claims that complex rules, corresponding to complex
non-salient covariations in the stimulus domain, are unlikely to be
learned, following a process of induction, under the condition t o
explicitly search fo r rules. This i s not the case with learning such
rules under implicit conditions: " . . . explicit processing of complex
30 Second language rule complexity
materials has a decided disadvantage in relation to implicit
processing" ( 1 989:223). However, simple rules, corresponding t o
non-complex covariations, can be induced and so learned under the
explicit rule-search condition. This is of great potential interest to
L2 research, since it makes a prediction, which it should be possible
to examine empirically, about the question of what is learnable
under different conditions, and therefore of the effects of
manipulating the conditions under which aspects of linguistic
structure are noticed.
It is clear from Reber' s discussion of these issues that he believes
that the complexity of the covariance in the stimulus domain
affects saliency. Even if the learning condition is an implicit one,
following instructions simply to memorize presented strings of the
artificial grammar, Reber claims that the presence of simple
patterns of covariance in the stimulus domain will become salient
and trigger explicit search for rules, which are likely to be
subsequently discovered. Hence his statement that the rules to be
learned, and the stimulus domain which they regulate, must be
complex if implicit learning is to be displayed: "If the system in use
is too simple, or if the code can be broken by conscious effort, then
one will not see implicit processes" ( 1989:220). Consequently Reber
uses artificial finite-state grammars to generate the strings of letters
used as stimuli in his studies (see section 4. 1 ) because they are
"sufficiently complex . . .that the underlying structure is not within
the conscious memorial domain of the typical subject
upon . . .entering the laboratory" ( 1 989:220).
The grammars of natural languages, of course, are considerably
more complex than the artificial grammars used by Reber. However,
rules regulating subsystems of the total grammar may differ
sufficiently in complexity to allow the claims made by Reber to be
examined using L2 learners as subjects.
2.2.2 Pedagogic implication of the claims
The issue of what is learnable under implicit versus explicit
conditions is of major interest to L2 research. If the effects o f
consciousness-ra1smg, or noticing features o f the L2, are
differentially sensitive to features of the linguistic system to be
learned, then it follows that the structural criteria for grading and
sequencing the units of the language presented to learners adopted
by many of the synthetic, Type A syllabuses described by White
(1 988), Nunan ( 1 9 88), and Long and Crookes ( 1 992), must be
reevaluated in the light of evidence of what is learnable under
Second language rule complexity 3 1
implicit and explicit conditions. Either such syllabuses would ne�d to
restrict the inventory of forms to be learned to those that are hkely
to be learnable under explicit conditions (perhaps adopting criteria
for sequencing such forms from developmental evidence, where
available), or the syllabus designer could incorporate infor� ation
about the appropriate learning condition into the specific�tton . o_f
items, leaving the teacher to methodologically :veave the 1mphc1t
and explicit strands of the syllabus together durmg the process o f
instruction.
There are also likely to be consequences for the methodological
implementation of some an�lytic, Ty�e B, s)'.ll�buses that employ
task as the unit of analysis, regardmg dec1s10ns about how t o
facilitate a ' focus on form' (Long 1 985, 1 99 1 ; Long & Crookes
1 992), since forms that vary in complexity may be differei;itially
suited to options for focusing on form that range along a contmu�m
_
of explicitness e.g., from pedagogic rule yre� entat1on, to mak mg
_
input to a task salient through highhghtmg o� phonological
emphasis, to simple repetition of examples of mput structure
experienced on task (Jourdenais et al. 1 995 ; Leeman et al. 1 99 5 ;
Sharwood Smith 1 993).
2.2.3 Speculation about rule complexity in L2 pedagogy
.
The relevance of the distinction between levels of complexity of
the linguistic stimuli to be learned under explicit and implicit
conditions has been acknowledged by L2 researchers but has
resulted, to date, in either post-hoe reasoning about results o� studies
intended to address other research questions (see e.g., B ialystok
1 979; Green & Hecht 1 992; Lyster 1 994; Pica 1 983 ; Ramat 1 9 ? 2),
or unmotivated assertions that stimuli used in experiments differ
along an unspecified dimension of complexity. Hulstijn ( 1 990 a), ip
particular, has claimed that the target structure presente� . m _ his
study of implicit learning was complex, but no . empmcal or
theoretical motivation for this claim was offered by him (see 1 .6 . 1 ) .
Without such empirical o r theoretical support claims for
complexity differences between rules often end up as simple matters
of dispute between individual researchers. Tarone ( 1 985), for
example, in a study of the differential maste� of rules under
different task conditions (a narrative, a conversat10n and a formal
grammar test) follows Krashen ( 1 982) in distinguishing th� simple
rule for third person ' s' from the complex rule for article use.
Tarone also includes the rule for direct object pronoun occurrence
along with third person 's' as an example of a simple rule. However,
32 Second language rule complexity
Preston ( 1 989) disagrees: " . . . object pronoun occurrence and article
use are more subtle morpho-syntactic and semantic processes.
Neither can have its rule for use stated easily... " ( 1 989:259).
Krashen ( 1 982:97-98) distinguishes between formal and
functional dimensions of complexity in distinguishing 'easy' from
'hard' rules: "The rules that we can learn and carry around in our
heads for use as a Monitor are not those that are the earliest
acquired, nor are they those that are important for communication.
Rather, they are the simple rules, rules that are easiest to describe
and remember" ( 1 982:97). Wh-question formation, requiring
extensive permutations of word order, is claimed to be a formally
complex rule, in contrast to the suppliance of the morpheme for
third person agreement, which is formally simple. The use of plural
forms is functionally simple, whereas choice between definite and
indefinite articles is functionally complex.
Ellis ( 1 990: 1 67) makes essentially the same distinction between
structures that are amenable to explicit instruction and those that
are not; structures that can be explicitly taught are those that are
formally simple and are transparent with regard to the form­
function relationship the grammatical rule regulates. Examples of
formally simple and functionally transparent structures, Ellis
suggests, are plural 's' and the copula 'be' ( 1 990: 1 67). Ellis notes,
however, that formally complex structures can be explicitly taught
in some cases if the learner is developmentally ready. This, in
contrast to Krashen, assumes an interface position between implicit
and explicit knowledge in which explicit knowledge can feed into
the implicit system at developmentally appropriate points and
cause restructuring (see Ellis 1 993a and Robinson 1 993a, 1 994a).
The distinction between formal simplicity and semantic
transparency, which both Krashen and Ellis use to distinguish simple
from complex rules, relates to a similar distinction made by Slobin,
who distinguishes operating principles for the acquisition o f
language motivated by perceptual saliency from those motivated by
degree of semantic transparency ( 1 979, 1 985). However, Krashen
nowhere motivates his distinction by specifying the extent to which
the simple and complex structures they relate to differentially
affect processing mechanisms or make differential demands o n
processing resources. In fact, as Gregg ( 1 98 8 : 73 ) points out,
Krashen's distinction between easy and hard rules, and his notion o f
rules i n general, is one based on pedagogic formulation o f the rules,
and is not licensed either by a full theory of the mental
representation of grammar or by operating principles of the type
Second language rule complexity 33
proposed by Slobin. In other words, Krashen's candidates for easy
versus hard rules, and Ellis' s too as described above, are descriptive
rules, and as such are examples of what Chomsky ( 1 9 8 6 : 1 5-24)
terms grammars of E-language, i.e., theories of the characterization
of language independently of any concern for how it is
psychologically and biologically represented in the mind (cf.
Chomsky 1 990; Searle 1 990, 1 992).
Schwartz ( 1 993) and Zobl ( 1 992, 1 995) both attempt to
reconcile Krashen's distinction between acquisition and learning
with recent work in Universal Grammar (Chomsky 1 986), and in
particular relate the notion of rule of language used by Krashen to a
more appropriate grammar of I-language, a theory that purports t o
depict what one knows when one possesses language in the
Chomskyan sense of competence. To this extent they invoke the
interaction between rule complexity (as defined by the descriptions
of language in Universal Grammar), and implicit knowledge (in the
innatist sense described in section 1 .2.2 above), to explain the
course of L2 development.
2.3 Second language rule complexity
In addition to the formal and functional complexity distinction
cited by Krashen ( 1 982) and Ellis ( 1 990), L2 researchers have
appealed to a variety of criteria for distinguishing complex from
less complex or simple rules. Some examples of these are given
below.
2.3.1 Perceptual salience
In a study of the acquisition of wh-question formation rules by
college level students of ESL, Bardovi-Harlig ( 1 987) cites salience
as the criterion distinguishing preposition pied-piping in wh­
questions of the form 'To whom did John give the book?' from wh­
questions involving preposition stranding of the form 'Who did
John give the book to? ' . It is this criterion, she claims, which
accounts for the fact that preposition stranding was acquired before
pied-piping. Her results run counter to the predictions of
markedness theory which claims that the unmarked form,
preposition pied-piping, will be learned before the marked form,
preposition stranding.
By salience Bardovi-Harlig appears to mean frequency in the
input, since preposition stranding, though marked, is much more
frequent as a wh-question form than pied-piping. However, it is
clear that salience could also be operationalized in perceptual terms
34 Second language rule complexity
to explain her results, since the relationship between the wh-word
and the preposition which is crucial to understanding this form of
wh-question formation is visually and acoustically more salient in
preposition stranding than in pied-piping by virtue of the fact that
they occupy positions at the beginning and at the end of the
structures in question. In pied-piping structures the wh-word is less
salient by virtue of being sentence-internal, and is therefore less
easily 'noticed' . Learners, that is, may simply have been using
strategies for allocating attention by adopting what Slobin ( 1 9 8 5 )
calls the operating principle o f paying attention to the end o f words
or strings. This, rather than linguistically defined markedness, m ay
have led them to be more aware of the critical aspect of structure
regulating preposition stranding than pied-piping. Salience in this
perceptual sense may therefore relate to the ease with which
structural correlates of rules can be noticed and so learned. Similar
considerations underlie Meisel, Clahsen and Pienemann' s proposal
( 19 8 1 ) that the acquisition of word order rules in German may be
regulated by the use of processing strategies which lead learners first
to attend to word order permutations at the beginning and ends o f
strings before attending to the details o f sentence-internal
permutations. However, while one of Slobin's operating principles
can be invoked to support the claim that preposition stranding is
more salient than pied-piping, and therefore more easily noticed
and so learned, it is also true that another of Slobin's principles
'avoid discontinuities' makes the reverse prediction about
learnability. Preposition stranding, 'who-to' is a discontinuous
structure, in contrast to the continuous 'to whom' .
2.3.2 Lexical rules versus abstract rules
Bialystok ( 1 979) attempted to identify the extent to which
manipulating time allowed and level of learner in an L2
grammaticality judgment task elicited implicit versus explicit
knowledge of the rules regulating the grammaticality of structures
presented for judgment. The subjects, English speaking high school
learners of French, were required to identify errors and articulate
rules for stimulus sentences. She found that adjective errors were
easier to detect than pronoun errors, which in turn were easier than
verb errors. This relates to the difficulty of the rules with which the
errors were associated, "rules pertaining to single lexical items were
easiest and those to general structures were most difficult"
( 1 979 : 1 00). An example of the former is "Color adjectives always
come before the noun." An example of the latter is "To form the
Second language rule complexity 3 5
passe compose, use the correct form of avoir or etre plus the past
participle of the verb" ( 1 979: 1 0 1 ).
2.3.3 Size of context and degree of semantic opacity
Green and Hecht ( 1 992), in a similar study to that of Bialystok
( 1 979) and Seliger ( 1 979), required school age German speaking
learners of English to both identify errors in sentences by making
an appropriate correction and to state the rule which had been
broken. Where learners were able to do the former, but not the
latter, they claimed they were displaying 'implicit' knowledge. The
rules that were most consistently correctly articulated they term
'easy' rules. These include, "those that ( 1 ) referred to easily
recognized categories; (2) could be applied mechanically; (3) were
not dependent on large contexts: for example the morphological
dichotomies like a/an, who/which, straightforward cases o f
some/any, and simple word order" ( 1 992: 1 79). Hard rules, which
were rarely identified or articulated, involved permutations and
additions and deletions that were applied over larger structural
contexts, and invoked semantically opaque principles. These
included:
those that involve aspect, such as the use of the continuous form or the
perfect tense . . . These are semantic distinctions that express a speaker' s
perspective on a situation . . . They do not allow of simple exhaustive
descriptions and they are not always governed by features of the
immediate linguistic context (Green & Hecht 1 992: 1 80).
2.3.4 Processing load and rule complexity
The speculations about rule complexity reviewed immediately above
are diverse in the factors they consider, but all invoke, to some
degree, general information processing load as a determinant of rule
complexity. For example, degree of perceptual salience, size o f
context o f rule application and degree of semantic opacity have all
been claimed to affect the complexity of a rule, and all affect the
amount of attention and processing effort expended in learning and
remembering a rule (easy rules are supposedly those with
perceptually salient structural features, which are applied to small
contexts, and which involve transparent meaning-to-form
relationships). Only one of these proposals invokes a specific
linguistic theory to explain the complexity of the structures
regulated by the rule (a version of markedness theory in the case o f
Bardovi-Harlig 1 987), and she does so only to observe that its
36 Second language rule complexity
predictions fail to account for the difference in learning difficulty
the structures pose to learners.
As described in Figure 1 (see section 2. 1 .3) the complexity of a
pedagogic rule has two dimensions, the complexity of the structure
which the rule attempts to explain, and the complexity of the
explanation itself. While one may select the structures upon which
a rule is based according to the linguistically derived notions of
accessibility, generalizability and entailability described in Chapter
1 , the process of describing the structures in pedagogic terms creates
an additional set of processing demands over and above those
associated with attending to the structural features of the examples.
It is likely that such general information processing load
considerations, along with the evidence provided by classroom
contact with learners, are also influential in the decisions about the
relative complexity of rules made by experienced, practicing ESL
teachers. However, there is currently no agreed upon taxonomy o f
these processing demands, or an accepted model of the way they
interact with the complexity of the structure a pedagogic rule
describes which could be used as the basis of an objective,
scientifically approved metric of rule complexity.
2.3.5 The consistency of expert judgment in problem classification
and problem-solving
Given the lack of consensus over the nature of the objective criteria
to apply in distinguishing easy from hard rules described above how
are such rules to be identified? The solution adopted in the present
study was to use the expert judgment of experienced L2 teachers to
identify the rules by seeking to establish consensus about rule
complexity based on the subjective criteria underlying their shared
expertise. The procedure of using informants to rate the
complexity of problems, in the absence of any accepted a priori
criteria that could be used to direct classification, is a standard one
in experimental studies of problem-solving ability (see e.g., Reed,
Ackinclose & Voss 1 990). Second language teachers, in addition to
any formal training they may have in ESL or linguistics, also have
experience of teaching to draw on in making judgments about
pedagogic rule complexity. Evidence exists from studies of expert
judgment in other areas of the curriculum that consistency of
judgment in problem classification tasks can be achieved as a result
of formal training in a knowledge domain like Physics (see Chi,
Glaser & Rees 1 982). However, experience in applying th at
knowledge domain to practical problems in everyday life can also be
Second language rule complexity 3 7
drawn o n to establish consistency o f judgment, and there is some
evidence that the knowledge base established by experience is
superior to that established by purely formal academic training. For
example, Voss ( 1 986) found that success in solving real world
economics problems correlated more with practical experience than
with formal training in economics. Ceci and Liker ( 1 986) studied
the knowledge base underlying the expertise of expert horse racing
handicappers and found their decisions to be based on "a complex
internal calculus in which a number of variables are related to one
another in multiplicative rather than additive ways" (Best
1 992:484). Experts were not necessarily more intelligent than non­
experts (using measures obtained from the Wechsler Adult
Intelligence Scale), but experts did consistently outperform non­
experts in handicapping ability.
These studies lend support to the decision to ask teachers of ESL
with substantial experience to provide the estimate of complexity
used in identifying the dependent variables for the present study.
The procedure was felt to have the following advantages: (i) it
provided empirical support for an easy/ hard rule distinction; (ii) it
is replicable; (iii) it has high face validity, since experienced ESL
teachers are the group most qualified to make expert decisions about
the factors influencing the complexity of pedagogic rule formats;
and (iv) it was expected that such teachers would use predominantly
information processing load criteria similar in kind to those used by
Krashen to make their decisions about rule complexity-it was the
comparability of claims by Krashen with those of Reber regarding
the relationship of rule complexity to learning condition that the
study sought to address.
2.4 Empirically motivating the easy/hard rule distinction
The following steps were followed in identifying the easy and hard
rules for the present study. (i) A grammaticality judgment test o f
structures that were judged to range in degree o f complexity and
potential familiarity was given to a group of subjects. These subjects
were comparable to the intended subjects for the present study in
terms of their age, proficiency level, language background and
length of exposure to English. (ii) T�ose items which the subjects
most consistently and incorrectly rated ungrammatical were selected
as the basis for pedagogic rules. (iii) Pedagogic rules based on these
structures were written by the researcher by analyzing the elements
of the constructions. (iv) The resulting rules were given t o
experienced ESL teachers to rate for complexity using a seven
38 Second language rule complexity Second language rule complexity 39

point rating scale. (v) Subsequently, the same teachers were asked t o
rate the structures described by the rules for their complexity using a 1. * Who did John give the book to?
Q-sort of cards illustrating the structures. (vi) The results were 2. Who did you suggest I apply to?
examined to see if the teachers had distinguished between the 3. Who did you suggest that should apply?
complexity of candidate rules, and if their rating of the complexity 4. Who did you suggest should apply?
of the rules corresponded with the results of the Q-sort of the 5. Who did you suggest that I apply to ?
relevant structures. (vii) The rules identified in this way were then 6. * That there cow gives a lot Qf milk.
examined to see if a plausible explanation could be found confirming 7. That cow there gives a lot of milk.
the easy/hard rule distinction. These procedures are described ill 8. * To whom did John give the book?
detail below. 9. * What Mary does is eat cake.
1 O. * Where the cheese is is in the bag not in the
2.4.1 IdentiJYing unfamiliar structures basket.
For the purposes of the present study, the rules to be prese�ted � o 1 1. *John 's hitting him angered me.
_
learners, and the structures exemplifying them, had to be d1stmct m 1 2. There exists a copy of that book.
terms of their degree of complexity for the learner, and at the same 13. I sent to John the book that I read yesterday.
time unfamiliar to the learners. Unfamiliarity was necessary in order 14. Is what Mary does good?
to control for prior learning of the rules, which could have 15. *In the morning he jogs.
invalidated the claims for learning based on the treatments. 1 6. This cow here gives a lot of milk.
Although a number of criteria for distinguishing the com? l �xit� of 1 7. This here cow gives a lot of milk.
rules have been proposed it is clear that not all of these d1stmct1ons 1 8. It's Mary who eats cake.
apply to rules and structures that are also likely to be unfamiliar t o 19. Never have I seen him.
_
learners. That is, rules for third person ' s ' , o r wh-quest10n 20 . I saw the dog that barked.
formation cited by Krashen as easy and hard rules, for example, are 21. I saw the dog you fed.
likely to be known by most L2 learners of the intermediate level 22. I saw the dog that you fed.
targeted for the present study. 23 . I saw the dog barked.
..
To identify these rules a pilot study of rule fam1hanty was
.

24. Into the house John ran

conducted at the University of Hawai'i in the Spring semester o f 25 . *Into the house ran John.
1 993. Since the demands o f the training and transfer phase i n the
present study required more than basic �roficienc� in Eng�ish, it was * = items eventually selected as the basis of pedagogic
decided to target subjects of approximately mtermediate level. rules
Consequently, a comparable group of students was targeted for the
pilot study of rule unfamiliarity. Further, in o�der to ensure
. Figure 2 : The Targeted Items In The Pilot Study Of Rule
maximum generalizability the subjects for both studies had, ideally,
Familiarity
to share similar language learning backgrounds, and be o f
approximately the same age, with approximately the same length
and quality of exposure to English. It was decided to target Japanese
Accordingly, twenty students aged 1 9-20 years from a Japanese
learners of ESL who had completed school level education in
university, who were engaged in a two month intensive English
English, and who were either currently enrolled in or recently
program in Hawai' i, were used as subjects for the pilot study. The
graduated from a junior college or university, since these would be _
format adopted for the pilot study was a forty two item
representative of the great majority of subjects who would take part
grammaticality judgment test that required subjects to c �rcle
in the eventual study.
'grammatical,' 'ungrammatical,' or 'not sure' in response to wntten
 Second language rule complexity 4 1
40 Second language rule complexity

Table 1 . Percentage Scores Of Correct, Incorrect And Not

sentences (see Appendix A). Subjects were told to base their decision
Sure Responses To Items On The Pilot StudyTest Of Rule
on grammaticality, not on what they thought might be errors in
Familia rity
punctuation or spelling. Twenty-five target structures were chosen
for inclusion, and subsequent analysis, based on some of the claims
for differences in complexity previously reviewed in this chapter. Item no. % % %
These target structures included a number of sentences illustrating Correct Incorrect Not sure
the notions of Subjacency and the Empty Category Principle from
Universal Grammar (Bley Vroman et al. 1 988; White 1 989a). A l .* 39 56 5
number of focus constructions were also selected, since they involve 2. 28 56 16
unusual and sometimes complex permutations of word order, and 3. 33 51 16
because many of them are rarely covered in ESL/EFL textbooks 4. 16 51 33
(Celce-Murcia & Larsen-Freeman 1 98 3 : 403-4 1 6). Eighteen, 5. 66 11 23
mostly ungrammatical, items were added as distractors. The twenty­ 6.* 5 90 5
five target items are listed in Figure 2. 7. 28 56 16
Items 1 and 8 are examples of question formation with 8.* 66 23 11
preposition stranding and pied-piping. Items 2, 3, 4 and 5, 1 3 , 20, 9.* 0 1 00 0
2 1 , 22 and 23 are examples relating to Subjacency and the 1 0. * 5 90 5
possibility of 'that' deletion. Items 6 and 7 are examples of the pre­ 1 1 .* 28 56 16
and post-subject use of emphatic 'there' . Items 9 and 1 0 are pseudo­ 1 2. 23 72 5
cleft constructions. Item 1 1 is an example of gerundivization. Item 13. 79 16 5
1 2 is an example of the use of existential 'there' to focus on a verb. 14. 33 56 11
Item 14 is an example of a pseudo-cleft question form. Items 1 5 1 5.* 56 39 5
and 24 are examples of adverbial fronting of time and location 16. 23 61 16
phrases. Items 1 6 and 1 7 are similar to items 6 and 7, but with 1 7. 11 78 11
different demonstrative pronouns and adverbs. Item 1 8 is a cleft 1 8. 78 22 0
construction. Item 1 9 is an example of adverbial of frequency 19. 33 56 11
fronting, with subject-verb inversion. Item 25 is an example of 20. 67 33 0
adverbial of location fronting with subject-verb inversion. 21. 45 39 16
The percentage correct, incorrect and not sure for responses t o 22 . 90 5 5
these items are given in Table 1 . Two items, 9 and 2 5 , received 23 . 33 67 0
1 00% incorrect responses, while item 1 0 received 90% incorrect 24. 28 56 16
responses. It appeared, then, that pseudo-cleft structures were 25.* 0 1 00 0
unfamiliar to this group, as was the structure illustrating subject-verb
inversion following fronted adverbials of location. Item 6, with pre­
subject emphatic 'there' was also rated ungrammatical by 90% o f * = items selected as the basis of pedagogic rules
the subjects, although accuracy was higher for the related item
number 7, with post-subject emphatic 'there' . These items, together The other items chosen as the basis of pedagogic rules were item
with other examples of the targeted items that seemed obviously t o 1 5, adverbial of time fronting, which seemed to illustrate an easy
differ in complexity, were consequently selected as the basis o f the rule, and item 1 1 , gerundivization, which seemed to illustrate a more
pedagogic rules written by the researcher for the present study. complex rule. Items 1 and 8 were also chosen as the basis for writing
rules of wh-question formation with pied-piping and preposition
42 Second language rule complexity Second language rule complexity 43

stranding, since these structures had been claimed to differ in ordered and given to fifteen experienced teachers of ESL (mean
complexity for L2 learners (see Bardovi-Harlig 1 987 and section length of ESL teaching experience = 1 0.6 years, see Table 2).
2.3 . 1 above). However, such structures, and possibly therefore
versions of their associated pedagogic rules, were obviously familiar
to subjects, since the illustrative items, 1 and 8 , were judged Table 2. Details Of Teachers Who Took Part In The Rating
relatively accurately (only 55% and 20% inaccurate). Scale Survey Of Rule Complexity
It was hoped that rules based on the four clearly unfamiliar items
(6, 9, 1 0 and 25) would also be distinguished by the teachers in Teacher Ll ESL experience Qualifications
terms of their complexity. Rules based on the other four items, 1 , 8,
1 1 and 15 were primarily added as distractors to ensure an even
1. English 1 0 years B .A. American Literature,
distribution in terms of the relative complexity of rules offered as
M.A. ESL
choices. The rules were written by the researcher and subsequently
2. English 1 0 years B.A. Music, M.A.Linguistics,
presented to a group of 1 5 experienced teachers of ESL to rate fo r
M.Ed. Educational Foundations
complexity.
3. English 9 years B .A. Philosophy, M.A. Asian
It was decided that the format for identifying unfamiliar rules
Studies, M.A. ESL
adopted in the pilot study would also be used as the pretest of rule
4. English 20 years M.A. International Management,
knowledge for the present study. I have referred already to the
M.A. ESL
problems of using extensive pretests which consequently run the
5. English 1 5 years B.A. Asian Studies, M.A. ESL
risk of alerting subjects to the formal characteristics of the
6. English 6 years B .A. English Literature, M.A. ESL
structures to be presented during implicit, incidental and instructed
7. English 1 1 years B .A. Elementary Education,
conditions, and so invalidating claims about differences between the
M.A. ESL
conditions based on degree of rule awareness or consciousness of
8. English 3 years B.A. & M.A. American Literature,
form (see the review of studies by Hulstijn ( 1 990a) and Doughty
Ph.D English
( 1 988) in Chapter 1 ). The version of the pilot study used for the
9. Japanese 5 years B .A. American Literature,
pretest consisted of the first thirty items, to which subjects had t o
M.A. ESL
respond ' grammatical,' ' ungrammatical,' o r ' not sure ' . A s in the
1 0. English 8 years M.A. International Management
pilot study, during the pretest subjects were told to ignore what they
1 1. English 23 years B.A. English, M.A. ESL
thought may be punctuation or spelling errors and to focus on
12. English 1 4 years B .A. Speech & Drama,
grammaticality. ' Ungrammatical' responses to the two grammatical
R.S.A. Diploma T.E.F.L.
items subsequently identified as unfamiliar and differing in
13. English 9 years M.A. Japanese, M.A. ESL
complexity were used as the basis for selecting subjects for the
14. English 4 years M.B.A.
study. It was felt that responses to these items would be masked by
1 5. English 1 2 years M.A. American Literature
the task of making responses to other items in the pretest, and by
completing subsequent pretests of aptitude. Consequently it was felt
that this pretest would not alert subjects to the structures to be used
as exemplars of easy and hard rules.
The teachers were asked to rate each rule for complexity using a
2.4.2 Rating scale surveys of rule complexity scale from +3 for most complex, or hard, to -3 for least complex,
The rules written by the researcher, which were based on the eight or easy. Subsequently, after an interval of more than a week, the
items described above, are given in Appendix B. These rules, same teachers were asked to complete a Q-sort of the struct�res
together with instructions to rate each rule for the relative which were the focus of each of the eight rules. Each of the eight
complexity it would present to learners of ESL, were randomly structures was written on a separate card. The cards were randomly
44 Second language rule complexity Second language rule complexity 45

ordered before being presented to each teacher. The teachers were

instructed to sort the cards in order of the difficulty they felt 7
learning each of the structures would pose a non-native speaker of 6
English, with the most difficult at the top, and the least difficult at
5
the bottom. The teachers were also told that they could indicate
equivalent difficulty by clipping cards together using paper clips. .5 4
This separate measure of the difficulty of the structures was then
compared with the difficulty ratings for each rule given using the
� 3
rating scale response. In this way the correlation of the judgments 2
of the difficulty elicited via the rating scale survey, and via the Q­ 1
sort was examined with a view to establishing the consistency of the -11- Eas
0 -1-!:::;:::::;==::r:==;::=..�r--.--.-�r---r-"""T"--ir--,
teachers' judgments. 2 3 4 5 6 7 8 9 1 0 1 1 12 1 3 14 1 5
Teacher
2.4.3 The simple and complex rules identifiedfor the study
Following the procedure described above two rules were identified as
clearly distinct in complexity by teachers. These rules were also Figure 3. Results Of The Rating Scale Survey Of The
based on two of the items which the Japanese ESL learners had most Complexity Of Two Pedagogic Rules
consistently rated ungrammatical, item 1 0 (90% inaccurate) and
item 25 ( 1 00% inaccurate). The rules describing these structures
were therefore assumed to be potentially unfamiliar, and different
along a dimension of complexity. These rules are those for the 8
optional inversion of subject and verb following fronting cif
7
adverbials of direction (e.g., Into the house ran John or John ran),
and the more difficult rule for deriving pseudo-cleft sentences with a 6
focus on a prepositional phrase (e.g., Where the baby sleeps is in the 5
01!
bed not on the floor). =

Figure 3 gives the results of the rating scale survey of the � 4

complexity of these two rules, and Figure 4 gives the results of the � 3
Q-sort rating for the learning difficulty of the two sentences around 2
which the rules were based. As can be seen from Figure 3 , only one ___._ Hard structure
teacher failed to distinguish the two rules as differing in complexity. -II- Easy stncture
Scores for Figure 3 have been converted from the range -3 to +3 t o o ..i.--r--r--.--r---.---r---r--r--r--=;r==r=T=:;=T=,
a range from 1 to 7 , where 1 corresponds to -3, very easy, and 7 2 3 4 5 6 7 8 9 1 0 1 1 1 2 13 14 1 5
corresponds to +3, very hard. Teacher

Figure 4. Results Of The Q-Sort Of The Difficulty Of

Structures Described By The Pedagogic Rules
46 Second language rule complexity
As can be seen from Figure 4, all teachers distinguish between the
order of difficulty of the two structures described by the rules. Scores
for Figure 4 have been converted from the rank ordered Q-sort; a
score of 8 indicates the structure was stacked on top and judged to
be the most difficult, and a score of 1 indicates it was stacked at the
bottom and judged to be the easiest. The formats for these rules
presented to teachers to rate for complexity are given in Figures 5
and 6 below, and the complete set of eight rule formats used in the
rating scale survey are given in Appendix B.
RULE 1
Some sentences contain a suqject a verb and a location phrase or a
time phrase. Here are some examples:
Joan (subject) crashed (verb) into the wall. (location phrase)
John (subject) slept (verb) in the morninf{. (time phrase)
Both location and time phrases can be moved to the front of the
sentence:
Into the ·wall Joan crashed.
On Tuesday morninf{ John slept.
The verb can also come before the suqject, but only when location
phrases are at the front of the sentence:
Into the wall (location phrase) crashed (verb) Joan. (su�ject)
* On Tuesday morninf{ (time phrase) slept (verb) John. (subject)
* means ungrammatical
Figure 5. Format Of The Word Order Rule Of Focus
Presented To Teachers To Rate For Complexity
Second language rule complexity 47
RULE 2
Some sentences tell about the locations of two things e.g.,
Alice stands on the rif{ht and Judy stands on the lefi.
We can change this sentence to focus on one of the locations e.g.,
Where Alice stands is on the rif{ht.
To make sentences like these, first, choose the su�ject you want t o
emphasize, e.g., 'Alice', then place 'where' in front o f it:
Where Alice...
Next, follow the subject with the verb, e.g., ' stands' :
Where Alice (su�iect) stands (verb) ...
Note that the verb cannot come before the subject:
* Where stands (verb) Alice (su�ject) . . .
* means ungrammatical
Next add a singular form of the verb 'be' which agrees in tense
followed by the phrase describing the location of the subject:
Where Alice stands is on the rif{ht.
Where Alice stood was on the rif{ht.
If the verbs do not agree in tense the sentence is ungrammatical:
* Where Alice stands was on the ri�ht.
For special emphasis, the second location can be contrasted with
the focus location by joining the locations using 'not'. This 'not' is
required. Without it the sentence is ungrammatical.
Where Alice stands is on the right not on the left.
*Where Alice stands is on the right the left.
Figure 6. Format Of The Pseudo-Cleft Rule Of Focus
Presented To Teachers To Rate For Complexity
48 Second language rule complexity
Into the house John ran.
Into the house ran John.
(- deletion - addition + permutation)
/ knowledge of constraints on possible word order permutations, but
Figure 7. Elements Of The 'Easy' Word Order Rule Of Focus
The baby sleeps in the bed and the dog sleeps on the floor.
(Where) the baby sleeps (is) in the bed (not) on the floor.
addition addition deletion addition
+agreement +agreement -agreement
(location) (tense)
(+ addition + deletion - permutation)
Figure 8. Elements Of The 'Hard' Pseudo-Cleft Rule Of
Focus
Once the easy and hard rules were identified, a number of post­
hoc comparisons and contrasts were made between them. Both of
these rules are rules of 'focus' . Celce-Murcia and Larsen-Freeman
( 1 983) note that pseudo-clefts in particular, like other rules o f
focus, chara�terize spoken rather than written language, "which
may explam _
m part why they have often been ignored in ESL/EFL
texts" 0 ? 83 :408). In the case of adverbial fronting (see Figure 7)
. switch positions, but if time is at the front they can't), whereas this
the rule mvolves constramts on permutation, or changes in word
order, of the type that allows a sentence-internal constituent (the
s�bj �ct, 'John' � to be optionally placed in sentence-final position.
S1m1lar adverbially conditioned constraints on subject-verb inversion
have been observed in languages other than English (see Bresnan &
Kanerva 1 989).
In contr�st, the pseu�o-cleft construction (see Figure 8) appears
to be specific to Enghsh. The rule involves constraints on the
addition of items requiring sentence-internal agreement and
deletion of unnecessary items. However, in contrast to the r�le for
fronting adverbials of direction, which allows subject-verb inversion,
Second language rule complexity 49
in pseudo-clefts the canonical SVO word order is preserved (for
discussion of pseudo-cleft structures see Baker 1 9 8 9: 3 6 7-3 7 1 ;
Celce-Murcia & Larsen-Freeman 1 983 :404-408; Declerck 1 994;
Fletcher 1 980; Mccawley 1 988:59-6 1 ; Prince 1 978).
The two rules can be distinguished, then, in that Rule 1 requires
no knowledge of possible additions or deletions, while Rule 2
requires knowledge of constraints on additions and deletions, but n o
knowledge o f word order permutations. The word order
permutations described by Rule 1 are typical of those required at
Stage 3 (the final-to-initial movement of adverbials of time and
location), and Stage 4 (the optional internal-to-final movement o f
the subject to end position following fronting o f adverbials of
location) of Pienemann, Johnston and Brindley's ( 1 98 8) proposed
taxonomy of word order stages for the acquisition of English. The
constraints on addition of elements involved in Rule 2 requires
knowledge of agreement between the wh-word used and the focused
location phrase, knowledge of the contrasting location phrase which
is preceded by 'not', and knowledge of tense agreement between the
two verbs used. The deletion constraint requires the learner t o
identify and omit the subject and predicate o f the parallel, non­
focused clause (see Figure 8). These more complex sentence­
internal operations are characteristic of Pienemann' s et al. 's Stage
5 structures (see Pienemann et al. 1 988 :224).
When asked to comment on why they had judged the two rules
to differ in complexity, teachers frequently referred to the length of
the rules, (the easy rule was shorter), and the number of conditions
on the rules, (the hard rule had a greater number of conditions). A
number commented that it was possible to reduce the easy rule to a
'rule of thumb,' (if place is at the front then subject and verb can
was not possible with the hard rule. Many also commented that the
use of two verbs in the hard rule was unusual and possibly confusing,
as was the fact that the wh-word fronting the sentences had n o
interrogative function. The use o f these rules to motivate the
classification of sentence types presented in the training and
transfer sessions of the present study is described in detail in
Chapter 5, sections 5 .3 .4-5 .
 Chapter 3

Implicit And Explicit Information

Processing

3. 0 Introduction
The aim of this chapter is to address in greater detail the issue of
the relationship between noticing, consciousness and the operations
of implicit and explicit processes during language learning. I n
sections 3.2 and 3 .3 two causal variables in explanations o f
dissociations between implicit and explicit processes, the concepts
of attention and memory, are described and related to models o f
information processing and definitions o f consciousness i n
experimental psychology and SLA theory. I n particular the role o f
'noticing' i n filter theories and capacity theories o f attention is
described and related to three functional classifications of
attentional resources; 'alerting,' ' orienting' and 'detecting' . Various
type classifications of memory, 'episodic' versus ' semantic,'
'procedural' versus 'declarative,' and 'short-term' versus ' long­
term,' are then described, and the question of whether these are
indeed different systems, or simply functional distinctions, is
discussed. Following this a summary model of the relationship o f
attention to memory during implicit and explicit processing is
proposed. Finally, in section 3 .4 of this chapter the issue of
individual differences in language learning aptitude is considered ill
light of the proposed model.

3.1 Resources, information processing and consciousness

Schmidt's claim that awareness at the level of noticing is necessary
for converting L2 input to intake invokes, but does not explain in
detail, attentional mechanisms and their relationship to encoding
and retrieval from the various subsystems of memory that have
been proposed. Neither are these attentional mechanisms and
information processing relationships explained in detail in other
proposals which, following Schmidt, have attributed an important
role to noticing in L2 pedagogy (e.g., Ellis l 993a, 1 994b, 1 994c,
52 Implicit and explicit information processing
1 99?; Fotos 1 993 ; Fotos & Ellis 1 99 1 ; Long 1 994; Zalewski 1 99 3 ) .
While th�re have b.een attempts to describe the relationship
between mput and mtake at the neurophysiological level o f
explanation (Sato & Jacobs 1 992) there has been, to date, n o
attempt at a systems-level characterization of why attention is
allocated to input under certain task conditions and not others.
Attentional mechanisms, in Schmidt's view, are causal factors in
L2 learning, since they are responsible for allocating the cognitive
resou�ces . that lead to .noticing, and subsequent encoding in m emory.
Schmidt is clear, as is Reber, about the necessity of attention to
input during both implicit and explicit learning. There are, though,
currently competing characterizations of the role of attention
during implicit information processing, with one recent explanation
of implicit learning even claiming that the learning effects observed
during a sequence learning task are due to a 'nonattentional' form
of processing (Curran & Keele 1 993).
Part of the difficulty in motivating an explanation of such
mechanisms using current cognitive theory lies in the fact, as
Baddeley ( 1 986:225) notes, that the study of attention has been
d?minated by theories of the role of attention in perception and
visual processes, particularly signal detection and pattern
recognition. The role of attention in the control of memory and
action, arguably areas of greater potential application to
descriptions of SLA processes, have, until recently, been less well
studied. Recent work on the memory/attention interface has
progressed considerably beyond the early multistore model of
m �m �ry and attentional control of Atkinson and Shiffrin ( 1 968),
reJectmg many of its fundamental assumptions (Rumelhart &
McClelland 1 985), though as Cowan ( 1 988) notes, cognitive
psychology has yet to settle on an accepted view of the mutual
constraints imposed by memory and attention during information
processing. Two established frameworks for describing skill
development and performance, Shiffrin and Schneider's theory o f
automaticity ( 1 977) and Anderson's ACT* theory o f skill
acquisition ( 1 983), have been heavily cited in the SLA literature
(e.g., Ellis 1 993a, 1 993b; Faerch & Kasper 1 984; Kohonen 1 992;
McLaughlin 1 987; Mohle & Raupach 1 989; O'Malley & Chamot
1 990; Robinson 1 989; Schmidt 1 992). However, more recent work
in the study of action (see Holding 1 989; Navon 1 984; Navon &
Gopher 1 ? 80; Schneider & Detweiler 1 988; Wickens 1 980, 1 9 84,
1 989) which has studied the role of attentional processes in skilled
performance using the dual-task paradigm for examining attentional
Implicit and explicit information processing 53
allocation, has rarely been invoked by SLA researchers, despite its
potential relevance to such current issues in SLA as task complexity
and grading in task-based approaches to syllabus design (Long &
Crookes 1 992; Nunan 1 988, 1 989; Robinson 1 993b, 1 99 5 a;
Robinson & Ross 1 996; Robinson, Ting & Urwin 1 995).
Part of the difficulty in motivating a theory of attentional
mechanisms in SLA by drawing upon an accepted body of relevant
findings from cognitive psychology research lies also in the recent
challenge to traditional information processing accounts o f
attention posed by more recent connectionist accounts.
Information processing models, such as that of Broadbent ( 1 9 5 8),
view attention as an executive process directing the serial passage of
information between separate short-term and long-term memory
stores. In contrast, connectionist accounts dispute the modular
metaphor for cognitive architecture that the information
processing views are based on, as well as the assumption of seriality,
arguing that executive attentional control is distributed throughout
the entire processing system, in local patterns of neuronal
excitation and inhibition, rather than in some central executive
processor. Recent attempts to reconcile connectionist and control
architectures in the study of attention have been proposed (e.g.,
Schneider 1 993; Schneider & Detweiler 1 988) but are not yet widely
accepted.
3.2 Attention
The concept of attention has three uses. It can be used to describe
the processes involved in 'selecting' the information to be
processed and stored in memory. Dichotic listening tasks, for
example, have been used to examine the fact that attention has a
variable focus, and can select information to be processed, to the
exclusion of other information (Cherry 1 953; Moray 1 959). It can
be used to describe our ' capacity' for processing information.
Studies of divided attention show that attention is capacity limited
and that decrements in performance increase as the number of task
dimensions, or components to be processed increase (Taylor,
Lindsay & Forbes 1 967). Finally it can be used to describe the
mental 'effort' involved in processing information. Pupillary
dilation, for example, can be measured as a physical index of the
degree of mental effort required in attending to increasingly
complex tasks (Kahneman 1 973). Each of these uses has influenced
the development of theories of attention.
54 Implicit and explicit information processing Implicit and explicit information processing 5 5

3.2.1 Filter theories of attention concurrently are not excessive (see Figure 9). I n this respect
Early filter theories of attention were based on pipeline models o f capacity theories such as Kahneman's differ from filter theories,
information processing, in which information i s conveyed i n a fixed which characterize incoming stimuli as inevitably involved in
serial order from one storage structure to the next. In Broadbent's competition for limited resources.
( 1 95 8) 'bottleneck' model voluntary control of information
processing is exercised by a selective attention mechanism or filter
which selects information from a sensory register and relays it to a
Miscellaneous
detection device. Once past the selective filter, information is manifestalions
Arousal
analyzed for meaning, rather than for physical properties alone, of arousal

enters awareness, and is encoded in short-term memory. Treismah

( 1 964 ), based on evidence that subjects noticed their own name �
: Available :

when repeated in an unshadowed ear during dichotic listening tasks, ; capacity ;

Enduring .. ..
argued for a filter mechanism that was sensitive to semantic disposi tions

information, as well as sensory information, and against the

proposal that all information in an unselected channel is completely
tuned out, and unavailable for detection. Arguments against
Treisman's attenuated filter model are that the preattentive Momentary
---" ,
/
.
'
.----­
'

processing, or analysis before detection, it requires is too complete

I ' '
intentions ' I '

and resource demanding. Late selection theories (Norman 1 96 8 ;

Watanabe 1 980) propose that all information i s processed i n
01t�------���o
Possible activities
parallel, and enters working memory, where a decision is made about
the importance of the information. Information that is judged
important is elaborated or rehearsed; that which is judged
unimportant is not selected for elaboration and is forgotten. Responses

3.2.2 Capacity theories of attention (from Kahneman 1 973 : 1 0)

Underlying filter theories of attention and their associated
mechanisms of selective attention is the metaphor of a limited
capacity channel, in which information competes for limited
attentional resources available to the passive processor. More
recent theories emphasize the voluntariness of the subject's control
of attentional resources and the task specificity of decisions about
attentional allocation. The metaphor most suited to these theories
is that of attention as a spotlight with a variable focus, which can be
narrowed and intensified, or broadened and dissipated, as task
conditions demand. In Kahneman' s model ( 1 973) resources are
allocated to incoming stimuli from a pool of cognitive resources
that varies as a function of the subject's state of arousal. Allocation
is divided between enduring predispositions (e.g., to recognize one ' s
own name) and momentary intentions (e.g., to eavesdrop). Divided
attention does not necessarily lead to decrements in performance,
given sufficient arousal and that the demands of the tasks performed
Figure 9. Kahneman's Capacity Model Of Attention
Wickens ( 1 980, 1 984, 1 989) has recently expanded Kahneman's
view o f attentional resource allocation, arguing that rather than a
single pool of resources, there are multiple pools. These pools
occupy different points on three intersecting dimensions of resource
systems. These are: (a) the dimension representing perceptual/
cognitive activities versus response processes; (b) the dimension
representing processing codes required by analog/spatial activities
versus verbal linguistic activities; and (c) the dimension representing
processing modalities e.g., auditory versus visual perception and
vocal versus manual responses (see Figure 1 0). Wickens ( 1 980,
1 984, 1 989) argues that the attentional demands of tasks, and so
 56 Implicit and explicit information processing
Implicit and explicit information processing 57

their relative difficulty, will be increased when concurrently
performed tasks draw simultaneously on the same pool of resources.
In the worst cases, like studying for two exams, interference effects
_
may m �ke time-sharing impossible and the attentional mechanism
of 'senal processing' will be adopted whereby separate task
compo!lents are completed in succession. In other cases where
'
there is less 'global similarity' between the tasks and so less
'
resourc� competition, as in driving a car while talking, the
m �chamsm o � parallel processing may be adopted. However, while
this may avoid the need for time-swapping, degradations in the
_ which draws on distinct, not identical resource pools, smce m this
quality of the attention allocated to both activities may lead t o
poor performance. When tasks draw o n completely different pools
of resources, or when one of the tasks is automatized then
successf�l time-sharing and dual-task performance is possib le. I n
these ctrcu1!1stances the me�hanism of 'parallel processing' is
alway� appl1ed .. However, Wickens notes, individuals may differ
_
both m thetr time-sharing ability, and in their store of available
resources : therefore individual differences, as well as task
_
charactenstics, may determine which of these two mechanisms of
attentional allocation is adopted.
"""'"(11----
STAGE S
Central
Encoding Processing Responding
' development in SLA, is a conceptual model for relating th� extent
t
�
Sootiol
3.2.3 Noticing and attentional theory
Capacity theories, like those of Kahneman and Wickens, are
properly seen as extensions of the late selection filter theory. I n
addition to mechanisms o f selective attention, they propose
mechanisms for adjusting the deployment of attentional resources
to suit the particular conditions of task demands. In this �ense more
recent attentional theory provides a framework for relatmg the act
of noticing to those task conditions which facilitate it. Noticing
following detection is more likely to occur in dual-task p erfor� an� e
_
case more attentional resources can be allocated to input processing.
It could be argued that Wickens' distinctions between the resource
pools drawn on in task performance are still too co�rse grained t o
allow a successful analysis of the workload and attentional demands
of L2 tasks. Schneider and Detweiler ( 1 988) argue that mechanisms
are insufficiently specified in Wickens' model, proposing a five
phase model of skill acquisition which specifies the type of
attention switching that occurs at each level.
However, the dispute between cognitive theorists over the
proper characterization of the mechanisms . used to . resolve
competing task demands is arguably one which is _ o ts1?e the
�
preserve of SLA theory construction per se, but ?ne which it m u�t
be informed by, and conceivably, in tum, contnbute to. What is
necessary, I would suggest, at the current stage of theory
_

R-------''<--..::....
Manual '?<"J',o
o�
to which subtle changes in task demands can affect the cond1t10ns of
implicit and explicit information processing. Th� ne�� for such a
i;! J'<"
model is prompted by the fact that the study of 1mp �1c1� proc�sses,
Visual Verbal Vocal
�
<t
0
J'� in laboratory settings like those of Reber ( 1 989), is mcreasmgly
0
acknowledging the fact that implicit processi?� is not a
:::<
I _
Auditory
homogeneous, undifferentiated alternative to exRl � c1t processm�,
but is itself susceptible to variance in task cond1t1ons, and their
�c 00
Spatial different processing consequences (Shanks & St. John 1 � 94;
� Whittlesea & Dorkin 1 993). It is prompted also by the observat10ns
J'�Vertiol
of Schmidt ( 1 990 : 1 43) that task demands are powerful determinants
of what gets noticed, and by speculation that certain L2 tasks may,
of their nature, facilitate noticing of different amounts, or aspects
(from Wickens 1 984: 69) of input (Ellis 1 995; Fotos & Ellis 1 99 1 ; Long 1 989; Loschky &
Bley-Vroman 1 993; Rankin 1 990; Robinson 1 995a; Robi�son et a !.
Figu re 1 0. Wic ken s' Mod el Of The
Stru ctur e Of Mul tiple 1 995). Wickens' model is at a sufficient level of generality that it
Reso urce s can allow the resource demands of both controlled laboratory tasks
and more complex classroom tasks to be categorized with a view to
58 Implicit and explicit information processing
identifying the extent to which such resource demands implicate, or
create the conditions for implicit and explicit processes.
3.2.4 Alertness, orientation and detection
While theories of attention, then, have moved away from their
early preoccupation with problems of signal detection, and the
question of 'where' selective attention occurs in relation to the
information processing sequence connecting the sensory register
and short-term memory, there is still a need for a theory which
identifies the mechanism whereby, and the point at which, selective
attention takes place. Signal detection theory, acknowledging the
parallelism of input processing, has moved from unidimensional t o
multidimensional formats for statistically modeling this process (see
Cohen & Massaro 1 992; Kadlec & Townsend 1 992), though as
Mandler ( 1 992) points out, purely connectionist accounts o f
parallel input processing and representation are left without a
mechanism for selectively directing attention to input, suggesting
the need for "consciousness as an intervening limited serial
process," functioning as a "gate between external information and
internal representations" ( 1 992 : 54).
Tomlin and Villa ( 1 994), purposefully avoiding reference to
consciousness, argue that a fine grained analysis of the process of
detection that leads to noticing is also necessary for SLA theory,
and distinguish between the alerting, orienting and detection
functions of attention during the allocation of selective attention.
Alertness concerns an individual' s "general readiness to deal with
incoming stimuli or data" (Tomlin & Villa 1 994: 1 90). Warning
signals, for example, can increase alertness or vigilance leading t o
an increased likelihood o f signal detection. Orientation concerns the
allocation of resources based on expectations about the particular
class of incoming sensory information, and involves activation of
some higher level schema or plan of action and events. During
detection a specific bit of information is engaged by attention.
Detection is more demanding of attentional resources and makes
further processing of a stimulus possible at higher levels, such as
storage and rehearsal in short-term memory.
Tomlin and Villa claim that alertness is of interest to SLA to the
extent that learners must be ready to process information.
Orientation is of interest to the extent that prior experience m ay
predispose learners to attend, for example, to form or meaning in
processing a stimulus. In this sense orientation, and the related
processes of input facilitation and inhibition, are clearly implicated
Implicit and explicit information processing 59
in transfer effects in task performance. A trammg task making
similar demands to a transfer task will facilitate detection on the
latter, since the subject is oriented to the particular stimuli relevant
to meet its demands, whereas differences in the processing demands
of training and transfer tasks will lead to inhibition, or a relative
delay in allocating attention to the detection of relevant stimuli.
3.2.5 Attention, awareness and detection
It is clear that of the three functions of attention just described it is
detection that is most similar to what Schmidt terms ( 1 990)
noticing. Detection is responsible for encoding in memory, and
therefore is the attentional level at which Tomlin and Villa claim
( 1 994: 1 92) learning must begin. However, Tomlin and Villa point
out that detection does not necessarily· imply awareness. There can
be detection without awareness (e.g., subliminally presented items
can cause semantic priming, see Balota 1 983, 1 986; Cowan
1 988: 1 74). It appears to follow from Tomlin and Villa's definition
of detection and awareness, then, that there can be learning without
awareness. This brings them, apparently, into conflict with Schmidt,
who claims conscious noticing is necessary for learning.
These different positions can be reconciled if the concept o f
noticing i s defined to mean detection plus rehearsal i n short-term
memory, prior to encoding in long-term memory, a view associated
with a number of late selection theories of attentional allocation
(Cowan 1 988, 1 993; Norman 1 968). Although some have suggested
that the contents of awareness and the contents of short-term
storage are identical (e.g., Stem 1 985), the position illustrated in
Figure 1 1 is that activation in short-term memory must exceed a
certain threshold before it becomes part of awareness (Cowan
1 988 : 1 65; Shiffrin 1 993 : 1 95), and further, that short-term memory
is the subset of long-term memory in a currently active state (see
section 3 .3 . 1 ). In this way noticing can be identified with what is
both detected and then further activated following the allocation of
attentional resources from a central executive. Rehearsal (Baddeley
1 986:99) following detection would be a consequence of the
allocation of resources to fulfill task demands, as, for example, in
the capacity model of Wickens ( 1 989). The nature of rehearsal and
elaboration would vary according to whether the task demanded
data-driven or conceptually driven processing (see Jacoby 1 9 8 3 ;
Roediger, Weldon & Challis 1 9 89 :24, and section 3 .2.7). I n data­
driven processing stimuli are encoded in "small pieces which are
later assembled in working memory" (Best 1 992:76). In contrast,
60 Implicit and explicit information processing Implicit and explicit information processing 6 1

conceptually driven processing involves more effortful integration
of encoded stimuli within the context of surrounding stimuli,
drawing on 'expectations or plans,' themselves the result of the
activation of schemata in long-term memory.
These different types of processing are initiated in response t o
task demands, and affect the nature o f encoding. Roediger et al.
( 1 989) claim that the nature of the encoding required during study is
a powerful predictor of performance during recognition tests, and
that "memory tests benefit to the extent that the operations
required at test recapitulate or overlap the encoding operations
performed during prior learning" ( 1 989: 1 6). They point out that
effects supposedly demonstrating dissociations between
performance on implicit and explicit memory tests can largely be
explained by the fact that in implicit tests encoding and retrieval
demands are matched, whereas this is not the case in explicit tests
(see section 4.2 for further discussion). In the context of Figure 1 1
the roles of detection, attention and awareness in learning are now
briefly summarized.
.,,. ... - .... ....
..
tection ,. '' , ,. noticing ' '
claim, but such subliminal exposure effects are unlikely to have
effects over intervals longer than a few hundred milliseconds, are
rapidly lost from memory, and cannot in any useful sense be
claimed to be evidence of learning (Bolender 1 986; Shanks & St.
John 1 994).
(ii) A ttention. Both detection and noticing require attention.
Consequently, the view taken here is that there can be no learning,
or encoding in memory, without attention. Although some have
described particular forms of learning as being non-attentional
(Curran & Keele 1 993), this description refers to the degree o f
attention directed at the stimulus, and such learning mechanisms are
properly considered less attention demanding, not attentionless.
(iii) Awareness. Awareness is critical to noticing, and
distinguishes it from simple detection. Noticing is a consequence o f
encoding in short-term memory, and is necessary for learning.
What is noticed may be subsequently encoded in long-term episodic
memory. It is possible to briefly notice and forget, and to notice
and remember over time. More permanent encoding in long-term
memory is a consequence of the level of activation of information
Long termmemory in short-term memory, itself the result of rehearsal and elaboration.
The nature of the rehearsal in short-term memory is a consequence
of the processing demands of particular tasks, and can involve data­
,\ driven processing, and conceptually driven processing (see Jacoby

;_
ttention : +attention & rehearsal -> ', - > elaboration 1 983). Data-driven processing involves simple maintenance
' :"<.. rehearsal of instances in memory. Conceptually driven processing
-•worene� • +owo�e� : / involves elaborative rehearsal and the activation of schemata or
higher order relations from long-term memory which are used t o

�-;;;J
Figure 1 1 . Noticing As Detection With Awareness In Short­
Term Memory
(i) Detection. Detection requires attention. There is evidence
that detection can occur without awareness, as Tomlin and Villa
organize instances into more abstract configurations and can be
induced, for example, by the instruction to search for rules
underlying a sequence of stimuli. Measures of awareness are difficult
to operationalize given that (i) the experience of noticing may be
fleeting and thus difficult to recall, and that (ii) one may be aware
of, yet unable to verbalize or otherwise articulate the nature of that
which one is aware of.
3.2. 6 Attentional and non-attentional learning versus encoding
specificity
The position taken here is that all learning requires attention.
Curran and Keele ( 1 993) claim that performance on a sequence
learning task using a dual-task methodology reveals effects for a
62 Implicit and explicit information processing
non-attentional form of learning. Their results can be explained by
invoking the notion of encoding specificity and the distinction
between data-driven and conceptually driven processing, both
mediated by attention, illustrated in Figure 1 1 .
(i) Single-task training and dual-task transfer. Curran and Keele
( 1 993) used a single-task training to dual-task transfer paradigm for
examining the influence of attention on the learning of sequences
of stimuli (the letter X) displayed in numbered quadrants of a
computer screen. One group were shown the rules regulating the
presentation of stimuli in these numbered quadrants, another group
were not. Subsequently, subjects in the non-instructed group were
divided into aware and unaware groups on the basis of their ability to
correctly identify the nature of the - rules regulating the presented
sequences. The finding was that learning, as measured in decreases in
reaction times, was better for instructed learners, and for aware non­
instructed learners than for unaware non-instructed learners in
single-task conditions. All groups showed evidence of learning, in
the sense of improved reaction times over the period of single-task
training. However, when a secondary task was added, counting high
pitched tones occurring during each trial of the transfer session, the
performances of the two groups were at the same level, though
again the performances of all groups showed evidence of learning, in
the sense of relatively faster reaction times compared to initial
practice. Curran and Keele argue that this is evidence of two forms
of learning, an attentional form, displayed by the instructed and
aware non-instructed learners under single task conditions, and non­
attentional learning, displayed by the non-aware subjects under
single task conditions. The attentional form is degraded by the
attentional demands of the secondary task, but the non-attentional
form is not. Since the instructed and aware non-instructed subjects
showed some evidence of learning in the dual-task transfer phase,
Curran and Keele argue that this is the result of a switch to the non­
attentional form of learning that had been occurring in parallel with
the attentional form in the single-task training sessions. Curran and
Keele conclude that this is evidence for two distinct, non-interfaced,
parallel forms of learning. They thus· differ from Hayes and
Broadbent ( 1 988) who claim that selective, attention demanding,
and unselective, less attention demanding forms of learning
compete with each other, and are applied serially.
Implicit and explicit information processing 63
(ii) Attention and awareness. It is clear, though, from Curran and
Keele's discussion that both forms of learning require attention. It is
less clear that the two forms of learning they identify are
differentially dependent on awareness. Awareness was assessed as the
ability to verbalize the rules of the sequence in answer to a question
about whether they noticed or could describe a pattern underlying
the stimulus presentations. This is precisely the awareness induced
by the task instructions for the instructed group, and the
consequence of the interpretation of the task made by the aware
non-instructed group. But awareness may also be of the frequency of
repetition of co-occurring stimuli. This may not qualify as a
pattern, or rule, and so subjects may have failed to report it, but it
could still have contributed to performance, and such co­
occurrences could have been consciously noticed. This type o f
awareness was not assessed by Curran and Keele, but it is precisely
the type of awareness one could expect the learners in the non­
attentional mode to have had. Shanks and St. John ( 1 994) make the
same point: "If subjects have learned something other than rules,
then asking them about rules may lead to erroneous conclusions. O n
the other hand, i f we ask the subjects questions about what they did
in fact learn, we may get reasonable answers" ( 1 994:427).
Awareness, that is, is a function of the interpretation of the nature
of the encoding and retrieval processes required by the task, and
both forms of awareness I have described are likely to have
accompanied performance by subjects in the experiment (cf.
Schmidt 1 995; Stadler 1 995). The problem lies in the insensitivity
of Curran and Keele's measure of awareness.
(iii) Encoding specificity and awareness. Rather than invoking
two distinct, non-interfaced forms of learning, one could invoke the
distinction between data-driven and conceptually driven processing,
as a consequence of task demands, to account for Curran and
Keele's results. Data-driven processing requires accumulation and
rehearsal of instances encountered in the input in memory, and m ay
lead to the development of simple patterns of association, or the
strength of dependency between co-occurring items. Conceptually
driven processing requires the elaboration of input following
activation of schemata for, in this case, sequence relations, and the
attendant rehearsal of more abstract patterns of hierarchical
organization. Both require awareness, though the nature of what is
noticed during processing will be relative to the interpretation
placed on the task by the learner.
64 Implicit and explicit information processing
3.2. 7 Summary of the role of attention
Four . poi �ts can . now be made about the model illustrated in Figure
_ the consciousness hypothesis of Schmidt
1 1 . ( 1) It is consistent with
( 1 990) which claims there is no learning without awareness at the
level of noticing. (ii) It is consistent with one interpretation o f
claims by Reber ( 1 989) and Krashen ( 1 98 l a, 1 982) that learning is
!he result of both explicit and implicit information processing; this
IS th �t explici� and implicit information processing require
conscious attention to form at input, but that implicit information
processing is data-driven and results in the accumulation o f
in�tance � , whereas explicit information processing is conceptually
driven, mvolvmg _ access to schemata in long-term memory. (iii)
The . n_iodel aJlows little, if any influence (outside of fleeting
subhmmal exposure effects) of nonconscious information
processing on the accumulation of L2 knowledge. (iv) It is
consistent with the position that dissociations in the extent o f
knowledge, and awareness o f knowledge arising during learning, are
consequences of the particular encoding and retrieval operations
requi�ed by particular tasks (Green & Shanks 1 993 ; Jacoby 1 98 3 ;
Roediger et al. 1 989), which implicates the operation of a central
executive attentional mechanism allocating limited resources t o
satisfy known task demands (Whittlesea & Dorken 1 993), rather
than the operation of two distinct, consciously and unconsciously
accessed systems.
3.3 Memory
The debate about the role of consciousness in L2 learning has, to
date, involved considerable speculation about the role of attention
in information yrocessing, but little discussion of the memory
systems underlymg such processing. Possibly this neglect is due t o
the desire by contemporary theorists and practitioners alike t o
disso�iate t�emselves from the audio-lingual approach to language
teachmg which placed a heavy emphasis on memorization and rote
learning of patterns. It may also be due to the sense, perhaps
confirmed by the discussion so far, that descriptions of memory
should properly take place within the context of a theory o f
attention and its role in information processing. This may be true,
but there are reasons for thinking that discussions of memory are
due to receive more prominent treatment in SLA theory than they
have to date. Firstly, more extensive evidence of the
neu�ophysiological substrata ?f memory are currently becoming
available, as a result of experimental methodologies like magnetic
Implicit and explicit information processing 65
resonance imaging and evoked brain potential (Cotman & Lynch
1 989; Squire & Zola-Morgan 1 99 1 ; Thompson 1 986), which appear
to lend firmer support to previously speculative distinctions
between memory systems. Secondly, a rapidly expanding literature
on dissociations between conscious awareness and performance o n
'implicit' versus ' explicit' memory tests has been interpreted as
supporting the claim that the separate systems support functionally
differentiated forms of learning. This literature is critically reviewed
in Chapter 4. The remainder of this chapter is devoted to a review
of the separate memory systems that have been proposed, the
extent of their supposed implication in implicit and explicit
information processing, and a summary of the evidence for
individual differences in L2 learning ability as a function of the
availability of cognitive resources like memory and attention.
3.3.1 Short-term and long-term memory
Most current models of memory assume the need for a functional
distinction between short-term and long-term storage (Raaijmakers
& Shiffrin 1 992), although whether this is represented as a
distinction between separate systems and stores (Shiffrin 1 993), or
whether the separate storage functions take place within a single
interconnected workspace (Baddeley & Hitch 1 993 ; Carlson, Khoo,
Yaure & Schneider 1 990) is currently a matter of dispute.
Proceduralist views of memory incline to the latter position, and
are evolutions of Hebb's views ( 1 949) that one mechanism for the
storage of memories "was the continued activity, or reverberation,
of the cells and cell assemblies recruited by a perceptual act"
(Crowder 1 993 : 1 42), and that should this reverberation continue
long enough, a mechanism causing structural changes in the neural
units would lead to indefinite memory. Versions of proceduralist
positions include those of Cowan ( 1 993), that short-term memory
is the subset of long-term memory in a current state of activation,
and the connectionist/control architecture of Schneider ( 1 993 ).
Even those who argue for a more modular approach (Shiffrin 1 99 3)
agree that the early modal model of Atkinson and Shiffrin ( 1 96 8),
which proposed distinct, serially connected, short and long-term
stores, with separate phonological and semantic codes respectively,
is wrong. All are in agreement about three basic properties of short­
term storage: (i) it involves temporary activation of neural
structures; (ii) it is the site of control processes such as directing
focal and peripheral attention, rehearsing current information and
coding new inputs; and (iii) it is capacity limited. As such short-term
 66 Implicit and explicit information processing
memo� "serves as the interface between everything we know and
everythmg we can see or do" (Cowan 1 993 : 1 66), and is traditionally
thought of as the workspace where skill development begins
(Anderson 1 983 ; Baddeley 1 986; Shiffrin & Schneider 1 977) and
where knowled�e is encoded into and retrieved from long-term
memory. For this rea�on meas�res of short-term memory capacity
_
have been used as mdices of aptitude for skill learning and cognitive
deve �opment in SLA (e.g., Carroll 1 993 ; Carroll & Sapon 1 95 9 ;
Harrmgton & Sawyer 1 992; Sasaki 1 993) and in a wide variety o f
other areas of educational research (e.g., Ackerman 1 988; Kyllonen
& Christal 1 990).
T�o current areas of memory research, however, imply that the
domam of the control processes in short-term memory may not be
as extensive as has traditionally been claimed. The first of these
challenges the traditional role attributed to short-term memory in
the development of automatic processing; the second challenges its
role in controlling access to long-term memory.
3.3.2 A utomaticity and memory
Automatic processes are often cited as classic examples of
the
operatio?s of implicit systems (Reber 1 993 ). The properties
of
auto�atic processes are that they are fast, effortless, obliga
. tory,
consistent, unconscious and learned. Traditional resource
based
theories of automaticity and skill acquisition, such as Ander
son's
ACT* theor?' , or Shiffrin and Schneider's theory (see McLa
ughlin
1 987; Schmidt 1 992) locate the mechanisms which are respon
sible
for the development of the learned responses drawn on in autom
atic
skilled performance in short-term memory. In Shiffrin
and
Schneider's theory ( 1 977; cf. McLaughlin et al. 1 983), contro
lled
processes activate sequential and temporary links in short-term
memory. Following practice in consistent environment
s
strengthening of corresponding links in long-term memory occurs
Eventually automatic processes, requiring little attent ional
'.
�upervisory contr� l, activate permanent links in long-term memory
m parallel each time exposure to a particular pattern of inputs
occurs.
Logan ( 1 98 8a, 1 988b, 1 990) has argued that modal resource
based theories of automaticity merely describe the pheno mena
associated with autom aticity, but do not explain 'why' practice
in
constant � nvi:onments is necessary to its development, or 'why'
. _ by a gradua
automattcity is accompamed l withdrawal of attent ion.
Memory-based theories of automaticity like those of Logan claim
Implicit and explicit information processing 67
initial performance is limited by lack of knowledge, not simply
availability of attentional resources in a capacity-limited short-term
memory, and that automaticity is the result of an increasing long­
term knowledge base which is eventually drawn on as the basis for
automatic responses. Logan's ' instance' theory rests on three main
assumptions: (i) stimuli are obligatorily encoded in memory once
attended to; (ii) each instance together with whatever aspects of the
context were attended to is stored separately; and (iii) retrieval o f
the memories i s an obligatory consequence o f attending to similar
stimuli. Together these assumptions imply a learning mechanism­
the accumulation of instance representations. Memory becomes
stronger not through strengthening (Cohen, Dunbar & McClelland
1 990; Schneider & Detweiler 1 98 8), but because "each experience
lays down a trace that may be recruited at the time of retrieval"
(Logan 1 988a:494 ). As memory accumulates so retrieval becomes
faster, until the speed of decisions about task performance made o n
the basis o f retrieval from memory exceeds the speed o f decisions
made by a general algorithm. At this point memory-based solutions
become the basis of automatic responses.
Instance theory, therefore, raises questions about previous views
of automaticity which placed the responsibility for developing
automaticity on the speed up of processes under attentional
supervision in short-term memory. These questions are raised by
three aspects of Logan's proposals: (i) the parallelism of memory­
based and algorithm-based processing; (ii) the obligatoriness o f
encoding into memory; and (iii) the obligatoriness of retrieval from
memory. Firstly, instance theory claims automatic, implicit
information processing does not follow consecutively from an
explicit learning phase of controlled processing, but develops in
parallel with it, and is to a large extent independent of it. Secondly,
the assumption of obligatory encoding of attended stimuli appears
to allow some role for memory without awareness in the
development of automaticity, since there can be detection o f
attended stimuli which operates outside o f awareness and which does
not involve noticing (see Figure 1 1 ) . However, it appears that such
memories would not contribute much to the knowledge base drawn
on in automatic responses, since the quality of attention allocated
at encoding is, for Logan, directly implicated in the speed o f
retrieval from memory: "Attention to an item will have some
impact on memory; it does not assume that all conditions o f
attention produce the same impact" ( 1 988a:494). Consequently,
noticing of instances appears to be necessary to establish a
68 Implicit and explicit information processing
knowledge base which can prompt retrieval with sufficient speed t o
outperform algorithm-based processing as the basis o f performance.
Thirdly, the obligatoriness of retrieval also implies a process that
develops in parallel with algorithm-based processing in the early
stages of skill development, but is not dependent on the attentional
control processes in short-term memory.
3.3.3 Access to long-term memory
Claims that memory without awareness may play a more significant
role in cognitive processes than has hitherto been acknowledged
have also come from studies of apparent dissociations between
perfo�mance . on recognition and recall tasks requiring direct,
conscious retneval of material presented during a study phase, and
indirect tasks which facilitate retrieval of the material presented at
study, but without apparent conscious attempts to recall (Merickle
& Reingold 1 989; Richardson-Klavehn & Bjork 1 988; Schacter
1 987, 1. 989; . schacte.r, . Chiu & Ochsner 1 993). Repetition priming
effects m lexical decision tasks, for example, have been claimed t o
provide evidence for the use o f implicit memory. I n these tasks
subjects are presented with letter strings and asked to judge if they
are words or non-words. Implicit memory is inferred from the
reduced latency, or speed of responses to previously presented words
compared to novel words. The claim that implicit and explicit
memory are dissociated is based on the observation that
manipulating one variable during the study phase will affect
p�rformance on the later implicit and explicit test phases
differently. For example, during the study phase, task instructions
may be varied to focus attention on the semantic aspects of the
studied words, or on the physical features of the words, thereby
manipulating differences in the depth of processing (Craik &
Lockhart 1 972). This manipulation is found to affect performance
on explicit tests, which are facilitated by the greater depth o f
processing required by focusing on semantics, but not to affect
performance on implicit tests. In tests using amnesics, with damage
to the hippocampus, it has been found that repetition priming
effects are preserved on implicit tests, relative to normal
populations, while performance on tasks requiring explicit recall
from previously presented word lists is impaired (Vallar & Shallice
1 990).
Such dissociations have been used by some to support the
argument that two distinct subsystems are responsible for these
effects-an explicit memory, accessed by conscious retrieval, and
Implicit and explicit information processing 69
an unconsciously accessed implicit memory (Schacter 1 98 9 ;
Schacter & Tulving 1 982; Squire & Cohen 1 984). The argument for
dissociable subsystems of memory has recently been invoked by
Paradis ( 1 994) to explain aspects of L2 development, and it is
implied, if not specified in detail in the Monitor model of Krashen
( 1 979, 1 9 8 l a, 1 982). The evidence for and against such
dissociations is reviewed in detail in Chapter 4. Here explanations of
dissociations that have been proposed and their implications for
models of the architecture of memory are briefly considered.
3.3.3.1 Activation views
rary
Activation views hold that priming effects are due to the tempo
entatio ns and are based on
activation of pre-existing semantic repres
ying the develo pment of
theories o f the processes underl
automaticity.
(i) Auto-associative semantic activation. Graf and Mandler
( 1 984) claim that implicit memory performance is due t o
temporary activation o f a representation that occurs automatically,
but which carries with it no contextual information which could
contribute to explicit remembering. Presentation of a word in a
study list activates its representation, thus making it more accessible
than non-studied words. However, this activation "is not sufficient
to determine recognition or recall operations, which are dependent
on the relationships of the event to be retrieved and the
characteristics of the context in which this event first occurred"
(Besson, Boaz, Fischler & Raney 1 992:90). Activation theories like
that of Graf and Mandler thus propose a mechanism, spreading
activation, which is variably related to the processing demands � f
tasks, which can either create the conditions for such automatic
access, or lead to more attentionally demanding processing which
impedes it (see the processing level explanations given below).
(ii) Similarity based recruitment and activation. Logan ( 1 990)
also offers an automatic activation explanation of repetition
priming effects. In lexical decision tasks judgments about wor�s or
non-words are made initially on the basis of an algonthm.
Subsequently, when the same words are presented a second or third
time, decisions are made on the basis of single-step access t o
previous encounters with the stimulus which are activated i n
memory. However, Logan is clear that instances are encoded i n
memory together with whatever contextual coordinates happen to
70 Implicit and explicit information processing
be attached to them. Consequently, at least some episodic, personal
memories could be activated by obligatory retrieval of instances.
The whole context of that which was attended is thus used as the
basis for computing the similarity of instances encoded during a
study phase to those attended during a test phase.
3.3.3.2 Processing views
Processing views argue that the nature of the match between the
processing demands encouraged by the task instructions at study and
at test contributes to differences in performance on tests.
Activation views and processing views are, to an extent,
complementary, as I will argue in section 3 .3 .4 below.
(i) Data-driven and conceptually driven processing. Jacoby
( 1 983) has argued that conceptually driven processing, such as
organizing and reconstructing, which is 'subject initiated' and guided
by knowledge and expectations from long-term memory, can be
distinguished from data-driven processing, which is initiated and
guided by data, for example the shapes of letters (see also Blaxton
1 989; Hamann 1 990; Roediger & Blaxton 1 987; Roediger et al.
1 989). Study tasks encourage a blend of both forms of processing,
but task instructions can bias the subject to performing more of one
than the other. Task instructions in the test phase are also claimed
to bias subjects to draw on one or the other form of processing;
explicit tests draw primarily on conceptually driven processing, and
implicit tests draw primarily on data-driven processing. What
distinguishes test performance is the transfer of appropriate
processing strategies initiated during the study phase of memory
tests to the transfer test. Implicit test performance reflects study­
test overlaps in data-driven processing; explicit test performance
reflects study-test overlaps in conceptually driven processing.
Different memory systems are not necessarily involved.
(ii) Integrative and elaborative memory processes. Graf and
Ryan ( 1 990) also take a processing perspective, and distinguish
integrative from elaborative processing. Integrative processing
"results from processing that bonds the features of a target into a
coordinated whole," whereas elaborative processing "associates a
target with other mental contents" ( 1 990:990). Following the same
reasoning as Schacter, they claim study task performance
encourages a blend of both types of memorization, but implicit tests
Implicit and explicit information processing 7 1
draw on integrative processes whereas explicit tests draw o n
elaborative processes.
3.3.3.3 A-fulti]Jle systeltlS views
Some researchers have attributed differences in performance o n
implicit and explicit tests to differences in the properties o f
underlying memory systems which are differentially activated by
attempts at explicit, conscious recall. The difference between this
type of explanation and the preceding explanations lies in the
preference of researchers for associating functionally differentiated
evidence of memory performance with system level
characterizations of the architecture of memory. Processing views
prefer to avoid this and explain functional differentiations at the
level of task demands. Multiple systems views often refer to
neurophysiological evidence of different patterns of brain activity,
and neural architecture, to support their claims for different
subsystems. Taken to an extreme this approach ceases to explain
data, since each time performance on memory tasks appears to be
functionally differentiated, a new subsystem of memory is proposed
to account for it (McKoon, Ratcliff & Dell 1 986; Richardson­
Klavehn & Bjork 1 992; Tulving 1 985). Nonetheless, a number o f
distinctions between subsystems of memory have attracted support
among some memory researchers.
(i) Procedural and declarative memory. Squire and Cohen
( 1 9 84) distinguish between consciously accessed declarative
memory, and unconscious procedural memory, the former
responsible for performance on explicit tests, the latter for
performance on implicit tests. This explanation is inspired by
neurophysiological evidence of the relation of brain impairments t o
memory functions. Combining studies o f human amnesia with
animal models of human amnesia, Squire and Zola-Morgan ( 1 99 1 )
have identified the medial temporal lobe as the site of memory
functions. The medial temporal lobe binds together distributed
storage sites in the hippocampus and neocortex. The role of the
medial temporal lobe in distributing information in neocortex is
only short-lived, and after time memories stored in neocortex
become independent of the medial temporal lobe. Squire and Zola­
Morgan interpret this as evidence of the neurophysiological basis o f
declarative memory for facts. The medial temporal lobe functions
as a short-term organizer of information in long-term storage in
neocortex. Lesions to the hippocampus and neocortex surrounding
72 Implicit and explicit information processing
the amygdala produce memory impairments, while lesions to the
amygdala do not, suggesting that the amygdala is responsible, in
part, for the memories that are preserved in amnesics-the ability
to make associative links between stimuli and sensory modalities
and �ther fu�ctions important to skill development and procedural
learmng (Sqmre, Amaral & Press 1 990). At present, evidence from
neurophysiology is suggestive of, though far from clear evidence
for, the claim that different aspects of memory functioning are
centralized in different sites. Impairments in performance o n
explicit tests o f recognition and recall appear to b e linked t o
impairments of .the hippocampal system; performance o n implicit
memory tests 1s not affected by such impairments. Studies o f
differences in synaptic plasticity within anatomically distinct sites
have also been identified and used to support claims for different
forms of memory (Gluck & Granger 1 993 :668).
(ii) Episodic and semantic memory. Tulving ( 1 983, 1 985, 1 986)
has developed a multiple systems theory of memory which
distinguishes between episodic memory for personal involvement
with an event, and semantic, decontextualised memory. This
distinction has been used in explanations of dissociations between
implicit and explicit memory (Schacter 1 9 89). The episodic
memory system is argued to be responsible for performance o n
tasks that require explicit remembering o f recent events, while
semantic memory is argued to be responsible for performance o n
tests that require access to pre-existing decontextualised semantic
representations. Episodic memory deals with, " unique, concrete,
personal and temporally dated events, " while semantic memory
involves "general, abstract, timeless knowledge that a person shares
with others" ( 1 986:307). Tulving' s multiple memory model is
hierarchical. Episodic memory is seen as a specialized subsystem o f
semantic memory, accessed by autonoetic, intentionally
manipulated, consciousness. Both episodic and semantic systems
deal with declarative knowledge. Declarative memory is viewed as a
separate system embedded within a procedural memory. This
distinction, then, is in some ways similar to the
declarative/procedural distinction made above. However, it differs
from the accounts of Cohen ( 1 984) and Squire and Zola-Morgan
( 1 99 1 ) by viewing declarative memory as nested within procedural
memory, rather than viewing them as separate co-existing systems
functioning in parallel. . It is also similar to the
procedural/declarative distinction made in Anderson's ACT* model
Implicit and explicit information processing 73
of knowledge compilation ( 1 983). However, Anderson's ACT*
theory does not propose that episodic and semantic memory are
separable systems within declarative memory. Rather than motivate
his procedural/declarative distinction with reference to
neurophysiological evidence alone, or with reference to a model o f
information processing during skill acquisition, Tulving, like Reber
( 1 993 ), has claimed that distinctions between memory systems are
the result of evolutionary adaptive processes. Episodic memory is a
late arrival, and phylogenetically derived from semantic memory.
Like Reber, Tulving claims that ontogeny recapitulates phylogeny,
thereby explaining why in early childhood children are able t o
perform on indirect measures o f memory, which, he claims, access
the semantic system, and only later are able to perform on direct
measures which access the episodic system.
3.3.4 Multiple systems and multiple explanations
As Schacter ( 1 987) points out, there are advantages and
disadvantages to each of these explanations, and they are not all
mutually exclusive. One major disadvantages of multiple systems
views, however, is that they are inconsistent with each other. Note,
for example, the inconsistency between the use of procedural and
declarative to describe memory systems in the work of Squire and
Zola-Morgan ( 1 99 1 ), Tulving ( 1 986) and Anderson ( 1 983),
described above. Multiple explanations underlie the multiple systems
that have been proposed. If confusion is to be avoided with regard
to the role of key variables like attention and consciousness, the use
of the same pair of terms to indicate a contrast between memory
mechanisms must be made with reference to the assumptions of the
specific explanation they are based on. It is clear, for example, that
explanations of forms of memory and their relationship to learning
differ with respect to the role of consciousness. For Anderson
( 1 983 ), skill learning begins with conscious declarative knowledge
which is later proceduralised. This is an interface position. For
Cohen ( 1 9 84) and Squire and Zola-Morgan ( 1 99 1 ), consciously
accessed memory and unconsciously accessed memory are
dissociated and distinct forms which are responsible for different
kinds of learning.
Confusions may also arise through failure to consider the
different levels at which each explanation is pitched. Claims for
distinctions between procedural and declarative memory based o n
neurophysiological evidence may be true at that level, but they are
interacting systems at the level of information processing during
74 Implicit and explicit information processing
task performance (Bunge 1 973 ; Robinson 1 993b). The question o f
whether the function o f different memory systems can be isolated
in task performance by amnesics and monkeys with lesions to the
hippocampus is, to a large extent, irrelevant to the question of how
they interact and contribute to learning during task performance by
normal L2 learners (though see Robbins 1 992, and Paradis 1 994,
who base large claims for L2 learning and pedagogy on this
evidence). Of interest to the information processing level o f
explanation is a close examination o f the attentional demands o f
tasks, and the extent to which manipulations o f these demands can
lead to differences in measures of learning and retention.
3.3. 5 A utomaticity, retrieval and context instantiation
An alternative to inferring two distinct forms of memory, then,
would be to reconcile an explanation based on automatic activation
of pre-existing representations, as in Logan's instance theory, with
an explanation based on processing distinctions between
conceptually driven and data-driven processes. In this account,
explicit instructions to remember are assumed to lead subjects t o
associate each currently attended target with each previously
experienced target, and additionally to involve comparing the
currently attended context of recall with the whole of the
previously experienced situational context of study. This requires
conceptually driven processing, and Graf and Ryan's closely related
elaborative memory strategy, rather than data-driven processing
alone. However, while a currently attended target may find a
perceptual match with a previously encoded target, the surrounding
situational context differs, and context is coded as a part of instance
representation in Logan's theory (Logan & Etherton 1 994). For
this reason, single step access to a matching instance may be slower
in situations requiring explicit recall, than in situations which do
not, like implicit tests, since the currently attended contextual
coordinates of the stimuli will not find a satisfactory match. In
implicit tests recognition is data-driven, and differences in the
study-test contexts are not invoked by task demands. Therefore
access may be more automatic, since instances find a clearer match.
This hybrid 'task-based' position would have the virtue of
accommodating two sources of implicit memory which Richardson­
Klavehn and Bjork ( 1 988) argue need to be accounted for "ahistoric
traces that depend on pre-existing codified representations, and
historic traces that incorporate contextual information"
( 1 988:53 3), without the need to invoke separate systems.
Implicit and explicit information processing 75
3.3. 6 Summary of the role of memory
In summary, this review of the nature of the relationship of
memory to attention during information processing leads to the
following conclusions. (i) There is minimal evidence of encoding
into long-term memory without awareness-what evidence there is,
reviewed in section 3 .2.6, is largely irrelevant to the debate over the
nature of implicit memory. By far the larger part of this debate, and
the issue of the role of attentional control processes in short-term
memory, concerns evidence of memory retrieval without awareness.
(ii) Short-term memory should be distinguished from long-term
memory, adopting the definition that short-term memory is that
subset of long-term memory in a currently activated state. (iii)
Retrieval from long-term memory can be a consequence o f
conceptually driven processing under attentional control in short­
term memory, as in explicit memory tasks. When this occurs the
contents of retrieval are determined by the interaction o f
conceptually driven processing and the specific nature o f task
demands. (iv) There can be automatic activation of previously
attended information encoded into long-term memory. When this
occurs the contents of retrieval are determined by the interaction of
data-driven processing and the nature of the specific task demands.
(v) This position is inconsistent with the view that separate l � ng­
term memory systems are differentially responsible for consc10us
and unconscious information access during informatic . 1 <irocessing
(Paradis 1 994; Schacter 1 989; Tulving 1 9 86). (vi) H is consistent
with the view that level of attentional awareness dunng retrieval is a
function of task demands and automatic processes, which jointly
determine access to a single long-term memory store.
3.4 Individual differences, coguitive resources, and L2 acquisition
The nature of the interaction between cognitive resources during
information processing and language learning is, as yet, little
understood. Nonetheless, a certain number of predictions follow
from Schmidt's claim that noticing is necessary to language
learning, and from the description of the relationship of attention
to memory during implicit and explicit information processing
which has been described above.
3.4.1 Individual differences and conscious learning
Firstly, short-term memory capacity should be important, since this
determines how much can be noticed, and subsequently rehearsed,
while performing a task. It should be related to differential
76 Implicit and explicit information processing
performance on single tasks, when resources in the L2 are limited
and skills are relatively non-automatic, and to a greater extent to
performance on dual-tasks which require time-sharing and attention
switching. Success in workload management, using Wickens' model
(see Figure 1 0), should be most clearly related to short-term
memory capacity when concurrently performed tasks draw on
similar resource pools. Secondly, when tasks require conceptually
driven processing, the availability of knowledge schemas to organize
perception, and direct attention to relevant aspects of the stimulus
domain to be learned, should also be important. This prediction,
then, claims that the extent to which relevant pre-existing
representations are available will determine the efficiency of
attentional allocation, which in turn will lead to more successful
task performance.
3.4.2 I,ndividual differences and unconscious acquisition
It should be noted, however, that those who claim SLA is a largely
unconscious process (Krashen 1 985), or who claim that learning
without awareness is a more robust and extensive form of learning
than that which arises as a consequence of explicit instruction to
look for rules or to apply consciously learned rules to examples of
relevant structures (Abrams & Reber 1 988; Reber 1 989, 1 993),
argue that ultimate attainment in second language, or artificial
grammar learning is likely to be unrelated to currently available
measures of cognitive abilities. These, it is argued, measure only the
abilities drawn on in explicit information processing, where
individual differences can be expected. Implicit learning, or
acquisition, which are independent of conscious efforts to learn, are
abilities which are relatively homogeneous in the population, and
unrelated to short-term memory capacity or the ability t o
consciously divide and maintain attention.
3.4.3 Short-term memory capacity and learning
The relationship of short-term memory capacity to processing in
the first and second language has been discussed by Lado ( 1 965),
Stevick ( 1 976), Cook ( 1 977, 1 979, 1 986), Papagno, Valentine and
Baddeley ( 1 99 1 ) and Harrington and Sawyer ( 1 992). Cook ( 1 977)
assessed the extent to which adult subjects differed on identical tests
of short-term and long-term memory using the first and then the
second language. He concluded that performance on short-term
memory tests was more similar in the L 1 and L2 than performance
on long-term memory tests, and that short-term memory capacity
Implicit and explicit information processing 77
was therefore more directly transferable to L2 processing than long­
term memory capacity. Cook ( 1979) extended the research to child
L2 learners, finding greater ability on digit span measures of short­
term memory than on recall of words, and greater ability in both o f
these when the digit span and word list were presented i n the L 1 .
Short-term memory for sentences was tested by presenting subjects
with L2 sentences of between four and sixteen words in length and
then asking for recall. Performance was significantly worse than
when the sentences were presented to adults in their L 1 .
Papagno et al. ( 1 99 1 ) examined the influence of a proposed
subsystem of short-term working memory, the phonological store,
on adult acquisition of L2 vocabulary. Phonological storage was
disrupted by articulatory suppression of subvocal rehearsal. They
contrasted performance on paired-associate tasks in the L I ,
English, and in the L2, using pairs of English and Italian, Russian
and Finnish words, finding that this disrupted the learning o f
Finnish-English and Italian-English paired-associates, but not
Russian-English or English-English. They speculate that the formal
similarity of word structure in English and Russian was greater than
for the other two languages and this allowed subjects to circumvent
the use of rehearsal in phonological short-term memory in learning
the associates.
Harrington and Sawyer ( 1992) examined the effect of differences
in L2 reading skill and working memory capacity, using a measure
developed by Daneman and Carpenter ( 1 980) for assessing reading
span. They found that subjects with larger measures of working
memory capacity scored higher on tests of reading skill. In contrast,
reading skill did not appear to correlate highly with more passive
measures of short-term storage such as backward digit span or
random word span.
These studies lend some support to the claim, consistent with
Schmidt's noticing hypothesis, that short-term memory capacity is
a predictor of success in performing L2 tasks. Carroll ( 1 97 1 , 1 99 3 ;
Carroll & Sapon 1 959) and Sasaki ( 1 993) have also used measures
of short-term memory to predict differences in aptitude for L2
learning, finding significant correlations between scores on memory
components of aptitude tests and subsequent foreign language
learning achievement. It should be noted, however, as Carroll has
acknowledged, that the use of memory tests in foreign language
aptitude research has not drawn much on recent findings in memory
research reviewed above. In particular no measures of implicit
memory ability are currently used in aptitude tests. Given the claims
 Implicit and explicit iriformation processing 79
78 Implicit and explicit iriformation processing

of Krashen and Reber that acquisition and implicit learning are of metalinguistic knowledge with the aim of developing such
likely to be insensitive to explicit measures of memory, it remains sensitivity is certainly possible (O'Malley & Chamot 1 990).
t� b� se�n "':hether implici� measures provide more homogeneous Nonetheless, Schmidt's noticing hypothesis predicts that efficiency
d1stnbutions m the test takmg populations than explicit measures, in directing attention to relevant areas of the structural domain t o
and whether they correlate more with language learning under be learned i s likely to lead to more effective learning, and eventual
conditions like those advocated by the Natural Approach (Krashen understanding, given the possible condition that such noticing and
& Tei:rell 1 983), which aim to develop unconscious acquisition, hypothesis testing takes place within an appropriate developmental
assummg, of course, that such a phenomenon could be empirically schedule (see e.g., Pienemann 1 989). It seems plausible to expect
de�onstrated and distinguished in a principled way from learning that such efficiency will be related to the availability o f
which proceeds by conscious noticing and hypothesis testing. metalinguistic knowledge and the ability to identify grammatical
analogies between sentences as revealed, for example in Carroll and
3.4.4 Attention, salience and grammatical sensitivity Sapon's ( 1 959) measure of grammatical sensitivity.
The ability to notice features of target language input should not
only be related to short-term memory capacity, but should also be 3. 5 Summary
affected by the existence of pre-existing knowledge representations In summary, Schmidt's claim that noticing is necessary to learning,
which can selectively guide the allocation of attentional resources. and that awareness at the level of understanding is strongly
Such representations would facilitate the act of selectively attending facilitative of subsequent learning can be related to measures o f
to �spects of the input for the purposes of conscious hypothesis cognitive resources that facilitate noticing-short-term memory
testmg about the target language, or may simply lead to effective capacity and grammatical sensitivity. The further claim is that
priming of attended features, making recall and use of such features these abilities are drawn on in explicit information processing, and
more automatic (cf. Mellow & Cumming 1 994 ). Hypothesis testing that since explicit information processing is the mode necessary for
may be a consequence of explicit instructions to search for, or t o adult L2 learning, then individual differences in these abilities will
apply rules o f the target language during form-focused tasks, or correlate with measures of L2 learning success. Possible differences
based on an individual impulse to do so during meaning-focused between adults and pre-critical period children in this respect are
activities. Noticing and hypothesis testing, that is, may be discussed in section 4.3 .4 of the following chapter.
constrained by task demands (Fotos 1 993), but are not limited by
them (Allwright 1 984; Slimani 1 992). General measures of field
independence and dependence have been used to address the issue of
individual differences in the ability to manage the deployment of
selective attention (Chapelle & Green 1 992; Chapelle & Roberts
1 986; Naiman, Frolich, Stern & Todesco 1 978), but there have
been doubts expressed about the usefulness of these measures in the
domain of L2 learning (see Skehan 1 989). More domain specific
schematic knowledge such as text schemata have been claimed to
facilitate L2 reading comprehension (Carrell 1 984; Swaffer 1 9 89),
and it seems likely that in the domain of grammatical knowledge,
familiarity with the basic metalinguistic principles for describing
stru�tural patterns and structural analogies would aid hypothesis
testmg. Carroll ( 1 97 1 , 1 993 ; Carroll & Sapon 1 959) has claimed
that grammatical sensitivity to structural patterns is a component
of aptitude for learning languages and not necessarily related to
previous foreign language study. However, explicit training in areas
 Chapter 4

Implicit Learning, Memory And

Knowledge

4. 0 Introduction
This chapter reviews the experimental evidence for a distinction
between implicit and explicit processes at three different levels of
description, the levels of learning, memory and knowledge.
Following this review, the objectives of the present study of implicit
and explicit L2 learning are described and a rationale for the
research hypotheses is presented. Research into implicit learning
and memory has made use of different methodologies, and it is
important to distinguish the two (Ellis 1 994a; Reber 1 99 3 ;
Robinson 1 993a, 1 995d). Leaming i s the process whereby new
knowledge is acquired as a result of exposure, in this case t o
examples of language, under particular task conditions. Section 4 . 1
reviews the evidence of differences in learning under conditions that
manipulate degree of awareness of the rules describing the synthetic
grammars used in experiments by Reber ( 1 989) and others.
Describing learning must of necessity invoke memory and
knowledge, since knowledge is the end result of learning, and
memory is where knowledge is stored. However, dissociations
between implicit and explicit memory, as opposed to learning, are
properly studied under conditions in which the knowledge accessed
by memory pre-exists the experimental task conditions. Of course,
this is rarely the case in studies of implicit memory, since these
employ a study-test format in which stimuli are presented at study
as the basis for later measures of accessibility (see section 4.2). T o
this extent, then, the study phase of implicit memory tests involves
learning, in the sense of encoding items into memory, and the test
phase measures the extent of their implicit activation by the test
instructions. However, this is not the sense in which implicit
learning is used by Reber who argues that abstraction of rules
regulating the presentation of letter strings takes place during
artificial grammar learning experiments, following the instruction
82
Implicit learning, memory and knowledge 83
Implicit learning, memory and knowledge

grammatical/ungrammatical significantly better than chance, but t o

to simply memorize, and that this learned, item-independent
be unable to articulate the rules o f the grammar that produced the
knowledge is generalizable to novel letter strings, forming the basis
of grammaticality judgments made in the later transfer tests. acceptable strings. Based on this evidence Reber makes claims about
Knowledge is thus implicated in any study of learning, or the product and process of implicit learning: (i) the product is an
mem? ry, though t�e:� is a temporal dimension in the study o f abstract knowledge of the rules of the grammar; and (ii) the process
learnmg (the acqms1t10n o f new, or the modification o f old whereby this knowledge is accumulated is inaccessible to conscious
knowledge) that is absent in the study of memory (access of existin � awareness. Reber has also claimed that implicit learning is superior
to explicit learning. This claim is based on comparisons of the
knowledge). The representational issue addressed in section 4.3 thus
takes its bearings from the study of implicit learning (how does a transfer phase performance of subjects trained under the
knowledge base accumulate over time?), though it clearly must also memorization condition with that of subjects following the
account for evidence arising from the study of implicit memory instruction to search for rules. Subjects trained under the
(what knowledge can be accessed, and under what conditions?). The memorization condition perform better than those trained under
the rule-search condition.
central representational issue in the study of implicit learning
concerns the abstractness of the accumulated knowledge base. Is it
abstract, and independent of memory for specific instances
en�ountered duri�g learning, as Reber ( 1 989, 1 993) has consistently
_ s
claimed, or 1s 1t the result of access to specific examples or
fragments of examples presented during learning? Section 4 . 3 .4
summarizes the review of implicit learning, memory and knowledge,
s
proposing that implicit and explicit processes at each of these levels
T
are 'fundamentally similar,' while section 4.4 describes the
objectives and research hypotheses for the present study.

4.1 Implicit and explicit learning

Reber's research ( 1 969, 1 976, 1 9 89, 1 993) is based on two
meth? dologies for studying implicit learning, synthetic grammar v
learn1�g and probability learning. The synthetic grammar learning
expenments consist of an acquisition, or training phase, and a
T
testing phase. In the training phase subjects are presented with
strings of letters generated by an artificial, Markovian, left-to-right
finite-state grammar. A typical example of the grammar used is
sho�n in Figure 1 2. In the testing, or transfer phase of experiments, (from Reber 1 989:222)
subjects are presented with new examples of strings generated by the
rules of the grammar and asked to judge which of them are well Figure 12. The Finite-State ' Reber' Grammar
formed. Subjects have been found to perform differently as a
function of the different conditions under which training takes
place. Specific details of these training conditions, and the evidence In the probability learning experiments subjects are exposed to
of differential success in learning are reviewed below. rapidly presented sequences of events for three minutes, at a rate of
Reber claims that these experiments produce evidence of implicit two per second. In the transfer phase they are asked to make
learning, which is the result of following instructions to memorize predictions about the probability of a future event. In Reber and
the strings presented during training. Subjects trained under the Millward ( 1 97 1 ) the probability of an event occurring in a sequence
memorization condition have been found to classify the strings into
84 Implicit learning, memory and knowledge
was adjusted to steadily increase over fifty presentations, then
systematically decrease over the next fifty, then increase again over
the following fifty, etc. Reber's claim, based on the subjects'
patterns of correct responses, is that they had learned the deep
organizing principle, or structure, of increasing and decreasing
probabilities over periods of fifty trials, and not merely the specific
probabilities of occurrence of individual events. Other researchers
have made use of matrix scanning methodologies for studying
implicit learning, in which the occurrence of sequences of numbers
(Lewicki, Hill & Bizot 1 988) or lights (Nissen & Bullemer 1 987) in
quadrants of a computer screen have been the target of learning.
However, Reber has commented that the nature of the repeating
sequences used derives from rules which are less 'abstract' than
those induced during synthetic grammar learning experiments
( 1 993 :44). Consequently, in the following sections my discussion of
the experimental evidence for implicit learning is restricted to those
studies which have employed synthetic grammars, since, of the
methodologies briefly surveyed here, they involve the use of a
stimulus domain most representative of the level of complexity o f
the target o f learning during SLA.
4.1.1 Evidence of dissociations using artificial grammars
There are several reasons for caution in assuming that claims about
implicit learning using artificial grammars could be generalizable t o
SLA. (i) First, there is reason to doubt that learning artificial
grammars and learning natural languages involve comparable
processes given the difference in complexity of the two stimulus
domains. (ii) Secondly, there is the issue of comparability across the
studies that have so far been done since there is a lack of
consistency in many studies regarding the operationalization of
instructional variables. This inconsistency gives cause to doubt that
a single homogeneous process could properly be inferred from the
different instructional treatments that are consequently delivered.
(iii) Thirdly, there is the issue of the slender effect sizes used t o
support claims for the occurrence of implicit learning. Though
demonstrating that performance above the level of chance is made
possible by learning under implicit conditions, the studies so far
have failed to show a robust level of success. (iv) Finally, there is
the issue of the consistency of the findings themselves, since there
have been a number of failures to replicate results obtained on early
implicit learning studies.
Implicit learning, memory and knowledge 85
(i) The complexity of artificial grammars . Reber has argued that
the complexity of the synthetic grammar used to generate strings
viewed during training is a necessary prerequisite for the observation
or occurrence of implicit learning, since if the rules are simple
enough to be analyzed by conscious effort, then subjects in the
implicit learning condition will become aware of, and begin t o
search for and find, the rules of the grammar. Figure 1 2 represents
the grammar used in Reber ( 1 967, 1 969 and 1976), and in studies by
others (e.g., Servan-Schreiber & Anderson 1 990). Variants of this
grammar have also been used in experiments by Reber (Reber &
Allen 1 978) and others (e.g., Knowlton & Squire 1 994; Turner &
Fischler 1 993 ; Vokey & Brooks 1 992; Whittlesea & Dorken 1 99 3 ),
with the aim of varying the complexity of permissible strings, or o f
seeing whether subjects can distinguish between strings generated by
different grammars (Whittlesea & Dorken 1 993 ). Nonetheless,
although they are complex, it is clear that such grammars are
considerably less complex than those of natural languages, and differ
in their fundamental properties by lacking, for example, a semantic
and a phonological component, and many of the structure­
dependent characteristics of natural languages which allow syntactic
relations to occur over discontiguous elements (Baker 1 989; Braine
1 965; Chomsky 1 959, 1 965; Mccawley 1 988; Pinker 1 994).
(ii) Lack of consistency across studies. It is clear, also, that few
features of implicit learning studies, apart from the use of finite­
state grammars to generate test strings, are standardized from study
to study. Sometimes differences in treatments are motivated by
manipulations of variables of interest, such as the degree o f
similarity or complexity o f test strings (McAndrews & Moscovitch
1 985; Vokey & Brooks 1 992), or by the limitations of the
experimental setting and the nature of the subjects (for example the
amnesics studied by Knowlton & Squire 1 994). However, at other
times, key moderator variables which could be expected to affect
the nature and extent of learning, like length of stimulus
presentation time, form of presentation and the nature o f
instructions t o subjects, differ across studies for n o apparent reason.
The variation of length of stimulus string, presentation time,
number of runs or repetitions of the string, and manner o f
presentation during training across nine implicit learning studies is
presented in Table 3 , together with descriptions of instructions t o
subjects in the implicit condition of each study.
86 Implicit learning, memory and knowledge Implicit learning, memory and knowledge 87

Table 3. Instructional Variables In Nine Implicit Learning (iii) Effect size. The effect size of correct classification of novel
Stud ies grammatical sentences as a function of training under implicit
conditions is typically small, though statistically distinguishable
Study Length of Presentation Repetitions Format Instructions from performance under rule-search and instructed conditions in a
Stimulus Time to Subjects number of studies. Comparisons of results across conditions are
summarized below in sections 4. 1 . 1-3 . Here it is sufficient to
Reber 4-8 letters subject-paced 1 run, printed memorize illustrate the magnitude of the effect size used to support claims for
( 1967) (20 strings) 4 strings card for a test the occurrence of implicit learning.
on 5 cards

Reber 3-6 letters 1 0 seconds 3 runs, printed pay the

Table 4. Effect Sizes For Performance By Ten Implicit
& Allen (20 strings) per card 1 string card utmost
attention
Learning Groups
( 1976) per card

Matthews 5-1 0 letters subject-paced 1 run, printed memorize Study Effect Size
et al. (20 strings) 20 strings list for a test
( 1989) on one list Reber ( 1 967) .69
Reber ( 1 976) .66
McAndrews 3-8 letters about 24 3 runs, printed choose the
Reber & Allen ( 1 978) .73
& (24 strings) minutes 8 pairs card best name
per card to market
Reber et al. ( 1 980) .5 1
Moscovitch
( 1985) a computer Knowlton & Squire ( 1 994) normal = .59 amnesic = .55
Whittlesea & Dorken ( 1 993) .64
Vokey & 3-7 letters subject-paced 1 run, printed study then Turner & Fischler ( 1 993) . 62
Brooks ( 1 6 strings) 4 strings card recall and Servan-Schreiber & Anderson ( 1 990) . 6 8
( 1 992) on 4 cards write down McAndrews & Moscovitch ( 1 985) .53

Knowlton 3-7 letters 3 seconds 1 run, printed read then

& Squire (23 strings) per card 1 card card recall and
( 1994) per string write down
(3 tries)
Table 4 illustrates effect sizes for implicit learning of
unstructured stimulus displays in Reber ( 1 967, 1 976), Reber and
Turner & 4-8 letters subject-paced 1 run, computer memonze Allen ( 1 978), and Reber, Kassin, Cantor and Lewis ( 1 980). These
Fischler (2 1 strings) 3 strings screen as much results are based on Reber ( 1 989), in which Reber summarizes his
( 1993) in 7 sets as you can own research findings, using percentage scores for correct
classification of examples of strings occurring twice during transfer
Whittlesea 7 letters subject-paced 2 runs computer pronounce grammaticality judgment tests. Also included in Table 4 are results
& Dorken (80 strings) for each screen on the first from Servan-Schreiber and Anderson ( 1 990) (a weighted average o f
(1 993) string run then spell percentage correct to unstructured stimuli in the transfer
on the next grammaticality judgment session); Turner and Fischler ( 1 994)
(percentage correct across all items in a transfer grammaticality
Servan- 6-8 letters 5 seconds 1 run, computer read then
4 strings screen recall and judgment session); Whittlesea and Dorken ( 1 993) (the probability
Schreiber (20 strings)
& Anderson in 5 sets write out of judging grammatical items correctly in the transfer session);
( 1990) Knowlton and Squire ( 1 994) (percentage correct across all items o n
88 Implicit learning, memory and knowledge
transfer sessions by normal subjects and amnesics ); and McAndrews
and Moscovitch ( 1 985) (percentage correct across all items in a
transfer grammaticality judgment test).
These data, then, while obscuring the effects of manipulating
variables of interest within individual studies, like degree o f
similarity o f items (McAndrews & Moscovitch 1 985 ), or the effect
of response bias (Whittlesea & Dorken 1 993), do give an
approximation of the effect sizes observed across studies for
performance on transfer grammaticality judgment tests following
training under implicit conditions, where performance at 50 per
cent accuracy would represent chance.
iv) Replicability. Finally it should be noted that there have been a
number of failures to replicate the finding that implicit learning
occurs and is superior to explicit attempts to search for rules using
artificial grammars, e.g., Dulany et al. ( 1 984) and Abrams ( 1 987),
as well as a recent failure to replicate findings regarding the
effectiveness of implicit learning following Hayes and Broadbent's
methodology ( 1 988) by Green and Shanks ( 1 993).
4.1.1. 1 Implicit task conditions
Given these caveats Reber ( 1 989, 1 993) has noted that
performance under implicit conditions in artificial grammar learning
typically exceeds performance under rule-search conditions.
Implicit conditions are operationalized following the instruction t o
remember, or pay the utmost attention to the stimuli presented,
without informing subjects of their rule-governed nature. Reber's
studies, it should be noted, have no measure of whether this
condition is delivered successfully, whereas other studies (e.g.,
Servan-Schrieber & Anderson 1 990) have added a recall test with
feedback to the implicit condition and imposed a threshold o f
correct recall o n responses before allowing subjects to proceed t o
viewing the next stimuli.
4.1.1.2 Rule-search conditions
In rule-search conditions subjects are told that rules underlie the
stimuli presented and are asked to search for them during training.
The typical low performance of this group is due, Reber claims, t o
the fact that the stimulus domain i s complex. Where the stimulus
domain is simple, and the structural correlates of rules are made
salient, then Reber claims this condition can produce effective
learning-"Looking for rules will work if you can find them"
Implicit learning, memory and knowledge 89
( 1 993 :49). This is possible in simple stimulus domains, but
otherwise will result in the formation of correlated grammars, or
partial and idiosyncratic rule systems that are not isomorphic with
the underlying structure of the grammar.
4.1. 1.3 Effects ofproviding explicit instruction
The effects of providing explicit instruction in the rules have been
examined by Reber et al. ( 1 980). In this study instruction took the
form of showing subjects the schematic grammar illustrated in
Figure 1 2, and giving them a seven-minute course on how it could be
used to generate letter strings. Their basic finding was that providing
explicit instruction, and structuring the presentation of stimuli t o
conform to this instruction, will produce effective learning i f the
instruction is provided early in the training process. If the stimuli
are unstructured, or if the instruction occurs late, then this
condition will also be relatively ineffective compared to the implicit
condition. It should be noted that a similar caveat applies here
regarding the complexity of the rule taught, as applies t o
instructions to search fo r rules: i f the rule is too complex, Reber
claims, then this condition will also be ineffective relative to
implicit learning, regardless of the point at which instruction is
provided. In other words, instruction helps if the stimulus domain is
simple, and structured to conform to the rule taught, and if training
is provided early. However, Reber et al. 's ( 1 980) results show that
even when these conditions are met, instructed subjects perform
little better than implicit subjects (. 7 1 for percentage correct o n
transfer items following explicit instruction, compared to .69 for
implicit learners in the same study). By implication, following
Reber's claims, one might expect instructed learners to perform at
about the same level as implicit learners where the stimulus domain
is at the level of complexity of artificial grammars, but to fare
much worse when faced with a considerably more complex stimulus
domain such as natural language.
4.1.2 Arguments against dissociations
Arguments have been put forward disputing Reber's claims, a) t o
have demonstrated that implicit learning o f abstract rules underlying
the stimulus display takes place, and b) that the accumulation o f
knowledge during implicit learning takes place in the absence of
awareness, and that, to this extent, such learning is unconscious and
dissociable from conscious explicit learning. This dispute, then,
concerns both the product and the process of implicit learning. The
90 Implicit learning, memory and knowledge
counterarguments to Reber all involve the claim that memory is the
basis for later performance, not abstract rule knowledge, and that
such memories are accessible to awareness given appropriate and
sensitive measures.
4.1.2.1 Fragmented memory or abstract knowledge?
There are three versions of the memory-based counterarguments t o
Reber's claim about the abstract product o f implicit learning. Vokey
and Brooks ( 1 992) claim performance in the transfer
grammaticality judgment test is based on similarity of instances
encountered during training to those encountered at test, and report
findings that degree of similarity of training and transfer instances
affected transfer test judgment more than rule conformity (see also
McAndrews & Moscovitch 1 985; Turner & Fischler 1 993). Vokey
and Brooks suggest that dual knowledge, rule and instance, models of
representation (e.g., Logan 1 988a) as well as distributed memory
models (e.g., Hintzman 1 986; Rumelhart & McClelland 1 9 8 5 ),
could explain their findings (see section 4.3). Perruchet and Pacteau
( 1 990) argue that whole string similarity can be further analyzed
into substring knowledge of memory for bigrams, or letter pairings.
They found that transfer test strings that preserved frequently
presented bigrams tended to be accepted as grammatical, even when
these strings violated the rules of the grammar. Servan-Schrieber
and Anderson ( 1 990) found that inducing chunking by presenting
visually spaced groupings of bigrams and trigrams in the training
session affected transfer test performance, since ungrammatical
strings that contained chunks tended to be wrongly judged
grammatical whereas, of the ungrammatical sentences, those that
violated chunks were rejected most often.
None of these findings, however, can explain the apparent
stimulus independent nature of the knowledge gained during implicit
learning illustrated in Reber ( 1 969) and Matthews et al. ( 1 989).
Reber ( 1 969) found that changing the letters used to produce
stimulus strings presented in a transfer phase had no effect o n
accuracy, whereas changing the structure o f the grammar that
generated the strings caused a significant drop in accuracy.
Matthews et al. ( 1 989) also found that changing the stimuli used
throughout the course of a four-week implicit learning experiment
had no adverse effect on performance by one group, relative t o
another group who used the same letters throughout the
experiment. However, Matthews et al. conclude that implicit
learning is not a homogeneous process of rule abstraction, but is also
Implicit learning, memory and knowledge 91
the product of memory-based strategies, citing as evidence analyses
of verbal reports given by implicitly trained subjects who were
subsequently asked to search for rules. The verbal report data
showed them to be drawing on a knowledge base that was both item
specific and rule-based. Importantly, too, for the implicitly trained
group, such rules were accessible to consciousness, and were
successfully communicated to a yoked group of subjects using an
unknown letter set. This was evidenced by the yoked group's better
than chance performance on string discrimination tasks requiring
them to identify which of five strings was grammatical. However,
the yoked groups never did as well as their implicitly trained
informants, suggesting that the item specific contribution to their
knowledge base had been affected by the letter change.
4.1.2.2 Verbalizability of knowledge and awareness
The claim that memory-based mechanisms are, at least in part,
responsible for the knowledge base that develops during implicit
learning is linked to the claim that measures of awareness used t o
assess the extent o f subjects' knowledge o f the rules are inadequate
since they require subjects to verbalize knowledge that is disj oint
with the knowledge they gain by consciously attending to the
stimulus domain (Perruchet & Pacteau 1 990; Shanks & St. John
1 994). Shanks and St. John ( 1 994) have argued that the
retrospective protocols typically used by Reber to measure the
extent of subjects' awareness of the abstract rules of the grammar
thus fail to meet the "information criterion" of measuring the
precise nature of information they have gained by consciously
attending. Like Perruchet and Pacteau ( 1 990) and Brody ( 1 9 89)
Shanks and St. John also argue that the use of retrospectiv_e
protocols is relatively insensitive as a measure of awareness and
should be accompanied by measures that meet the "sensitivity
criterion" like concurrent think-aloud protocols and recognition
tests, rather than explicit recall alone. Similar claims are made by
Perruchet and Amorim ( 1 992) and Perruchet and Gallego ( 1 993)
about the sensitivity of measures of awareness used in sequence
learning experiments (Nissen & Bullemer 1 987), which claim t o
produce evidence o f implicit rule learning in the absence of
awareness.
4. 1.2.3 Processing accounts of the evidence
Claims for the memory-based nature of representations, and the
consequent insensitivity of measures of 'rule' awareness, are also
92 Implicit learning, memory and knowledge
invoked by processing accounts of implicit learning experiments. As
with the implicit memory experiments reviewed in section 3 . 3 ,
processing explanations explain dissociations in terms o f the
transfer of processing strategies adopted during encoding to test
performance. To this extent, implicit learning conditions result in a
distinct kind of processing, not necessarily a dissociable form o f
unconscious rule learning, which is a consequence o f the instruction
to memorize. Nonetheless, such processing is a conscious operation,
involving noticing and maintenance rehearsal in short-term
memory, prior to long-term storage. Subsequent long-term memory
for associations between items presented during training is a
consequence of this type of rehearsal, and retrieval is facilitated if
study-test processing operations are matched. Whittlesea and
Darken ( 1 993 ) have recently pointed to the importance of
encoding specificity, and of matching the nature of the processing
encouraged at study and at test. In their study subjects were found to
improve in grammaticality judgments when the study-test
processing strategies were the same. Subjects performed better o n
items which were encoded and tested in the same modality, either by
spelling or by pronouncing the letter strings.
In summary, this review of studies of implicit learning leads to a
similar conclusion to that drawn by Shanks and St. John ( 1 994). (i)
There is evidence that performance under implicit task conditions,
following the instruction to memorize or simply attend to stimuli,
can result in item-specific knowledge, as well as the induction of an
abstract, item-independent representation of the stimulus domain.
(ii) It is likely that both forms of knowledge are drawn on in
transfer grammaticality judgment tests. (iii) Both forms of
knowledge are the result of fully attended explicit information
processing. (iv) Subjects are conscious of processing strategies
regulating the development of both knowledge bases and can
demonstrate awareness given sufficiently sensitive measures of that
awareness. (v) The studies reviewed above do not support the claim
that two dissociable forms of learning, differentially susceptible to
awareness, result from exposure under implicit, versus rule-search
and instructed conditions
4.2 Implicit and explicit memory
A prerequisite for studies of implicit learning is exposure to a
stimulus domain which is unknown prior to the experimental
situation. Additionally Reber has proposed that the stimulus domain
must be complex to create the conditions for implicit, as opposed
Implicit learning, memory and knowledge 93
to explicit learning. In implicit memory studies, in contrast, the
stimulus domain does not contain new knowledge, but rather is
limited to the encoding, and subsequent retrieval of previously
known information, usually in the form of words or pictures. The
dissociation between explicit and implicit memory retrieval
processes has been studied experimentally by contrasting
performance on tasks that do not require conscious recollection o f
experiences at the time o f encoding, with performance o n tasks
that do require conscious reflection on experiences at the time o f
encoding (Merickle & Reingold 1 989; Richardson-Klavehn & Bjork
1 988; Roediger et al. 1 989; Schacter 1 987, 1 989; Schacter et al.
1 993). The distinction between implicit and explicit memory,
deriving as it does from differences in performance on such tasks,
thus relates, as Schacter points out ( 1 987:50 1 ), to the subject's
psychological experience at the time of retrieval from memory; it
does not necessarily imply the existence of two separate memory
stores. It is also true that conscious awareness of, or reflection on,
experiences at the time of encoding may not simply be a
consequence of task demands, under voluntary control, but m ay
sometimes occur involuntarily.
Test type Form of memory
Explicit Explicit Implicit
intentional involuntary
Direct 1 2 3
Indirect 4 5 6
Figure 13. Measures Of Memory And Forms Of Memory
For these reasons Richardson-Klavehn and Bjork ( 1988) propose
the model illustrated in Figure 1 3 for distinguishing between
measures of memory and forms of memory. Direct and indirect
tests relate to the presence of task instructions to consciously
remember a prior episode (direct tests) or to indirectly facilitate it
(indirect tests). Explicit and implicit memory refer to those mental
states that involve conscious access of prior events, and
nonconscious activation of memory for prior events, respectively.
Explicit remembering can be voluntary, or involuntary. Studies of
94 Implicit learning, memory and knowledge
dissociations between implicit and explicit forms of memory have
been based on a comparison of cells 1 and 6 in Figure 1 3 .
4.2.1 Methodologies for studying dissociations
Those who argue for the existence of two distinct capacities or
systems (Cohen 1 984; Squire 1 987, 1 992; Squire & Zola-Morgan
1 99 1 ; Tulving & Schacter 1 990) claim that direct tests, such as
recognition and recall measures, draw on explicit memory; indirect
tests, such as word completion and preference rating, are presumed
to access implicit memory. These measures are described below.
4.2.1.1 Recognition and recall
In recall tasks the subject is told to recall something from memory
that formed part of an earlier target event, for example words that
appeared on a previously presented list. Recall can be uncued or
cued. In cued recall subjects are presented with part of the
previously attended event to aid conscious recall (e.g., the first
letter of a word). Such cues can also be items that are semantically,
phonemically or graphemically related to, but not a part of the
earlier event (e.g., a semantically associated word-a synonym or
superordinate term, see Nelson & McEvoy 1 979). In recognition
tests the subject is asked to identify which of a series of presented
words or objects were previously presented and which are novel. In
both recall and recognition tests the need to draw on memory to
complete the task is thus made explicit in the task instructions.
4.2. 1.2 Subliminal exposure and preference rating
In indirect tests, which are presumed to access implicit memory, the
subject is asked to perform a task using materials only some of
which have been presented before. If the subject shows greater
facility using the previously presented materials than the novel
materials, then this is inferred to be an effect due to the use of
implicit memory. Subliminal exposure, for example, can be used t o
demonstrate the dissociation o f implicit and explicit memory.
Kunst-Wilson and Zajonc ( 1 980) exposed subjects to geometric
shapes for periods too brief, they claimed, to permit consci�us
perception. Explicit memory, as accessed by a forced-ch01ce
recognition test in which subjects had to indicate which of a new set
of shapes they had already seen, was at chance. However, when
asked to indicate a preference for one of a series of paired shapes
(consisting of one previously presented and one new shape), the
subjects showed a significant preference for the previously presented
Implicit learning, memory and knowledge 95
shapes. The same dissociation was observed by Lewicki ( 1 985) with
respect to explicit memory for subliminally presented noun
adjective pairs, which was at chance, and preference for the
subliminally presented adjectives in sentence completion tasks.
4.2. 1.3 Priming
There are a number of different types of test of the relationship
between priming and memory. In lexical decision tasks a list o f
words is presented, with n o instructions other than to attend.
Subsequently subjects are shown another list containing some words
from the old list plus some new words. In some cases the
instructions accompanying the presentation of the second list
require the subject to identify ' acceptable' words of English.
Implicit memory is then inferred from the decreased latency in the
making of a decision about the legality or acceptability of a
previously presented word, when compared to words not previously
presented or 'primed. ' Priming can also be contrived prior to
decisions about 'preferences' for words on the second list, rather
than acceptability judgments.
4.2.1.4 Word completion
In direct measures of explicit memory, like the word identification
tests of recall and recognition cited above, the metric for assessing
the extent of access to memory is the accuracy of responses when
subjects are asked to identify previously presented words. In
contrast, indirect measures of implicit memory like word
completion tests require subjects to complete a partially presented
letter string with the most appropriate word. The partial
representation can be a word fragment, in which letters have been
systematically deleted from a word, or provision of the word stem.
Memory for previously presented words is inferred from the
frequency with which they are used to complete the letter string.
4.2.2 Evidence for dissociations
Four types of evidence are used to support the claim for
dissociations in normal populations. (i) Explicit memory, using
recognition or recall measures, is sensitive to the type of processing
required during study. Jacoby and Dallas ( 1 9 8 1 ) required subjects to
answer questions about the meaning of words presented at study
(elaborative processing) or identify whether they contained
particular letters (non-elaborative processing). They found
subsequent recognition and recall was better for the group who
96 Implicit learning, memory and knowledge
received the study treatment requmng elaborative processing.
However, performance on implicit tests was insensitive to the
different study treatments. (ii) Indirect tests of implicit memory are
sensitive to shifts in the modality of study and test, whereas direct
tests of explicit memory are not. Jacoby and Dallas ( 1 9 8 1 ) found
that auditory study followed by visual test conditions impaired
implicit memory performance but had little or no effect on explicit
recall and recognition. (iii) Priming effects on word fragment
completion and word stem completion tests have also been found t o
lead to retention over much longer periods o f time than recognition
memory for explicitly remembered material (Tulving, Schacter &
Stark 1 982). (iv) Performance on implicit and explicit tests has
been found to be statistically independent, implying that different
abilities are drawn on in these tests (Schacter 1 987: 508).
Other evidence for dissociations comes from the performance o f
amnesics who are impaired in performance on explicit tests o f
recognition and recall, but are not impaired i n performance o n
implicit tests relative to a normal population (e.g., Squire 1 9 8 7 ;
Squire & McKee 1 993). These findings are relatively
uncontroversial; however, there is disagreement about the extent t o
which a dissociation can be inferred o n the basis o f contrasting
young and elderly adult performance on implicit memory tests.
Light and Singh ( 1 987) and Graf ( 1 990) claim elderly adults
perform poorly on explicit memory tests relative to young adults,
showing a 5 0% decline in performance, compared to only a 4%
decline in performance on implicit tests. Russo and Parkin ( 1 993 )
claim steeper declines in implicit memory performance are
apparent when memory is tested for visual, rather than verbal,
material, and further that elderly subjects' reduced implicit memory
was caused by the decline in explicit memory, which was therefore
unavailable to facilitate implicit retrieval, suggesting age related
differences in the interaction of processing modes rather than
differences in the relative capacities of separate systems.
4.2.3 Explanations, dissociations and task demands
Explanations for dissociations can be divided into those which
propose that different memory systems underlie performance o n
direct and indirect tests, and that these two systems are
differentially regulated by conscious awareness, and explanations
that attribute differences in performance on the tests to differences
in the task demands of study and test formats, which encourage
different types of processing (3 .3 .3). Both types of explanation
Implicit learning, memory and knowledge 97
require that the precise nature of memory task demands be
identified so that they can be manipulated with a view to examining
the processing consequences for performance on implicit and
explicit tests.
Like Richardson-Klavehn and Bjork ( 1 988), Merickle and
Reingold ( 1 99 1 ) use the terms 'direct' and ' indirect' measures t o
distinguish the types o f experimental tasks described above, which
are presumed to access implicit and explicit memory. Since direct
measures (like the recognition task in the Kunst-Wilson & Zaj onc
1 980 experiment) and indirect measures (like the preference rating
task) of memory are quite dissimilar Merickle and Reingold n ote
that it is possible the observed dissociations may simply reflect the
different task conditions used. This means that attempts to make
claims about the dissociability of explicit and implicit memory need
to control for as many of the following as possible across the two
types of measures.
(i) Retrieval or test cues. There is a need to control for the use o f
particular cues across recognition and recall experiments, and
preference and word completion measures. Nelson, Canas, Bajo and
Keelean ( 1 987) have shown that in cued recall of word lists, using
word stems, response accuracy can vary with the lexical set size o f
the cue (cues with uncommon o r small set sizes like 'ood' are m ore
likely to cue recall than cues with larger set sizes like ' ine' . Recall
has also been shown to vary depending on the semantic set size o f
the target. Words from a large semantic set like fruit, e.g., apple,
are less likely to be recalled accurately than words from a smaller
semantic set (Nelson 1 987; Nelson et al. 1 987; Snodgrass 1 98 9 ) .
Finally, recall has been shown to vary depending on the nature o f
the relation o f the cue to the target. 'Intralist' cues, where the cue
forms a physical part of the initially presented target list, result in
more successful recall than 'extralist' cues, where the cue is
semantically, phonologically or graphemically related to, but not a
physical part of the initially presented list (Nelson 1 987; Nelson &
McEvoy 1 979).
(ii) Response bias. Response bias can be assessed on direct
measures like recognition tasks, for example, by comparing
recognition performance requiring responses in different modalities,
whereas this cannot be assessed on indirect measures like word
completion tests. As a result it is not possible to tell if dissociations
are a result of a difference in response bias on the two tasks.
98 Implicit learning, memory and knowledge
(iii) Response metric. The response metrics on word completion,
recognition and recall tests are different. Merikle and Reingold
suggest that consequently "dissociations between direct and indirect
memory tasks may simply reflect differences in measurement scales
and not.. .underlying processes" ( 1 99 1 :225). Merickle and Reingold
suggest, therefore, that task-specific differences relating to the
three dimensions described above must first be controlled for before
making claims about differences in underlying processes based o n
task performance. However, this was clearly not done i n earlier
implicit memory research which simply associated different stu�y
test formats with different systems (Schacter 1 987). More vahd
procedures for theorizing memory performance would be to identify
"format-specific memory effects," as in Graf and Ryan ( 1 990), and
then propose processing explanations based on t�em, or to com�are
performance on tasks which controlled for the differences descn?ed
above while differing only in task instructions, and then base claims
'
about implicit and explicit memory on the observed differences (if
any) in performance (for a related discussion see Robinson 1 995c).
4.3 Implicit knowledge
Knowledge is what is learned over time, following exposure t o
stimuli presented under the different task conditions described i n
section 4. 1 , and what i s activated by the direct and indirect memory
tests described in 4.2. Current discussions of the representation of
implicit knowledge have tended to side either with the claim that it
is rule-based, and abstract, or with the claim that it is the result of
access to memory for instances, or fragments of instai:ices
encountered during training. There is evidence that both views
capture important aspects of the knowledge drawn on in trans�er
task performance during implicit learning studies (and also durmg
SLA, see e.g., Ellis 1 993a; Schmidt 1 992 for discussion). On the one
hand some process of abstraction seems necessary to account for
findings by Reber ( 1 967), McAndrews and Moscovitch ( 1 985), and
Matthews et al. ( 1989), that transfer of knowledge took place from
the grammar and exemplars experienced during the training phase
of their experiments, to new letter strings in transfer test
performance, a finding which poses pro?lems for an a�count based
purely on the development of item-specific representat10ns . . C?n � he
other hand, significant effects have been found for the facihtat10n
of transfer task performance on items which preserved the
similarity of whole instances (Vokey & Brooks 1 992), or preser�ed
bigrams and trigrams (Perruchet & Pacteau 1 990; Servan-Schreiber
Implicit learning, memory and knowledge 99
& Anderson 1 990) in the letter strings requmng well-formedness
judgments. Consequently Matthews et al. ( 1 989) and McAndrews
and Moscovitz ( 1 985) argue both rule-based and exemplar-based
representations are implied by transfer task performance, while
Shanks and St. John ( 1 994) argue that implicit learning is no
different from any other form of general learning, or from learning
under the rule-search and instructed conditions, since these also
result from a blend of these two forms of classification. Reber has
also acknowledged that memorization plays a large part in implicit
learning, but his insistence on the abstractness of the resulting
representation is no doubt motivated by the need to give substance
to his larger theoretical position, that implicit learning is "primary"
in the sense of ontogenetically and phylogenetically earlier evolved,
and consequently deeper and more robust, than later evolved
conscious forms of learning (Abrams & Reber 1 988). In the sections
below rule-based and exemplar-based positions on the representation
of implicit knowledge are described and then related to seemingly
complementary models of categorization and rule-based reasoning.
4.3.1 The rule-based abstractionist position
Reber defines implicit learning as, "an unconscious process . . .
yielding abstract knowledge. Implicit knowledge results from the
induction of an abstract representation of the structure that the
stimulus environment displays, and this knowledge is acquired in the
absence of conscious, reflective strategies to learn" ( 1 989:2 1 9 ) .
The unconscious procedural knowledge resulting from implicit
learning, Reber claims, both anticipates, and is always in excess of,
conscious declarative knowledge of the stimulus domain represented
by the finite-state grammar. Reber also maintains a clear distinction
between implicit learning and memory ( 1 989: 2 1 9). The former is a
process of induction which results in an abstract representation of
the rules regulating allowable concatenations of elements in the
stimulus domain. Access to this knowledge base in making
grammaticality judgments is independent of access to memory for
specific items or trained exemplars. His position is thus
characterized by Vokey and Brooks ( 1 992) as ' abstractionist.'
4.3.1.1 Prototypes and abstraction
Abstraction of typical characteristics of category exemplars by a
process of feature counting or averaging is basic to prototype and
schema theory (Malt 1 989; Medin & Smith 1 984; Rosch 1 973). I n
this sense a schematic representation of the grammar could be
1 00 Implicit learning, memory and knowledge
derived by counting each legitimate bigram pairing as a feature, then
summing their distributional preferences to provide an abstract
representation of the mutual constraints on their positional
relations. This representation of grammatical dependency and
constituency relations (Robinson 1 986, 1 990), would be stored in
semantic memory, independently of knowledge of individual
experiences of bigram pairings, which would be stored in episodic
memory (Tulving 1 9 83). However, the assumptions of prototype
theory-that categorization is based on abstraction of defining
features from examples of a stimulus encountered during a period o f
exposure, and that this requires 'analytic' thinking skills (Malt
1989)-have been disputed by exemplar-based theories of
classification and recognition (e.g., Hintzman 1 986; Nosofsky
1 992). These are more closely related to the memory-based
representational position which Reber rejects.
4.3.1.2 Connectionist models of categorization
An alternative interpretation of the abstractionist claim could be
based, as Reber has himself suggested, on connectionist models of
categorization (see for discussion Adams, Aizawa & Fuller 1 992;
Dienes 1 992; Grossberg 1 980, 1 988; Kirsch 1 99 1 ; Lachter & Bever
1 988; MacWhinney, Leinbach, Taraban & McDonald 1 989; Pinker
& Prince 1 988; Smolensky 1 988) and would follow from the act of
noticing frequently co-occurring bigram pairings. Schmidt, for
example, has claimed that implicit learning is probably best
characterized as "the gradual accumulation of associations between
frequently co-occurring features," as in connectionist architectures,
rather than the "unconscious induction of abstract rule systems"
( 1990 : 1 49). Schmidt seems to have in mind here rule systems of the
type proposed by formal linguists (e.g., Chomsky 1 986; Pinker
1 99 1 ). However connectionist models have also been shown to be
capable of representing abstract rule-like knowledge, in the sense of
learning categories like noun, verb, animate and inanimate object
(represented as clusters of hidden units), as a consequence of
learning to predict the sequence of words in simple letter strings fed
to a network during training (Elman 1 988, 1 990). To this extent
connectionist networks are also capable of modeling the rule
abstraction process that Reber claims is responsible for implicit
learning, and, in his own words, "may be able to provide a formal
foundation for examining the underlying processes that we have
long felt lies at the core of implicit learning" ( 1 993 : 1 06).
Implicit learning, memory and knowledge 101
4.3. 1.3 Natural language categorization
Connectionist models have been developed in the domain of
natural language learning (e.g., Gasser 1 990; Henning 1 993 ; Lee &
Gasser 1 992; Robinson 1 994c; Rumelhart & McClelland 1 986;
Shirai 1 992), and a claim that the result of implicit L2 learning is
abstract could also be interpreted within this framework.
Alternatively, however, current linguistic theory offers another
interpretation of the claim that implicit L2 learning is abstract.
This is to identify implicit learning conditions as those which are
necessary for access to Universal Grammar (Chomsky 1 986; cf.
Pinker 1 989, 1 994; White 1 989a), and the rule-based
configurational constraints on parameterization and the shape of
possible languages. Schwartz ( 1 992, 1 993) and Zobl ( 1 992, 1 99 5 )
have recently made this claim with regard to the conditions Krashen
identifies as necessary for the process of language acquisition. They
claim unconscious acquisition processes, but not conscious learning
processes, can access Universal Grammar and facilitate L2
representational change and development by triggering parameter
resetting and recategorization of the target language grammar.
4.3.2 The exemplar-based distributive position.
Vokey and Brooks ( 1 992), along with many critics of Reber's
abstractionist position, propose a 'distributive' view of the
knowledge resulting from the experiments Reber describes in which
"experience with the distributional properties of the domain (e.g.,
frequency, characteristic features, privilege of occurrence) can
remain distributed in the knowledge base and can be exploited by a
process of forming immediate local models or analogies as the need
arises" ( 1 992:328). Similarly, McLaughlin ( 1 980: cf. McLaughlin et
al. 1 983) distinguishes implicit learning, of the purely abstract type
identified by Reber, from 'analogic' learning, of a type similar to
that proposed by Ross ( 1 989) and Skousen ( 1 990). The claim that
implicit learning may derive from exemplar-based learning and
analogic reasoning derives support from a number of recent models
of classification.
4.3.2.1 Estes 's model of classification
Estes' s ( 1 986, 1 992) model was proposed to account for the kind of
learning that occurs when, for example, a person is involved in
classifying plants into classes of fruits and vegetables. It is assumed
that experience results in specific memories for objects called fruits
or vegetables, and that when called on to classify a new plant, its
1 02 Implicit learning, memory and knowledge
overall similarity to exemplars of each of these classes is computed
and the person then assigns it to a class to which it seems more
similar based on a probability related to the ratio of similarities. One
problem with applying this model to the learning process that takes
place during artificial grammar learning experiments is that such
experiments typically contain only grammatical examples during
the training phase. There are no examples of ungrammatical strings
which could serve as the basis for forming two distinct categories.
4.3.2.2 Nosofsky 's exemplar-based classification model
Nosofsky also assumes that categorization is made on the basis o f
computing the similarity o f an experienced event to previously
experienced events. Since the context in which events are
experienced introduces a certain amount of contextual variance into
instance encoding Nosofsky adopts a multidimensional scaling
approach to representing context effects on similarity assessment.
"Exemplars are represented as points in a multidimensional
psychological space, with similarity assumed to be a decreasing
function of distance in the space" ( 1 992:3 64). Selective attention
can be used to "shrink the space along its component dimensions,"
thus facilitating the act of matching similar instances and so
categorizing newly experienced events. In this approach the
positive evidence of grammatical examples presented during
training in artificial grammar experiments would result in a faster
and clearer judgment about category membership in transfer well­
formedness tests, than previously unexperienced ungrammatical
examples.
4.3.2.3 Murphy and Medin 's theory model of classification
A different approach to identifying the process of computing
similarity is taken by Murphy and Medin ( 1 985). They also argue
that the features or units used to compute similarity are not
invariant across contexts, but change as a function of a person's
needs, interests and goals. Thus the judgment of similarity, and
decisions about the relevant attributes to use in computing
similarity, depends on a person's individual, task dependent, theory
for relating them.
4.3.3 Rules, instances, and task demands
The models of exemplar-based classification given above, it should
be stressed, cannot be used to predict how subjects will perform o n
classification tasks when they are given instructions to follow rules
Implicit learning, memory and knowledge 1 03
or to search �or the1!1. There is an extensive literature investigating
the hypothesis testmg and reasoning procedures subjects adopt in
rule discovery tasks (see Evans, Newstead & Byrne 1 993), which
have demonstrated consistent findings with regard to problem­
solving ability and reasoning. One such finding is that o f
conf�rmation bias in rule discovery experiments. Subjects are
consiste.ntly observed to test hypotheses formed during rule-search
by seekmg to confirm predictions the hypothetical rule makes about
t�e structur.e of the pr?blem domain, rather than by seeking t o
disconfirm it. Overlookmg negative, disconfirming evidence often
leads subjects to overly narrow hypotheses based on positive
evidence alone (Klayman & Ha 1 989; Wason 1 960; cf. Bley­
Vroman 1 983). There is also evidence that instructions to apply
rules can preempt the effects of exemplar-based learning. Nosofsky,
Clark and Shin ( 1 989) trained subjects by exposing them to multiple
examples of visual stimuli, then asked them to classify stimuli as old
?r n�w . in a transfer task. Additionally they were given rules for
identifymg old versus new stimuli which contradicted the evidence
of their previous experience. Nosofsky et al. report that subjects
overwhelmingly based their classification decisions on the rules· the
impact of the earlier experiences with the stimuli were forgotten.
As Shanks and St. John ( 1 994) point out, the fact that subjects in
this experiment were following instructions to apply rules does not
mean that the impulse to test and follow a rule cannot be self­
initia�ed unde� task conditions like those in the artificial grammar
le�rnmg expenmen�s of Reber. Shanks and St. John, after reviewing
.
evide�ce for the ab1hty to apply and follow rules in problem-solving
exper�ments, and relating this to artificial grammar learning
expent?ents, conc h-1;de, "t.here are in fact two separate learning
strategies ... The rule mductton strategy is characterized by conscious
effort to develop and evaluate hypotheses which are often
�nrepresentative of the actual grammar... The instance strategy is
mvoked by instructions simply to memorize or observe the training
strings" ( 1 994:427).
4.3.4 The Fundamental Similarity Hypothesis
The evidence of implicit and explicit learning and information
processing reviewed in this and the preceding chapter lend support
.
to the claim that they are the result of fundamentally similar
processes. I refer to this as the 'Fundamental Similarity Hypothesis'
(see. also Robinson 1 995b). Like Shanks and St. John ( 1 994), and
Whittlesea and Darken ( 1 993), I have taken the position that there
1 04 Implicit learning, memory and knowledge
is no evidence that what is learned during implicit learning
conditions is unconscious. Information is processed explicitly,
following the procedure of attending, noticing and rehearsal in
memory described in Chapter 3 . However, different learning
conditions are likely to lead to differential reliance on strategies for
accumulating knowledge of the stimulus domain. The tasks to be
used in this study exemplify such differences. The task instructions
for the implicit memorization condition facilitate the development
of a knowledge base in which conscious memory for strings and
substrings plays a large part. Task instructions for the incidental
condition require subjects to pay greater attention to coding and
remembering propositional meaning than instructions for other
conditions. In contrast, the task instructions for the rule-search and
instructed conditions facilitate the development of a knowledge base
in which the inductive search for, or deductive application of,
abstract configurational relations between syntactic categories plays
a large part. Each of these conditions, however, is likely to promote
a blend of the available means for constructing representations, just
as different memory tasks involve a blend of data-driven and
conceptually driven processing, though facilitating and emphasizing
one or the other through their task instructions. There is no reason
to believe that data-driven and conceptually driven processing, or
the processing requirements of the different tasks used in the
present study, lead to the development of distinct knowledge
sources either, or that they are related to distinct processing
systems differentially regulated by awareness. Rather, they are
mutually compensatory strategies for fulfilling task demands and are
coordinated by the attentional mechanisms described in Chapter 3 .
The Fundamental Similarity Hypothesis, it must be noted, is
based on evidence of adult learning. It may be, of course, that
implicit and explicit learning are more clearly dissociated in child
learning, though there has been little experimental research into
implicit learning processes in precritical period child language
learners, as noted in Chapter 1 , which could be used to confirm or
refute this speculation. The Fundamental Similarity Hypothesis is
thus a logical complement of B ley-Vroman's Fundamental
Difference Hypothesis ( 1 989) which holds that child and adult
language learning are distinct, the one allowing implicit access t o
Universal Grammar and associated domain specific mechanisms for
language learning, the other mediated by conscious problem-solving
using general, domain independent, cognitive mechanisms. In this
acconnt_ imnlicit and exnlicit 1earnin2: would be wavs of
Implicit learning, memory and knowledge 1 05
distinguishing child and adult learning of languages, but no longer
clearly dissociable processes once maturational constraints (Long
1 990) make access to ontogenetically earlier evolved implicit
learning mechanisms problematic for the adult, and once domain­
independent problem-solving procedures become established in the
adult.
4.3. 5 Summary
In summary, the Fundamental Similarity Hypothesis is based o n
three claims which have been motivated by the discussion o f
implicit and explicit learning in the preceding chapters. (i) I t
incorporates Schmidt's noticing hypothesis by claiming that
noticing, under any condition, is necessary to learning (see section
3 .2. 7). (ii) It assumes that learning under any condition results in
the development of a knowledge base in which rule-governed
abstract knowledge, as well as the accumulation of memories for
specific instances, plays a part (see section 4 . 1 .2.3). (iii) It does not
assume that different memory systems responsible for the contents
of implicitly and explicitly learned knowledge, or that different
systems play fundamentally different roles in conscious and
unconscious learning and memory retrieval (see section 3 .3 .6). This
hypothesis, of course, has been largely based on the evidence_ o f
controlled experimental research into artificial grammar learmng,
and studies of implicit memory for words and visual objects. There
is a lack of equivalently controlled experimental research findings in
the domain of natural L2 learning which could be used to either
support or refute this hypothesis. With this in mind the present
study, described in the following section, examines the extent t o
which Reber's methodology for studying implicit learning is
generalizable to the context of SLA.
4.4 The present study
The present study addresses the issue of implicit learning within th·e
context of the induction of rules of a natural language by adult L2
learners. The aim of the study was to examine the extent to which
evidence of learning is similar under Reber's and Krashen' s
conditions, and therefore the extent to which it is possible t o
generalize conclusions about the role o f consciousness i n learning
using Reber' s experimental methodology, which involves learning
the rules regulating letter strings generated by an artificial finite­
state grammar, to more ecologically valid L2 learning conditions.
1 06 Implicit learning, memory and knowledge
Krashen ( 1 982, 1 985) and Reber ( 1 989, 1 993) make similar
claims about the extent of learning possible under these conditions.
Complex rules are more likely to be learned under the implicit
condition for Reber, and under the incidental condition for Krashen,
than under explicit instructed or rule-search conditions. Both Reber
and Krashen agree that simple rules can be learned under explicit
conditions if the structural correlates of the rules are made salient.
To examine these claims 1 04 adult learners of ESL were exposed t o
structures conforming to simple and complex rules of English under
four experimental training conditions. These conditions were: (i) an
implicit condition in which subjects were instructed simply t o
remember presented sentences; (ii) an incidental condition in which
subjects were instructed to answer comprehension questions based on
the sentences; (iii) an explicit rule-search condition in which
subjects were instructed to search for the rules underlying the
sentences; and (iv) an explicit instructed condition in which subjects
were shown rules, then asked to relate them to the sentences.
Two of these conditions, the implicit condition and the explicit
rule-search condition, replicated conditions used by Reber ( 1 969,
1 976, 1 989) in research into the implicit learning of artificial
finite-state grammars. These were matched by two parallel
conditions, the incidental condition and the explicit instructed
condition, which were judged to have greater ecological validity for
L2 learners. It is these latter two conditions that are assumed in
most of the discussion of the role of consciousness in SLA. In
particular Krashen ( 1 98 l a, 1 982, 1 985, 1 994) has claimed that
language 'acquisition' takes place incidentally as a result o f
processing input during meaning-oriented tasks with n o conscious
focus on form. In contrast, Krashen's 'learning' process was
simulated by the explicit instructed condition in which subjects were
required to monitor input using consciously learned rules.
The rules presented during training in the explicit instructed
condition were of different orders of complexity. Sentence tokens
corresponding to these rules were presented during training in all
conditions, exactly matched across conditions for order o f
presentation. Only the instructional treatments differed. The rules
from which the sentence tokens were derived, and which were
themselves presented in the explicit instructed condition, were
identified following the procedures described in 2.4. A transfer
session followed the training session during which new sentences, o f
the same type as those presented i n the training sessions, were
presented. The ability to generalize from instruction on the specific
Implicit learning, memory and knowledge 1 07
trammg set exemplars to the new transfer set exemplars was
therefore assumed to demonstrate the extent of learning. The
dependent measures for performance on the transfer set were speed
and accuracy of response, with fast accurate responses to a
grammaticality judgment test in the transfer phase being taken t o
indicate successful learning o f the rules.
The issue of the influence of language learning aptitude on the
ability to induce rules under all conditions is also addressed in the
present study. Reber has claimed that implicit learning is insensitive
to individual differences in cognitive ability ( 1 989, 1 992; Reber,
Walkenfield & Hernstadt 1 99 1 ), whereas explicit learning is
sensitive to the influence of such abilities. Similarly Krashen
( 1 98 l b, 1 982) has claimed that measures of aptitude access those
cognitive faculties responsible for learning but not for acquisition.
The present study used scores on two subtests of the Modern
Language Aptitude Test (MLAT)(Carroll & Sapon 1 959): paired­
associate learning, a measure of explicit memory, and words-in­
sentences, a measure of grammatical sensitivity, as a test of Reber's
and Krashen' s claims. These were administered prior to the
computer training sessions.
Finally, written reports of the extent to which subjects were
aware of or looking for rules in the four experimental conditions
were elicited from subjects, in their own language, in a debriefing
session. The materials used, and the procedures followed in
administering them, are described in sections 5 .2 and 5.3 of th_e
following chapter. Sections 4.4. 1 and 4.4.2 below summarize the
research design and the research hypotheses for the study.
4.4.1 Summary of the research design
The following factors are included in the present study:
(a) Learning condition, with four levels: (i) implicit condition; (ii)
incidental condition; (iii) explicit rule-search condition; and (iv)
explicit instructed condition, a between subjects factor.
(b) Rules to be learned, with two levels of difficulty: simple versus
complex, a within-subject factor.
(c) Two measures of language learning aptitude: paired-associate
learning and grammatical sensitivity.
4. 4. 2 Research hypotheses
Hypothesis I. The implicit and incidental conditions will pattern
together in all comparisons and the explicit rule-search and explicit
instructed conditions will pattern together in all comparisons.
1 08 Implicit learning, memory and knowledge
Hypothesis 2. Performance on complex structures in the transfer
session will be faster and more accurate for subjects trained under
implicit and incidental conditions than for subjects trained under
explicit rule-search and explicit instructed conditions.
Hypothesis 3. Performance on simple rules will be superior t o
performance on complex rules under all conditions using measures
of speed and accuracy of response.
Hypothesis 4. In all conditions subjects will attest to being more
aware of simple rules than of complex rules during the debriefing
sess10n.
Hypothesis 5. The variance in performance under implicit and
incidental conditions will be equivalent and less than the variance in
performance on rule-search and instructed conditions, which will
also be equivalent.
Hypothesis 6. Measures of language learning aptitude will predict
differences in performance under the explicit rule-search and
explicit instructed conditions but not under the implicit or
incidental conditions.
4.4.3 Rationale for the research hypotheses
Hypothesis 1: The relationship of implicit, incidental, rule­
search and instructed conditions. The first hypothesis relates
Reber's implicit and explicit rule-search conditions to the incidental
and explicit instructed conditions described by SLA theorists and
researchers. The implicit and incidental conditions are alike in not
requiring a conscious focus on the grammatical form of the stimuli
presented during training. In the implicit condition the task
instructions are simply to remember co-occurring words, whereas in
the incidental condition the task instruction is to read for meaning.
Paradis ( 1 994:395) suggests that the implicit and incidental learning
conditions result in identical knowledge bases: "Implicit competence
refers to the knowledge inferred from individuals' performance . . . [it
is] . .. acquired incidentally . . . [and] . . . stored implicitly". To examine
the similarity of learning under implicit and incidental conditions it
was hypothesized that the two conditions would pattern together in
all comparisons.
In contrast to the above two conditions, both the rule-search and
instructed conditions involve a conscious focus on form. In the
latter this involves applying a previously learned rule during
training, and in the former it involves a search for the rules
underlying sentences presented during training. Both Krashen and
Implicit learning, memory and knowledge 1 09
Reber have claimed that learning conditions that involve a
conscious focus on form will produce inferior learning effects,
relative to an implicit condition (in the case of Reber) and an
incidental condition (in the case of Krashen) where there is no such
conscious focus on form. This difference will be most pronounced
where the stimulus domain is complex. However, Reber claims that
even in the case of a simple stimulus domain, where explicit
instruction can be effective, implicit learning will be at least equally
effective. If implicit learning and incidental learning display similar
effects then both conditions should produce evidence of learning
superior to that of explicit conditions on complex rules, and should
pattern together in performance on simple rules.
Hypothesis 2: Performance on complex structures. The second
hypothesis was that implicit and incidental conditions would be
clearly superior to the two explicit conditions in the case o f
learning complex rules. This derives from Reber ( 1 989:223):
"implicit processing of complex materials has an advantage over
explicit processing," and from Krashen ( 1 982:98): "only 'easy'
rules are learnable".
Hypothesis 3: Performance on simple structures. The third
hypothesis was that performance on simple rules would be superior
to performance on complex rules under all conditions, as
demonstrated by measures of speed and accuracy of performance.
This follows from the classification of the rules and structures by
experienced teachers described in 2.4.
Hypothesis 4: The relationship between rule complexity,
learning condition and awareness. Hypothesis 4 predicts that in all
conditions subjects would be more likely to attest to awareness of
rules regulating simple structures than rules regulating complex
structures. Verbalizability is a way of operationalising awareness and
of identifying the objects of awareness. It is clear, though, that
awareness need not imply the ability to verbalize. One can be aware
of something, without being able to put it into words. In the present
study, without time pressure, and using their L 1 , subjects were asked
to describe any rules that they had been aware of during training.
Verbalizability is thus used as a measure of awareness, and as
evidence for claims made about the relationship of awareness to rule
complexity. Reber ( 1 989:220) claims that even in the implicit
condition subjects will become aware of rules if they are salient, and
1 10 Implicit learning, memory and knowledge Implicit learning, memory and knowledge 111

that saliency will be induced by the simplicity of the stimulus Hypothesis 6: The relationship between learning condition and
domain: "A rich and complex stimulus domain is a prerequisite for aptitude. Reber, Walkenfield and Hernstadt ( 1 99 1 ) have also
the occurrence of implicit learning. If the system in use is too claimed that the degree of success achieved under implicit and
simple, or if the code can be broken by conscious effort, then one explicit conditions is differentially sensitive to differences in
will not see implicit processes". Similarly, in the rule-search relevant cognitive abilities, with measures of individual differences
condition Reber claims that subjects will only become aware of in such abilities predicting performance under explicit conditions,
simple rules. Complex rules can only be learned implicitly and will but not affecting performance under implicit conditions: "Implicit
not be accessible to awareness. Although there is no specific functions should operate largely independently of standard measures
motivation for the remaining conditions, it seems obvious that of cognitive capability" ( 1 99 1 :888). Similarly Krashen ( 19 8 1 b) has
subjects will be more likely to be aware of simple rules than o f claimed that aptitude batteries measure the abilities that are drawn
complex rules in the incidental condition. O f course, in the explicit on in ' learning,' but not those that are activated during
instructed condition subjects have been taught both rules, but they 'acquisition' . To examine this claim the present study used
are more likely to forget, or have an imperfect recollection of the performance on two subtests of the MLAT(Carroll & Sapon 1 95 9),
complex rule than of the simple rule. the words-in-sentences test (a measure of grammatical sensitivity)
and the paired-associates test (a measure of memory for foreign
Hypothesis 5: The relationship between learning condition and vocabulary). The final hypothesis was that performance on these
variance in performance. The fifth hypothesis concerns the two subtests would correlate significantly with differences in
relationship of variance in performance to learning under all performance under rule-search and instructed conditions, but n ot
conditions. The hypothesis is that there will be greater population with differences in performance under implicit and incidental
variance when learning takes place under the rule-search and conditions.
instructed conditions, compared to the implicit and incidental
conditions. This hypothesis is motivated by Reber (Reber,
Walkenfield & Hernstadt 1 99 1 ) and Zobl ( 1 992). Reber et al.
( 1 99 1 :888) claim that, "fewer individual differences and smaller
population variances are to be expected when comparing the covert
and unconscious with the overt and conscious functions". Likewise,
Zobl ( 1 992), in a meta-analysis of standard deviations of responses
to items on a variety of grammaticality judgment tests, claims that
performance that engages unconscious, tacit linguistic knowledge
will be more uniform, and display smaller population variances
across the species, than performance which engages learned
encyclopedic knowledge. Zobl's claim that scores will be more
variable where learning is taking place and more uniform when
learners are drawing on the tacit, acquired system, is apparently
corroborated by evidence from VanPatten ( 1 988) regarding the
higher standard deviations for the performance of a number of rule­
focused instructional groups than for meaning-focused instructional
groups on post treatment grammaticality judgment tests. Zobl
( 1 992, 1 995) argues, like Krashen ( 1 9 8 1 b, 1 982), that this evidence
of uniformity of performance is the result of unconscious
acquisition processes facilitated by the meaning-focused treatments.
 Chapter 5

Methodology For The Present Study

5. 0 Introduction
The present chapter describes the subjects for the study, the data
collection procedures followed, the pretest materials and computer
training and testing programs used, and details of the analyses
performed to address the research hypotheses listed in section 4.4.2.

5.1 Subjects
Subjects for the study were 1 04 non-native speakers of English,
enrolled in English language programs at four institutions in Hawai' i
during the period from October 1 993 to February 1 994. Subjects
were predominantly native speakers of Japanese. The programs
were the STEP-UP program at Tokai University at Honolulu, the
NICE program at the University of Hawai'i, and the English
language programs of the Japan America Institute of Management
Science (JAIMS) and Kansai Gaidai Junior College, both located in
Hawai'i Kai on the outskirts of Honolulu. These institutions were
selected as subject sources because of the relative homogeneity o f
their student populations. Tokai University and Kansai Gaidai are
both Japanese owned junior colleges which offer credit courses in
English and English language programs for those students needing t o
improve basic English ability before entering the credit programs.
Students at Tokai University tend to be from Japan, in their late
teens and early twenties. The Japanese students have all completed
school level education in Japan, including seven years of formal
classroom instruction in English. Students at Kansai Gaidai are all
native speakers of Japanese, and the group who participated in the
present study were all aged between eighteen and nineteen. Like the
students at Tokai they have all completed school level education in
Japan, including seven years of formal classroom instruction in
English. The students from JAIMS who participated in the present
study were all from a group enrolled in a management training
course. They were graduates from universities or colleges in Japan in
1 14 Methodology for the study
their early to mid twenties. Students from the NICE program were
predominantly native speakers of Japanese, in their early to mid
twenties, who had come to Hawai'i to improve their conversational
English. Initial analyses showed subjects from all source institutions
to be performing similarly under all conditions.
A small number of non-Japanese native speakers recruited from
Tokai University and the NICE program were also included in the
subject pool (five Chinese and five Korean) and randomly assigned
to the experimental conditions. These subjects also reported
themselves to have received between six and eight years of formal
schooling in English before coming to Hawai'i to enroll in English
programs. Since no hypotheses were motivated by claims about
specific differences in proficiency levels, no attempt was made to
control for proficiency other than by selecting students from low t o
high intermediate level classes within the source institutions. The
homogeneity of the subject pool was maintained to the extent that
no beginning or extremely advanced subjects were used. What
differences there might have been between the proficiency levels o f
subjects included in the study were assumed not to have unduly
affected performance on any particular condition, since all subjects
were randomly assigned. Details of subjects included in the study,
including age, gender, native country, L l , and condition assigned are
given in Appendix K.
Subjects were recrnited during visits to classes arranged by the
Directors of the source institutions. During these visits the nature of
the research was briefly described as research into the use of
computers to help language learners. The demands in terms of time,
pre-tests, treatment and post-tests were explained. Subjects were
offered $ 1 0 in return for their participation. Subjects were
subsequently selected for inclusion in the study from amongst those
who volunteered on the basis of the results of a pretest showing
them to be unfamiliar with the rules to be presented in the study.
5.2 Data collection procedures
The data collection took place on site at Tokai University, JAlMS
and Kansai Gaidai College. At Tokai University data collection took
place in a computer laboratory equipped with Macintosh computers.
At JAIMS and Kansai Gaidai the researcher brought a Macintosh
computer and set it up in a quiet, convenient location. The
researcher met students from the NICE program at an appointed
time and place on the University of Hawai' i campus and took them
to an office equipped with two Macintosh computers. No intact
Methodology for the study 1 15
classes were used during any period of data collection, and no more
than three subjects completed the computer training sessions at any
one time, since the individual attention of the researcher was needed
for each subject during the early familiarization phase of the
training sessions to ensure that the task directions were fully
understood.
5.2.1 Pretesting and aptitude testing
Subjects first completed a grammaticality judgment pretest. This
consisted of the first thirty items of the grammaticality judgment
test used in the pilot study to identify unfamiliar structures described
in Chapter 2, section 2.4 (Appendix A). Subjects were told to circle
either 'grammatical,' 'ungrammatical,' or ' not sure' beneath each
sentence. They were told to make only one answer and to ignore
what they thought might be punctuation or spelling errors in
making their decision. The procedure was illustrated for them using
an example sentence. If subjects indicated uncertainty about the
meaning of any of the terms after this example had been completed
the researcher attempted to further clarify them. Subjects then
completed the test. They were given as much time as they needed
and generally took about five minutes. If subjects circled
'ungrammatical' for the sentences corresponding to the simple and
complex rules, they were selected for inclusion in the study. These
were 'Into the house ran John' and 'Where the cheese is is in the
basket not in the bag' . More extensive testing of their knowledge o f
these rules was avoided since drawing the subjects' attention at the
pretest stage to the structures to be learned during the treatments
would have threatened the validity of the study, since both the
implicit and incidental conditions required that the subjects not be
aware of or be looking for grammatical rules in the training sessions.
It was felt that the other distractor items in the pretest would mask
the relevant items and divert attention from them, and that the
following aptitude tests would further disguise the structures that
were the focus of the treatments. Taken together with the evidence
of the pilot study reported in section 2.4. 1 , which confirmed these
structures and their associated rules to be unfamiliar to a comparable
group of subjects, it was felt that this pretest evidence of rule
knowledge alone was sufficient for subject selection.
Following the pretest the subjects selected for the experimental
treatment were asked to complete two aptitude tests. The memory
test consisted of a list of twenty English and Kurdish paired­
associates. It was explained to the subjects that the Kurdish word
1 16 Methodology for the study
meant the same as the paired English word in another language and
they were instructed to try and memorize the pairings for a
subsequent memory test. To aid them in this task a number of words
were given on a practice sheet and they were instructed to search
the two columns of pairs to match each word with its associate.
They were told that when they finished this they were to continue
studying the vocabulary and memorizing the pairings. They were
given two minutes to memorize the vocabulary, and then asked to
complete the twenty four item multiple-choice test. In this test a
Kurdish word was given together with a choice of five English
words. Subjects had to choose the correct English equivalent for
each of the twenty four Kurdish items. The time limit was four
minutes and the material used was from the short form of the
MLAT (Carroll & Sapon 1 959).
The remaining aptitude test was the words-in-sentences test of
grammatical sensitivity, also taken from the short form of the
MLAT. This is a test of the ability to identify analogies between
the grammatical function of underlined words in two sentences. The
procedure for answering these questions was explained to each
student by the researcher using the examples on the cover page of
the test. When the subjects indicated they understood the task they
were allowed to turn the page and complete the test. Subjects were
told they could take up to forty minutes to complete this test,
though they generally took much less time than this.
5.2.2 Assignment to experimental conditions
After completing the pretest and the aptitude tests subjects were
randomly assigned to one of the four conditions referred t o
previously: an implicit condition, an incidental condition, an
explicit rule-search condition and an explicit instructed condition.
All conditions consisted of two phases, a training and a transfer
phase. The training phase for each condition consisted of an initial
task familiarization phase, in which the demands of the training
task were explained to the subject on the computer screen. At this
point the researcher was available to help clarify procedures and
answer questions. After familiarizing themselves with the task t o
their own and the researcher's satisfaction, subjects began the first
of the two training trials. Each trial consisted of a sequence of forty
timed stimulus presentations, together with follow up questions t o
each stimulus requiring a yes/ n o response, and feedback to the
response. Stimulus sentences were timed to remain on screen for ten
seconds, after which they disappeared from the screen and were
Methodology for the study 1 17
replaced by the follow up question. This remained on screen until
the response keystroke, after which the feedback appeared o n
screen. This remained on screen until the 'press any key t o
continue' prompt was followed, whereupon the next stimulus
sentence appeared for ten seconds. The follow up questions differed
according to the treatment that was being delivered. The stimulus
presentation, follow up question, and feedback sequences for each
condition are described below in sections 5 .2.3-5.2.6. The complete
training set of sentence stimuli and follow up questions for each
condition are given in Appendix H. In each condition the stimuli
were the same matched sequence of forty sentences, twenty
generated by a simple rule and twenty generated by a complex rule.
5.2.3 The implicit learning condition
In the training phase for the implicit learning condition, subjects
were first given an opportunity to familiarize themselves with the
demands of the task by reading an explanation of the procedure
they were to follow on the computer screen (see Appendix C). A t
this point the researcher observed and answered questions about the
procedure. When the task procedure had been explained to their
satisfaction the familiarization session ended and the first of two
training trials began on the computer. In all conditions the second
trial was a repetition of the first. During each trial subjects were
presented with sentences conforming to simple or complex rules
and instructed to remember them. After presentation of each
stimulus sentence subjects were required to respond yes or no to a
question asking them if a particular sequence of two words had been
used in the previously presented stimulus sentence. They were
instructed to respond using the keys 'c' for correct, or yes, and ' m '
for mistake, or no. They were given feedback in the form o f a
correct/ incorrect message after each response.
5.2.4 The incidental learning condition
In the training phase for the incidental learning condition subjects
were also first given an opportunity to familiarize themselves with
the demands of the task by reading an explanation of the procedure
they were to follow on the computer screen (Appendix D). At this
point the researcher observed and answered questions about the
procedure. When the task procedure had been explained to their
satisfaction, the familiarization session ended and the first training
trial began on the computer. After presentation of each stimulus
sentence subjects were presented with a question about the
1 18 Methodology for the study
propositional content of the sentence they had just seen, to which
they had to respond by typing yes or no by using the 'c', for correct
and 'm' for mistake keys, as in the implicit condition. As in the
implicit condition they were also given feedback in the form of a
correct/ incorrect message after each response.
5.2.5 The explicit rule-search condition
As in the above conditions, in the training phase for the rule-search
condition subjects were first given an opportunity to familiarize
themselves with the demands of the task by reading an explanation
of the procedure they were to follow on the computer screen
(Appendix E). The researcher observed and answered questions
about the procedure until the task procedure had been explained t o
their satisfaction. The familiarization session then ended and the
first of the two training trials began on the computer. In the
familiarization phase for the explicit rule-search condition subjects
were told that the sentences they were about to see exemplified
rules of English. They were instructed to try and identify the rules
underlying the sentences and told that they would be asked t o
describe the rules they had found at the end of the training session.
After presentation of each stimulus sentence during training they
were asked questions about the rules they were searching for. These
questions were phrased so as to require a yes/ no response (e.g.,
Have you identified the rules yet? Are you still looking for the
rules? Do you know the rules for sentences like this?) which they
were to make by using the keys 'c' for yes, or correct, and 'm' for
no or mistake. Since these questions were open ended, unlike the
questions in the other conditions, no feedback was given after each
yes/ no response, and the next stimulus appeared on screen
immediately following the response keystroke.
5.2. 6 The explicit instructed learning condition
The familiarization phase for the explicit instructed condition was
similar in form to that for the other conditions. Subjects were first
asked to read through the two rules that were the focus of the study
(see Figures 5 and 6 in section 2.4). The researcher explained t o
subjects i n this condition that they would be asked t o learn the two
rules, and would be able to practice them on the computer. Subjects
were then asked to read the rules as the researcher explained them .
The key points of the explanations of these rules were rehearse.cl
during the familiarization phase for the explicit instructed
condition. The explanation, examples, and the sequence o f
Methodology for the study 1 19
comp_uter �creen disp�ays presented during the familiarization phase
a�e given m Appendix F. As can be seen in Appendix F, (screen
displays 2 and 6) the familiarization phase drew attention to four
key points regarding each of the rules. These related to violations
of the grammaticality of sentences which were constitutive of the
ungrammatical types presented during the transfer phase (see 5 . 3 .4
and _s . � .5). After subjects in this condition completed the
. _ _
famihanzatton exercise the training set began. Subjects were told
that they could consult the written versions of these rules in
answer�ng ques�io_ns duri�g the computerized training session. During
e�ch tnal of trammg, as m the other conditions, subjects first saw a
stimulus sentence then were asked a question requiring a yes/ n o
response. These questions were metalinguistic and related the
stimulus sentence to the explanation of the rule given in the
familiarization phase, e.g., 'Did the verbs used agree in tense?' 'Did
the �ubject of the sentence come after the verb?' Each question
reqmred a yes/no response. As in the implicit and incidental
conditions subjects were given feedback to each response in the
form of a correct/ incorrect message which remained on screen until
they responded to the 'press any key to continue' prompt.
5.2. 7 The transfer phase
The transfer pha� e was the same for all conditions and immediately
.
followed the trammg phase. After finishing training subjects were
told that they were to complete a grammaticality judgment test o n
the co�puter and instructed to respond as quickly as they could t o
each stimulus by pressing the ' c ' key if the sentence was a
grammatically acceptable one and 'm' if it was a grammatically
unacceptable one. Subjects first completed a familiarization task on
the co�puter before seeing the randomized set of twenty
grammatical and twenty ungrammatical sentences (Appendix I).
Accuracy for each subject's responses, together with the time taken
to make each response, measured in milliseconds, was recorded by
the software used to write the training and transfer set programs.
5.2.8 The post-experimental debriefing session
After completing the transfer phase subjects were given a set of five
written questions and instructed to answer them in the space
provided using their L l . Answers to three of the questions were used
�o �ode the a�areness of subjects for the present study, these were;
Did you notice any rules of grammar when you did the first two
exercises on the computer?' , ' Were you looking for grammar rules
1 20 Methodology for the study
when did the first two exercises on the computer?', and 'Can you
say what the rules are?' On the basis of these responses subjects
were coded as having noticed, or not; as having looked for rules, or
not; and as able to verbalize rules, or not. Answers were later
translated back into English for the purposes of data analysis.
Despite the researcher's instructions, some subjects chose t o
respond using English. The questions were written in Japanese for
Japanese subjects. The few non-Japanese subjects were given English
versions of the questions. In each case the researcher explained the
questions in English to the subject. If asked for clarification the
researcher responded by giving necessary examples, or by reminding
subjects of the experimental procedures they had just followed in
the particular condition to which they had been assigned.
5.3 Materials
5.3.1 Computer materials
All training and transfer sessions were completed at Macintosh
computer terminals using standard qwerty keyboards. The
Macintosh computers used were SE, SEII, and Hsi models. The
stimulus sentences were presented in black, on a white background,
and written in size 1 2 Geneva font. The training and transfer
programs which recorded response time and accuracy were written
using Mindlab software, version 2 . 1 (Meike 1 988). This software
allows the researcher to construct programs which consist of a series
of stimuli, in the form of a pictures, numerals, words, or sentences.
Each stimulus can be timed to remain on screen from any period
between 1 /60th of a second and 1 8 seconds, measured in units o f
1 /60th o f a second. Presentation time for the stimuli presented
during training in the present study was fixed at 1 0 seconds. In the
transfer session stimuli remained on screen until the response
keystroke. The time taken for each subject's responses during
training and transfer sessions was recorded in milliseconds, together
with accuracy for each response, which was recorded as 0, for
incorrect, and 1, for correct. Additionally the subject number, trial
number, and the key used to make each response were also recorded.
5.3.2 Pretest and aptitude measures
The pretest materials consisted of responses to the first thirty items
used in the grammaticality judgment pilot test of rule familiarity
(Appendix A). The materials for the associative memory and the
words-in-sentences subtests were photocopied from the short form
Methodology for the study 121
o f the MLAT (Carroll & Sapon 1 959). For each o f these tests,
subjects responded by circling answers on the test materials.
5.3.3 Rules
During the instructed condition, subjects were shown written
explanations of the simple and complex rules which they were asked
to read and consult during the training sessions (see Figures 5 and 6).
5.3.4 Training set sentences
The training set sentences consisted of twenty grammatical
structures illustrating the simple rule and twenty grammatical
structures illustrating the complex rule (Appendix G). These were
randomly ordered for presentation in the four conditions.
(i) Simple rule training set sentences. The twenty sentences
illustrating the simple rule were made up of four tokens of five
types. These types were the same as the types used in the
explanation of the rule given to teachers to rate for complexity and
given to subjects to study in the instructed condition. Type 1 were
sentences in the order, subject, verb, location phrase (e.g., Pau/
drove across the city). Type 2 were sentences in the order location
phrase, verb subject (e.g., Up the road came Eric). Type 3
sentences were in the order location phrase, subject, verb (e.g.,
Round the track Bill raced). Type 4 sentences were in the order
subject, verb, time phrase (e.g., Lori ate at five o 'clock), and Type 5
sentences were in the order time phrase, subject, verb (e.g., In the
afternoon Elaine arrived).
(ii) Complex rule training set sentences. Similarly, the twenty
grammatical sentences illustrating the complex rule were also made
up of four tokens of five types. These types were also motivated by
the description of the complex rule presented to teachers to rate for
complexity. Type 1 were sentences containing two conjoined
clauses which contrasted the locations of two things (e.g., The girl
plays in the park and the boy plays in the yard). Type 2 were
sentences with singular subjects, and two forms of the verb 'to be'
(e.g., Where Sue is is in the car not on the boat). Type 3 were
sentences with plural subjects, requiring agreement with a plural
form of the main verb 'be' (e.g., Where Jim and Amy are is in the
kitchen not in the garden). Type 4 were sentences with singular
subjects and lexical main verbs (e.g., Where the President lives is in
Washington not in Chicago), and Type 5 sentences contained plural
/
1 22 Methodology for the study
subjects and lexical main verbs (e.g., Where Steve and Mark swim is
in the pool not in the ocean). Additionally two of the four tokens
illustrating the complex Type 2, 3, 4 and 5 sentences consisted of
the optional phrase beginning with 'not', described at the end of the
complex rule. In writing tokens of each of these types, only simple
vocabulary and common English names were used. No location
phrases or lexical verbs were repeated in either simple or complex
sentence tokens.
5.3. 5 Transfer set sentences
The transfer set sentences were made up of twenty grammatical and
twenty ungrammatical sentences (Appendix J). The grammatical
sentences were two new tokens of each of the ten types presented
during the training sessions. The ungrammatical sentences were
systematic violations of these grammatical types, as described in the
rules presented to teachers to rate for complexity, and to subjects to
study in the instructed condition.
(i) Simple rule transfer set sentences. Two new tokens of each of
the five simple rule sentence types presented during training were
presented in the transfer set. In addition to these ten grammatical
sentences, there were ten ungrammatical sentences which violated
conditions on the rule. These were divided into two types, sentences
in which subject-verb inversion occurs following a time phrase (e.g.,
On Saturday night danced Charlie), and sentences in which subject­
verb inversion occurs preceding time and location phrases (e.g.,
Crawled Mick through the tunnel). These are violations of the
simple rule which states that subject-verb inversion can only occur
when location phrases are fronted. There were five tokens of each
of these two types of violation, making a total of ten
ungrammatical sentences.
(ii) Complex rule transfer set sentences. As with the simple rule,
two new examples of each of the five grammatical sentence types
illustrating the complex rule were presented during the transfer
session. In addition to these ten grammatical sentences, ten
ungrammatical sentences were presented. These ten ungrammatical
sentences were of five types, corresponding to violations of the
complex rule that were explained in the rule. The first
ungrammatical type violates the condition that subject-verb
inversion is not possible following ' Where' (e.g., Where is Bill is in
the country not in the city). The second ungrammatical type violates
Methodology for the study 1 23
the condition that the form of the verb 'be', following the m ain
verb must always be singular, and need not agree in number with the
subject (e.g., Where Carl writes are at a desk not on the floor.I Where
the soldiers fight are in Europe not in Hawaii). The third
ungrammatical type violates the condition that the two verbs used
must agree in tense (e.g., Where the bird is was in the sky not in the
sea). The fourth ungrammatical type violates the condition that
sentences of this kind must begin with ' Where' (e.g., What the King
lives is in a palace not in an apartment). The final ungrammatical
type violates the condition that when the second location is added
to contrast with the focused location, it must be introduced by
'not'(e.g., Where Mark works is in America in Japan). There were
two tokens of each of these five ungrammatical types making a
total of ten ungrammatical sentences. All sentences in the transfer
session were therefore either new examples of the previously
encountered grammatical sentence types, or new examples o f
violations o f those sentence types with regard to specific conditions
described above. As with the training session, only common verbs,
nouns and names were used, and no verbs, nouns, or location phrases
were repeated across the forty sentences presented in the transfer
session.
5.3. 6 Debriefing materials
The debriefing questions were written on two pieces of paper, in
Japanese, by a native speaker of Japanese, and in English by the
researcher. Enough space was left beneath each question for subjects
to complete their answers, though if they felt they needed to they
were told to use the blank pages on the reverse of the questionnaire
to complete their answers.
5.4 Analyses
Data was analysed using the statistical packages ' Statview SE +
Graphics,' and 'SuperANOVA' available for use on Macintosh
computers. A Macintosh Hsi was used for the data analysis.
To address Hypothesis 1 , that performance on the implicit
condition and incidental condition would pattern together, and that
performance on the rule-search condition and the instructed
condition would pattern together, two repeated measure ANOVA o f
accuracy and reaction time o f responses to all transfer set sentences
were performed. Planned comparisons between the implicit and
incidental conditions, and between the rule-search and instructed
conditions were made. A third planned comparison between the
/
124 Methodology for the study
joint performance of the implicit and incidental conditions versus
the rule-search and instructed conditions was also performed. The
planned comparisons were followed by post-hoe comparisons of all
training conditions.
To address Hypothesis 2, that performance on complex rules
would be superior for those trained in the implicit and incidental
condition, compared to performance by those trained in the rule­
search and instructed conditions, the factor Rule was removed from
the two repeated measure ANOVA. The ANOVA for accuracy and
reaction time on easy and hard rules were then followed by the same
series of planned and post-hoe comparisons used to address
Hypothesis 1 .
To address Hypothesis 3 , that performance on easy rules would
be superior to performance on hard rules for all conditions, the
factor Condition was removed from the two repeated measure
ANOVA for accuracy and reaction time. The ANOVA for each
condition was then performed separately to identify whether the
factor Rule was significant.
To address Hypothesis 4, that subjects in all conditions would be
more likely to attest to being aware of simple rules than of complex
rules during the postexperimental debriefing session, a Chi-square
analysis of subjects' responses to the debriefing questionnaire was
performed.
To address Hypothesis 5, that variance in performance under
implicit and incidental conditions will be equal and less than the
variance in performance under rule-search and instructed conditions,
which will also be equal, six paired F-tests were performed
comparing variance in accuracy and reaction time in the transfer
grammaticality judgment task.
To address Hypothesis 6, that measures of language learning
aptitude will be related to performance under rule-search and
instructed conditions, but not to performance under implicit or
incidental conditions, accuracy and reaction time scores for the
transfer task were correlated with scores on the aptitude tests.
Pearson's r was reported, together with the significance level of
each correlation.
Given the large number of comparisons made between groups in
the present study the alpha level for testing the six hypotheses was
set at the .0 1 level of significance.
Chapter 6
Analyses And Results
6. 0 Introduction
This chapter describes the analyses performed to address the six
hypotheses for the present study which were listed at the end o f
Chapter 4 . The presentation o f results is organized around the four
different types of data used to address these hypotheses. The
accuracy and reaction time data for performance on the transfer
grammaticality judgment task are first described in section 6. 1 . and
are used to address Hypotheses 1 , 2 and 3 in sections 6. 1 . 1-3 . Next,
the rule awareness data are described in section 6.2. These data are
used to address Hypothesis 4 in section 6.2. 1 . Thirdly, the variance
data are described in section 6.3, and are used to address Hypothesis
5 in section 6.3 . 1 . Finally the aptitude data are described in section
6.4 and are used to address Hypothesis 6 in section 6.4. 1 . A number
of issues arising from the analyses, which are not integral to the
research hypotheses addressed here, are examined in more detail in
the following chapter, for example the effects of grammaticality
and sentence type on reaction time and accuracy in the transfer
task, and the relationship of the awareness data to the aptitude data.
6.1 Accuracy and reaction time data
Accuracy and reaction time for grammaticality judgments on 4 O
sentences presented during the transfer session were recorded.
Correct responses were scored as 1 and incorrect responses were
scored as 0. Raw scores in milliseconds were entered for response
reaction time. Twenty six subjects in each condition made responses
to 20 grammatical items, and 20 ungrammatical items, forming a
total of 40 responses to sentence tokens. The accuracy data are
thus based on correct judgments of ' grammatical' for grammatical
sentences, and correct judgments of 'ungrammatical' for
ungrammatical sentences. For the population of 1 04 subjects the
expected frequencies for chance performance on all sentences in
training are thus 520 responses in each cell of an accurate/
126 Analyses and results Analyses and results 1 27

inaccurate two cell matrix for grammatical sentences. The same is rule is lower for all conditions than mean reaction time for
also true for ungrammatical sentences. Using the Chi-square statistic performance on the hard rule. Reaction time for both easy and hard
to calculate probabilities of accurate and inaccurate responses, a rules is faster for those trained in the rule-search condition than for
ratio of 562 accurate to 478 inaccurate responses, (JO (562) -fE those trained in the implicit, incidental or instructed conditions.
(520) = 32*32/520 = 3 .39*2 = 6.7) or an accuracy level of 5 4%, These data are also normally distributed.
(562/ 1 040) would be significantly above chance at the p<.0 1 level
of significance. Considering the easy rule and hard rule sentences
separately, a ratio of 288 accurate to 232 inaccurate responses, or Table 6. Descriptive Statistics For Transfer Test Reaction
an accuracy level of 56%, would be above chance at the p<.0 1 level. Time On Easy And Hard Rules Under Each Condition
The descriptive statistics for response accuracy to sentences
conforming to easy and hard rules, split by condition, are given in Condition/Rule Mean Reaction Time SD SE
Table 5. Mean accuracy on the easy rule is higher for all conditions
than mean accuracy on the hard rule. Accuracy on both easy and Implicit
hard rules is higher for those trained in the instructed condition than Easy rule 8 148.838 303 1 .08 5 94.444
for those in the implicit, incidental or rule-search conditions. Mean Hard rule 9958.427 3325.64 1 652.2 12
Incidental
accuracy by all conditions on easy and hard rules is above 5 6% and
Easy rule 86 1 7.8 1 7 2705.89 530.669
is therefore significantly above the level of chance. The data are Hard rule 1 0679.067 27 1 9.573 533.352
normally distributed with no excessive kurtosis or skewedness. Rule-search
Easy rule 671 6.496 2704.865 53 0.468
Hard rule 8 1 6 1 .8 8 1 2720. 1 7 533 .469
Table 5. Descriptive Statistics For Transfer Test Accuracy O n Instructed
Easy And Hard Rules Under Each Condition Easy rule 84 1 8.063 2 1 07.83 1 4 13.38
Hard rule 9980.9 2902.382 569.204
Condition/Rule Mean % Accurate SD SE

Implicit 6.1.1 Analyses for Hypothesis 1

Easy rule 66.346 1 5 .202 2.98 1 The first hypothesis of the present study is that the implicit and
Hard rule 6 1 .346 1 2.927 2.535 incidental conditions will pattern together in all comparisons, and
Incidental the rule-search and instructed conditions will 'pattern together in all
Easy rule 72.308 1 4.576 2.859 comparisons. To address this hypothesis two repeated measure
Hard rule 58.654 I 0 .729 2 . 1 04
ANOVA, for accuracy and reaction time, were carried out. Three
Rule-search
1 5 .679 3 .075
planned comparisons of the results of each ANOVA were also made.
Easy rule 69. 6 1 5
Hard rule 56.923 1 2.576 2.466 These were comparisons of the results for implicit and incidental
Instructed conditions, the rule-search and instructed conditions, and a
Easy rule 84 .6 1 5 1 3 .923 2 .73 comparison of the results for the combined implicit/incidental
Hard rule 66.923 1 3 .044 2.558 conditions versus the combined rule-search/instructed conditions.

6.1.1.1 Accuracy
The descriptive statistics for reaction time in milliseconds to all The results of the repeated measure ANOVA for accuracy are given
sentences conforming to easy and hard rules, split by condition, are in Table 7. There is a significant main effect for Condition (F
given in Table 6. Mean reaction time for performance on the easy (3/ 1 00) = 6.799, p<.O l ) and for Rule (F ( l/ 1 00) = 70.566, p<. 0 1 ).
1 28 Analyses and results Analyses and results 1 29

There is no significant main effect for Grammaticality. The means

for performance on easy and hard rules across conditions are 0 Hard rule
illustrated in Figure 1 4 . There is no significant interaction between • Easy rule
Rule and Condition. There is an almost significant interaction of
.9
Grammaticality and Condition (F (3/ 1 00) = 3 .953, p = . 0 1 04), and
a significant interaction of Rule and Grammaticality (F ( 1/ 1 00) = .85
9.253, p<.0 1 ). These interactions are discussed in section 6. 1 .2 and
in more detail in Chapter 7, since they are not integral to the >.
u .8
predictions of the main hypotheses addressed in this chapter � u
.75
Table 7. ANOVA Table For Accuracy To Transfer Set �
.......

Sentences
0 .7
"'
!a
tU
.65
Source df Sum of Mean F-Value P-Value �
......
Squares Square Q) .6
u
c 3 10.7 1 6 3 .572 6 .799 .0003
Subject(Group) 1 00 52.539 .525 .55
R 1 1 5 .63 1 1 5 .63 1 70.566 .000 1 Implicit Incidental Rule-search Instructed
R * C 3 2. 193 .73 1 3.300 .0235
R * Subject(Group) 1 00 22 . 1 5 1 .222 condition
G 1 3 .877 3 . 877 5 .232 .0243
G * C 3 8.789 2 .930 3 .953 . 0 1 04 Figure 14. Mean Accuracy For Performance On Easy And
G * Subject(Group) 1 00 74. 1 09 .74 1 Hard Rules Under Each Condition
s 9 2 1 .925 2.436 1 4.706 .000 1
S * C 27 5 .964 .22 1 1 .334 . 1 200
S * Subject(Group) 900 149.086 . 1 66
1 4.779 4.779 9.253 .0030
The result of the planned comparison of the implicit and
R* G
R* G * C 3 1 .945 .648 1 .255 .2940 incidental conditions shows no significant difference in accuracy
R * G * Subject(Group) 1 00 5 1 .65 1 .5 1 7 between these groups. However there is a significant difference
R * S 9 12.583 1 .398 8 .077 .000 1 between the rule-search and instructed conditions (F ( 1 /5 0) =
R* S * C 27 4.4 1 0 . 1 63 .944 . 5480 1 5 .465, p<.0 1 ). There is no significant difference the combined
R * S * Subject(Group) 900 1 5 5 .782 . 1 73 implicit/incidental conditions and rule-search/instructed conditions .
G * S 9 9.462 1 .05 1 6.91 1 .000 1 As Figure 1 4 shows, the difference between the rule-search and the
G * S * C 27 1 0 .6 1 2 .393 2. 584 .000 1 instructed condition is due to the greater accuracy of the instructed
G * S * Subject(Group) 900 1 36.90 1 . 1 52 subjects in performance on both rules. Post-hoe Scheffe
R* G * S 9 1 3 .444 1 .494 9.308 .000 1 comparisons also show the instructed condition to be significantly
R * G * S * C 27 5 .495 .204 1 .268 . 1 64 1
superior in accuracy to the implicit and the rule-search conditions
R * G * S * Subject{Grou£2 900 144.436 . 1 60
(p<.0 1 ), and an effect approaching significance for a difference
Dependent: Accuracy
between the instructed and incidental conditions (p = . 0 1 85).
Key: C Condition (Implicit, Incidental, Rule-search, Instructed)
6.1.1.2 Reaction time
=

R Rule (Easy rule, Hard rule)

The results of the repeated measure ANOVA for reaction time are
=

G = Grammaticality (Grammatical, Ungrammatical)

S = Sentence token given in Table 8. The main effect for Condition approaches
130 Analyses and results Analyses and results 131

significance (F (3/1 00) = 3 . 5 02, p = . 0 1 82) and there is a significant Table 8. ANOVA Table For Reaction Time To Transfer Set
main effect for Rule (F ( 1 /1 00) = 84.053, p<. 0 1 ). Mean reaction S e n te n c es
time for the rule-search condition is lower than for all other
conditions and mean performance on the easy rule is lower than Source df Sum of Mean F-Value P-Value
performance on the hard rule (see Figure 1 5). There is n o Squares Square
significant main effect o f Grammaticality o n reaction time, and n o c 3 2.902E9 9.673E8 3 .502 . 0 1 82
significant interactions o f Rule, Condition and Grammaticality. Subject(Group) 1 00 2.762E 1 0 2.762E8
The result of the planned comparison of reaction time for R 1 3 . 074E9 3 .074E9 84.053 .000 1
R*C 3 58301 1 26.7 l .943E7 .53 1 .66 1 8
implicit and incidentally trained subjects shows no significant
R * Subject(Group) 1 00 3 .657E9 3 .657E7
difference between these groups. However there is an effect G 1 5062 1 744.3 5 .062E7 1 .995 . 1 609
approaching an effect approaching significance for a difference G * C 3 39228742.5 l .3 08E7 .515 .6727
between these the faster rule-search condition and the instructed G * Subject(Group) 1 00 2.53 8E9 2.53 8E7
condition (F ( 1 150) = 5 . 83 5, p = . 0 1 75). There is no significant s 9 4.785E9 5.3 1 6E8 22. 1 7 1 .000 1
difference between the combined implicit/incidental versus rule­ S * C 27 77549565 1 .8 2.872E7 1 . 198 .2243
search/instructed conditions. Post-hoe Scheffe comparisons show no S * Subject(Group) 900 2 . 1 58E10 2.398E7
significant differences between any of the groups at the p<.0 1 level. R*G 1 7453 8 120.5 7.454E7 3 . 142 .0793
R*G*C 3 1 7 1 750 1 8 1 57250060.3 2.4 1 3 .07 1 1
R * G * Subject(Group) 1 00 2.372E9 2.372E7
0 Hard rule R * S 9 2.36E9 2.623E8 1 1 .986 .000 1
• Easy rule R * S * C 27 866 1 3 1 574.3 3 .208E7 1 .466 .0598
R * S * Subject(Group) 900 l .969E 1 0 2. 1 88E7
1 1 000 G * S 9 3 .592E9 3 .992E8 1 7.3 1 9 .000 1
1 0500 G * S * C 27 636080850.9 2.356E7 1 .022 .4340
<I)
G * S * Subject(Group) 900 2.074E 1 0 2.305E7
.§ 1 0000 R* G * S 9 l .952E9 2 . 1 69E8 9.58 1 .000 1
......
s:: 9500 R* G * S * C 27 3 9 1 44 1 9 8 1 .5 l .45E7 .640 .92 1 8
.9
...... R * G * S * Subject�GrouE} 900 2.038E l 0 2.264E7
<.)
ro 9000 Dependent: Reaction time
�
<+-< �500
0
"' Key: C = Condition (Implicit, Incidental, Rule-search, Instructed)
§<I) 8000 R = Rule (Easy rule, Hard rule)
� 7500 G = Grammaticality (Grammatical, Ungrammatical)
-
S Sentence token
Q)
=

u 7000
6500
Instructed
6.1.1.3 Summary of the evidence for Hypothesis 1
Implicit Incidental Rule-search
The results of the analyses of accuracy and reaction time data to all
condition sentences in the transfer task shows no significant difference
between the implicit and incidental conditions. To this extent they
Figure 15. Mean Reaction Time For Performance On Easy pattern together, partially confirming Hypothesis 1 . However,
And Hard Rules Under Each Condition when the factor Rule is considered differences between the
conditions emerge. For the incidental condition there is a significant
difference in accuracy for performance on easy and hard rules,
132 Analyses and results
whereas there is no significant difference in accuracy of
performance on easy and hard rules for the implicit condition. This
result is addressed in more detail in section 6. 1 .3 .
While the implicit and incidental conditions show no significant
differences in accuracy or reaction time, there are significant
differences between the rule-search and instructed conditions with
respect to accuracy. They cannot, therefore, be said to pattern
together. The instructed condition is significantly more accurate
than the rule-search group. There is also a trend supporting the
superior speed of the rule-search condition relative to other
conditions. The planned comparisons show an almost significant
effect for a difference between the rule-search and instructed
conditions (p = .0 1 75). The fast performance of those in the rule­
search condition may reflect the fact that they had learned little
during training and consequently that the guesses made during
transfer are faster than the more calculated decisions of the
instructed condition. Both the greater accuracy of the instructed
condition and the greater speed of the rule-search condition may
also reflect transfer of specific aspects of task demands made during
training. Task demands in the rule-search condition are more similar
to the demands of the transfer task than are the demands of the
other conditions, since in the rule-search condition subjects are
required to search sentence strings simply to identify grammatical
rules, without attending to meaning, or memorizing the strings, or
remembering previously taught rules. Both attending to meaning
and memorizing, however, would seem likely to slow reaction time
if these processing modes are carried over into the transfer task.
Similarly, transfer of the processing mode adopted during the
application of rules in the instructed condition would seem likely t o
lead to slower processing o f the transfer strings, relative to a rule­
search condition, since this would involve recall of the full details of
the pedagogic rule presented during training. It appears to be true
that the rules taught in the instructed condition are successfully
applied to the task of discriminating grammatical from
ungrammatical strings. However, there is a significant interaction o f
Grammaticality, Rule and accuracy fo r the instructed condition.
Subjects in the instructed condition are significantly more accurate
in judging grammatical strings based on the easy rule than are
subjects trained under other conditions, but significantly worse at
judging ungrammatical strings based on the hard rule. This result 1s
further examined in section 6 . 1 .2 and Chapter 7.
Analyses and results 133
These results, therefore, lend only partial support to Hypothesis
1 . Whereas implicit and incidental conditions produce similar levels
of accuracy and reaction time across all sentences in the transfer
session, there is a significant difference in accuracy of performance
between those trained under rule-search and instructed conditions.
6.1.2 Analyses for Hypothesis 2
Whereas the analyses for Hypothesis 1 were based on accuracy and
reaction time to all sentences in the training set, the analyses
performed to address Hypothesis 2 specifically distinguish
performance on easy and hard rules. The second hypothesis states
that the implicit and incidental will be clearly superior to the rule­
search and instructed conditions in learning complex rules. T o
address this hypothesis the factor Rule was removed from the
repeated measure ANOVA for accuracy and reaction time and
separate repeated measure ANOVA were performed for accuracy
and reaction time for hard rule sentences, and for easy rule
sentences.
6.1.2.1 Complex rule accuracy
Results of the repeated measure ANOVA for accuracy on hard rule
sentences shows no significant main effect for Condition, but a
significant main effect for Grammaticality (F ( 1 / 1 00) = 1 1 .3 08 ,
p<.0 1 ) (see Table 9 ) . The relationship o f hard rule accuracy t o
Grammaticality across conditions is illustrated in Figure 1 6. The
interaction between Grammaticality and Condition is almost
significance (F (3/1 00) = 3 .722, p = . 0 1 39). All subjects are more
accurate in judging grammatical sentences than they are in judging
ungrammatical sentences, with this difference being most
pronounced in the case of instructed subjects.
The planned comparisons show no significant difference between
the implicit and incidental conditions in accuracy of performance
on the hard rule. However, there is a significant difference between
the rule-search and instructed conditions (F ( 1 /50) = 8.5 1 7, p<.O 1 ).
There is no significant difference between the combined implicit
and incidental versus the combined rule-search and instructed
conditions. Results of Scheffe post-hoe comparisons of accuracy o n
the hard rule across all conditions show significant differences at the
p<. 0 1 level only between the instructed and rule-search conditions.
1 34 Analyses and results Analyses and results 135

Table 9. ANOVA Table For Accuracy T o Hard Rule S e ntences

0 Grammatical
In The Transfer Set
• Ungrammatical
.85 Source df Sum of Mean F-Value P-Value
Squares Square
.8 c 3 2.98 1 .994 3 .255 .0248
G'
C<! Subject(Group) 1 00 3 0 .527 .305
8
(,)
.75 G 1 8.633 8.633 1 1 .308 .OO l l
<!'. G * C 3 8.525 2 . 842 3 .722 .0 1 3 9
.7 G * Subject(Group) 100 76.342 .763
�
'""O .65
s 9 22 . 1 42 2.460 1 3 .02 1 .000 1
!il S * C 27 5.308 . 1 97 1 .040 .4087
::c S * Subject(Group) 900 1 70.050 . 1 89
<+-< .6
0 G * S 9 8.7 1 3 .968 5 .729 .000 1
"'
ta .55 G * S * C 27 9.706 .359 2. 1 27 .0008
tl)
G * S * Subject�GrouE} 900 1 52.0 8 1 . 1 69
�
.5 Dependent: Hard Rule Accuracy
�
u
.45 Key: C = Condition (Implicit, Incidental, Rule-search, Instructed)
Implicit Incidental Rule-search Instructed
G = Grammaticality (Grammatical, Ungrammatical)
S = Sentence token
condition

Table 10. ANOVA Table For Accuracy To Easy Rule

Figure 16. Mean Accuracy For Performance On
Sentences In The Transfer Set
Grammatical And Ungrammatical Hard Rule Sentences
Source df Sum of Mean F-Value P-Value
Squares Square
6.1.2.2 Simple rule accuracy
c 3 9 .928 3 .309 7.494 .000 1
Results of the repeated measure ANOVA for easy rule accuracy are Subject(Group) 100 44. 1 63 .442
given in Table 1 0 . There is a significant main effect for Condition G 1 .024 .024 .048 .8276
(F (3/ 1 00) = 7.494, p<. 0 1 ) . Mean accuracy on grammatical and G * C 3 2.209 .736 1 .490 .22 1 8
ungrammatical sentences conforming to the easy rule is highest for G * Subject(Group) 100 49.4 1 7 .494
those in the instructed condition (see Figure 1 7). There is no s 9 1 2.366 1 .3 74 9. 1 72 .000 1
significant main effect for Grammaticality, and no significant S * C 27 5 .067 . 1 88 1 .253 . 1 76 1
interaction of Grammaticality and Condition. Planned comparisons S * Subject(Group) 900 1 34. 8 1 7 . 1 50
show no significant difference between implicit and incidental G * S 9 14. 193 1 .577 1 0.980 .000 1
conditions, but a significant difference between the rule-search and G * S * C 27 6.40 1 .237 1 .65 1 .0202
instructed conditions (F ( 1/50) = 1 3 .246, p<. 0 1 ). There are also G * S * Subject�GrouE2 900 1 29.256 . 1 44
significant differences between the combined implicit and incidental Dependent: Easy Rule Accuracy
conditions versus the combined rule-search and instructed conditions
(F ( 1 / 1 00) = 7 . 1 42, p<.0 1 ), though this is attributable to the
Key: C = Condition (Implicit, Incidental, Rule-search, Instructed)
G = Grammaticality (Grammatical, Ungrammatical)
contribution made by instructed subjects to the total score for the S = Sentence token
136 Analyses and results Analyses and results 137

Table 1 1 . ANOVA Table For Reaction Time To Hard Rule 0 Grammatical

Sentences In The Transfer Set
• Ungrammatical
Source df Sum of Mean F-Value P-Value
Squares Square

f
c 3 1 . 1 56E9 3 .854E8 2.727 .04 8 1 .9
Subject(Group) 100 l .4 1 3E 1 0 l .4 1 3E8
G 1 1240067 1 5.569 l .24E8 5 .373 .0225 � .85
G * C 3 145853 1 78.527 4. 862E7 2. 106 . 1 042 <l)

G * Subject(Group) 100 2.308E9 2.308E7 2 .8

s 9 2.685E9 2.983E8 1 5 . 6 1 7 .000 1 i;;>
S * C 27 720750958.33 1 2.669E7 1 .398 .0865 � .75
......
S * Subject(Group) 900 l .7 1 9E 1 0 l .9 1E7 0

G * S 9 729766843.796 8. 1 09E7 3 .983 .000 1 fj) .7

G * S * C 27 403005996.03 1 l .493E7 .733 . 8369
§
<l)

G * S * Subject�Group) 900 l .832E 1 0 2.03 6E7 :E .65

Dependent: Hard Rule Reaction time

Key: C = Condition (Implicit, Incidental, Rule-search, Instructed) Implicit Incidental Rule-search Instructed
G = Grammaticality (Grammatical, Ungrammatical)
S Sentence token
condition
=

Table 12. ANOVA Table For Reaction Time To Easy Rule Figure 17. Mean Accuracy For Performance On
Sentences In The Transfer Set Grammatical And Ungrammatical Easy Rule Sentences

Source df
Sum of Mean F-Value P-Value
Squares Square combined explicit conditions. The Scheffe post-hoe analyses show
c 3 l .804E9 6.01 3E8 3 .508 .01 8 1 significantly greater accuracy on easy rules in the instructed
Subject(Group) 1 00 l .7 1 4E l 0 l .7 14E8 condition compared with the implicit and rule-search conditions and
G 1 1 1 53 149.33 1 1 53 1 49.33 .044 .8337 an almost significant difference between instructed and incidental
G * C 3 65 1 25745.06 2 . 1 7 1 E7 .834 .478 1
conditions (p = .02)
G * Subject(Group) 1 00 2.602E9 2.602E7
s 9 4.46E9 4.956E8 1 8 .52 1 .000 1
S * C 27 920876267.86 3 .4 1 1 E7 1 .275 . 1 592 6.1.2.3 Complex rule reaction time
S * Subject(Group) 900 2.408E l 0 2.676E7 The results of the repeated measure ANOVA for hard rule reaction
G * S 9 4.8 1 5E9 5 .35E8 2 1 . 1 2 1 .000 1 time are given in Table 1 1 . The main effect for Condition
G * S * C 27 6245 1 6836.44 2.3 1 3E7 .9 1 3 .5936 approaches significance (F (3/1 00) = 3 . 508, p = .0 1 8 1 ). There is n o
G * S * Subject�GrouE2 900 2.28E10 2.533E7 significant main effect for Grammaticality and no significant
Dependent: Easy Rule Reaction time interaction of Grammaticality and Condition (see Figure 1 8). The
planned comparisons reveal no significant difference in reaction
Key: C = Condition (Implicit, Incidental, Rule-search, Instructed) time to hard rule sentences between implicit and incidental
G = Grammaticality (Grammatical, Ungrammatical) conditions but an almost significant difference between the rule­
S = Sentence token search and instructed conditions (F ( 1 /50) = 5 .0 1 9, p = . 0273 ).
138 Analyses and results
Analyses and results 139

There is no significant difference between the combined implicit

and incidental versus the combined rule-search and instructed 0 Grammatical
conditions. Scheffe post-hoe comparisons also show no significant
• Ungrammatical
differences between the conditions.

6.1.2.4 Simple rule reaction time

The results of the repeated measure ANOVA for reaction time to § 9000
easy rule sentences are given in Table 1 2 . There is no significant
main effect for Condition, or for Grammaticality and no significant -�
interaction of Grammaticality and Condition. As illustrated in � 8500
0
Figure 1 9 the relative speed of conditions is similar to that for e 8000
performance on the complex rule, with faster resonses by rule­ �
�
search learners. As with the results for the complex rule, planned µ.)
comparisons and Scheffe post-hoe tests reveal no significant 'O 7500
differences in reaction time between the conditions. §
� 7000
0 Grammatical
• Ungrammatical
.§ 1 1 000 Implicit Incidental Rule-search Instructed

condition
.§
.....
1 0500
.....
()
o:s
0 1 0000 Figure 19. Mean Reaction Time For Grammatical And
i:c: Ungrammatical Easy Rule Sentences
0
;:;
.....
9500
--0
.....
o:s
::r: 9000 6.1.2.5 Summary of the evidence for Hypothesis 2
<.;..,
0
Hypothesis 2 states that the implicit and incidental conditions will
{/) 8500 be clearly superior to the explicit conditions in performance on
§
0 hard rules. The results of Hypothesis 1 support the claim for
�
-
8000 similarity between these conditions across all sentences, since there
Q) is no significant difference between them in accuracy and reaction
u 7500 time to transfer set sentences. The analyses presented above also
Implicit Incidental Rule-search Instructed support the hypothesis that the two conditions perform similarly
on easy and hard rules since there is no significant difference
condition between them in accuracy or reaction time on easy and hard rule
sentences. However, these conditions do not perform significantly
Figure 18. Mean Reaction Time For Grammatical And better than the explicit conditions on the hard rule, either in
Ungrammatical Hard Rule Sentences accuracy or reaction time. Hypothesis 2 is therefore not supported.
The analyses reported above also extend the finding that rule­
search and instructed conditions do not perform similarly, contrary
to Hypothesis 1 . The instructed condition is significantly more
1 40 Analyses and results
accurate than the rule-search condition in performance on the hard
rule and on the easy rule. In addition, the post-hoe comparisons
reported above also show a significant difference between the more
accurate instructed condition and the implicit condition, and an
almost significant difference between the instructed and incidental
conditions (p = .02).
6.1.3 Analyses for Hypothesis 3
Hypothesis 3 states that performance on easy rule sentences will be
faster and more accurate than performance on hard rule sentences.
To address this hypothesis the factor Condition was removed from
the repeated measure ANOVA for accuracy and reaction time and
separate ANOVA were performed for each condition. Significant
main effects for the factor Rule, when supported by evidence o f
higher mean accuracy and lower mean reaction time for the easy
rule as opposed to the hard rule, were interpreted as evidence
supporting the hypothesis for each particular condition.
6.1.3.1 Simple and complex rule accuracy
The results of the ANOVA for accuracy on easy and hard rules show
no significant main effect for the factor Rule at the p<.0 1 level fo r
subjects trained in the implicit condition (F ( 1/25) = 3 . 779, p = The awareness data is based on responses to a questionnaire given
.0632). For subjects trained in the incidental condition (F ( 1 /25) =
1 4 . 8 1 8, p<.0 1 ), the rule-search condition (F ( 1 /25) =23 . 73 6 ,
p<.O l ) and the instructed condition, however, the factor Rule is
significant. There is a significant main effect for Grammaticality (F
( 1 /25) = 32.529, p<.0 1 ), and a significant interaction between Rule
and Grammaticality (F ( 1 /25) = 1 0 .740, p<. 0 1 ) only for instructed
subjects, which, as Figure 1 7 shows is attibutable to their poor
performance on ungrammatical hard rule sentences, relative to
grammatical hard rule sentences.
6.1.3.2 Simple and complex rule reaction time
There is a significant difference between reaction times to sentences
conforming to easy and hard rules for subjects trained under the
implicit (F ( 1 /25) 3 0.254, p<. 0 1 ), incidental (F ( 1 /25) = 2 8 .696,
=
p<.0 1 ), rule-search (F ( 1/25) = 1 8 . 8 1 3 , p<.0 1 ) and instructed
conditions (F ( 1 /25) = 1 2 .489, p<. 0 1 ). There is no significant main
effect for Grammaticality for any condition, but there is a
significant interaction of Rule and Grammaticality for instructed
subjects, which as Figure 1 9 shows is the result of faster perfomance
Analyses and results 141
on easy rule grammatical sentences compared with ungrammatical
sentences (F ( 1 /25) = 7.7 1 4, p = . 0 1 02).
6.1.3.3 Summary of the evidence for Hypothesis 3
The analyses reported above support Hypothesis 3 . There is a
significant difference in accuracy of responses to sentences
conforming to easy and hard rules for subjects trained in the
incidental, rule-search and instructed conditions. In all cases
accuracy on the sentences conforming to the easy rule is higher
than accuracy on sentences conforming to the hard rule (see Figure
1 4). There is no significant difference, however, between accuracy
of judgment on the easy and hard rule for subjects trained under the
implicit condition, though, again, mean accuracy on the easy rule is
higher than mean accuracy on the hard rule and the difference
approaches significance (p = .06). Similarly the reaction time data
supports the hypothesized order of difficulty of sentences
conforming to these two rules. In all conditions reaction time t o
easy rule sentences is significantly faster than reaction time to hard
rule sentences (see Figure 1 5).
6.2 Rule awareness data
after the training and transfer sessions had been completed. Subjects
were asked to write responses in their L l . These were translated
into English by the researcher and native-speaker informants.
Answers to three of the questions; 'Did you notice any rules o f
grammar when you did the first two exercises o n the computer?',
' Were you looking for grammar rules when you did the first two
exercises on the computer?' , and 'Can you say what the rules are?' ,
were used to code subjects as aware at the levels of noticing, looking
for rules, and able to verbalize rules. Only answers to the third
question were used to address Hypothesis 4 (see Chapter 7 for
further analyses). Complete explanations of the rules were n o t
required. If subjects attempted to describe any elements o f the easy
and hard rules presented during training, either by a structural
description, or by presenting and commenting upon an example o f
an easy o r hard rule sentence, they were coded as able to verbalize.
The coding also included information about which rule was
verbalized. There are thus three categories of subjects who
verbalized the rules, those who verbalized only the easy rule, those
who verbalized only the hard rule, and those who verbalized both
142 Analyses and results Analyses and results 1 43

easy and hard rules. Frequencies for subjects who were coded as able 6.2.1 Analyses for Hypothesis 4
to verbalize, divided by condition, are given in Table 1 3 . Tables 1 4 The verbalizability data displayed above was used to address
and 1 5 give frequencies of subjects able to verbalize only one rule, Hypothesis 4, that subjects in all conditions would attest to being
and both rules . Table 1 6 gives frequencies for the total more aware of simple rules than of complex rules. Only subjects
verbalizations of easy and hard rules across conditions. who verbalized one rule were included in the analysis. Condition was
not entered as a factor since it is not integral to the hypothesis. A
one-way Chi-square was performed on the total frequencies
Table 13. Frequencies Of Subjects Able To Verbalize The displayed in Table 14. There is only a small difference between the
Rules groups verbalizing one rule, 9 verbalizing the easy rule and 1 1
verbalizing the hard rule. The results pf the Chi-square with Yates'
Implicit Incidental Rule-search Instructed
correction are not significant (X2 =.200, p>.0 1 ). Subjects in the
instructed condition who verbalized one rule, all reported the easy
didn't verbalize 20 19 19 12 70
did verbalize 6 7 7 14 34
rule. There is a trend in the reverse direction, towards reporting the
hard rule, for subjects in the other conditions. A Chi-square of the
26 26 26 26 Totals
frequency values for easy and hard rule verbalizability by those in
the combined implicit and incidental group, versus the combined
Table 14. Frequencies Of Subjects Verbalizing One Rule
rule-search and instructed group, with Yates' correction, was also
not significant (X2= 3 .3 6, p>.0 1 ). However, pairwise comparisons
of easy and hard rule verbalization for the instructed and implicit
Implicit Incidental Rule-search Instructed
learners (X2 = 1 1 .66), incidental learners (X2 = 8.67) and rule­
easy rule 1 1 1 6 9
search learners (X2 = 8.53) are all significant at the p<.0 1 level.
hard rule 5 4 2 0 11
6 5 3 6 Totals
6.2.1.1 Summary of the evidence for Hypothesis 4
There is no evidence to support Hypothesis 4. The number o f
subjects attempting to verbalize any o f the rules i s low, as Table 1 3
Table 1 5 . Freq uencies Of Subjects Verbalizing Both Rules shows. There is no significant difference between the numbers o f
subjects verbalizing the easy and the hard rules. There i s a trend,
Implicit Incidental Rule-search Instructed apparent from the raw frequencies displayed in Table 1 6, for
both rules 0 2 4 8 14 subjects in the instructed condition to report the easy rule, and for
Total subjects in the implicit and incidental conditions to report the hard
rule. The relationship of the verbalizability data to the other
questionnaire evidence of rule awareness and conscious rule-search is
Table 16. Frequency Of Verbalization Of Easy And Hard discussed in more detail in the following chapter.
Rules
6.3 Variance data
Implicit Incidental Rule-search Instructed Two types of variance data were used to address Hypothesis 5 ;
easy rule 1 3 5 14 23 variance in accuracy of performance across conditions in the
hard rule 5 6 6 8 25 transfer grammaticality judgment test, and variance across
6 9 11 22 Totals conditions in the time taken to make grammaticality judgments.
Descriptive statistics for this data are summarised in Table 1 7.
144 Analyses and results Analyses and results 145

Table 17. Descriptive Statistics For The Variance Data 6.3.J.2 Reaction time
The results of the paired F-tests for reaction time given in Table 1 9
Condition Accuracy Reaction Time also show no significant difference between the implicit and
Mean Variance SD Mean Variance SD incidental condition, but a significant difference between the rule­
Implicit .638 .23 1 .48 1 9052.633 3 6 8 78737.992 6072.787 search and instructed condition (F ( 1 039/1 039) = . 846, p<.0 1 ). I n
Incidental .655 .226 .476 9648 .442 3 4 1 30566.3 1 4 5842. 1 3 7 addition the variance o f performance i n the rule-search condition is
Rule-search .633 .233 .482 743 9 . 1 8 8 28055756.049 5296.768 significantly lower than the variance of performance in the implicit
Instructed .758 . 1 84 .429 9 1 99.482 3 3 1 8 1 1 42 1 .657 5760.332
condition (F ( 1 039/ 1 03 9) = 1 .3 1 4, p<.0 1 ), and the incidental
condition (F ( 1 039/ 1 03 9 = 1 .2 1 7, p<.0 1 ).
6.3.1 Analyses for Hypothesis 5
Hypothesis 5 states that variance in performance under implicit and Table 19. Results Of Paired F-Tests For Reaction Time
incidental conditions will be equivalent, and less than the variance in
performance under rule-search and explicit conditions, which wiH Comparison Variance ratio Num.df Den.df F-Value P-Value
also be equivalent. To address this hypothesis F-tests comparing the
variance of each experimental condition, were performed using Implicit v. Incidental 1 .0 8 1 1 03 9 1 03 9 1 .08 1 .2 122
measures of accuracy and reaction time to sentences presented in Implicit v. Rule-search 1 .3 1 4 1039 1 03 9 1 .3 1 4 .000 1
the transfer task. Implicit v. Instructed 1.1 1 1 1039 1 03 9 1.1 1 1 .0888
Incidental v. Rule-search 1 .2 1 7 1 03 9 1 03 9 1 .2 1 7 . 00 1 6
6.3.1.1 Accuracy Incidental v . Instructed 1 .209 1039 1 03 9 1 .029 .6496
The results of the paired F-tests for accuracy are given in Table 1 8 . Rule-search v. Instructed 0.846 1039 1 03 9 0 . 846 .0069
There is no significant difference between variance of performance
under the implicit and incidental conditions, though there is a
significant difference between the rule-search and instructed 6.3. 1.3 Summary of the evidence for Hypothesis 5
condition (F ( 1 039/ 1 039) = 1 .266, p<.O 1 ). In addition, variance in The results reported above do not support Hypothesis 5 . There is
accuracy in the instructed condition is significantly less than no significant difference between the variance of performance of
variance in accuracy in the implicit (F ( 1 039/ 1 039) = 1 .257, p<.0 1 ) those in the implicit and incidental conditions, there is a significant
and incidental conditions (F ( 1 039/1 039) = 1 .23 1 , p<. 0 1 ). difference between the rule-search and the instructed condition.
There is significantly less variance in reaction time under the rule­
search condition than under the instructed condition, and
Table 18. Results Of Paired F-Tests For Accuracy significantly less variance in accuracy under the instructed condition
than under the rule-search condition. The rule-search and instructed
Comparison Variance ratio Num.df Den.df F-Value P-Value conditions are therefore not equivalent. Further, it is the instructed
condition that has significantly lower variance in accuracy than all
Implicit v. Incidental 1 .021 1 03 9 1 03 9 1 .02 1 .7352 other conditions, and the rule-search condition that has
Implicit v. Rule-search 0 .993 1 03 9 1039 0 .993 .9 1 3 3 significantly lower variance in reaction time than all other
Implicit v. Instructed 1 .2 5 7 1 03 9 1039 1 .257 .0002
conditions. The implicit and incidental conditions do not exhibit
Incidental v. Rule-search 0.973 1 03 9 1 03 9 0.973 .6549
less variance in performance than the explicit conditions.
Incidental v. Instructed 1 .23 1 1 03 9 1 03 9 1 .2 3 1 .0008
Rule-search v. Instructed 1 . 266 1 03 9 1 03 9 1 .266 .000 1
6.4 Aptitude data
All subjects completed two subtests of the Modern Language
Aptitude Test (Carroll & Sapon 1 959), the words-in-sentences test
1 46 Analyses and results Analyses and results 1 47

of grammatical sensitivity and the paired-associates test o f
memory. The two subtest scores were combined to give a composite
measure of aptitude. Descriptive statistics for the aptitude data show
scores on the memory test are normally distributed. There is n o
excessive kurtosis (-.258) o r skewness (-.252). The total score
available on this test is twenty four. Scores are distributed from 2 t o
2 4 with a mean score o f 1 4 . Reliability o f the subtest is .7728,
(KR20). Scores on the grammatical sensitivity test are also
normally distributed, with no significant kurtosis (-. 762) or
skewness (. 1 49). The total score available on this test is forty five.
Scores are distributed from 9 to 39 with a mean score of 2 2 .
Reliability o f the subtest i s .8 1 3 1 , (KR20). Finally, scores o n the
composite aptitude test are normally distributed (kurtosis = - .63 8,
skewness = .006). The total score available on this test is ninety
three, based on a score out of forty five for grammatical sensitivity
and a score out of forty eight for the memory test. The memory
test contribution was obtained by doubling the scores obtained on
the subtest. Scores on the composite aptitude tests are distributed
from 20 to 80 with a mean of 5 1 . Reliability is .94 1 6, (KR20).
Three one-way ANOVA showed no significant main effect for
condition on memory scores, grammatical sensitivity scores, and
the combined aptitude scores, providing support for the assumption
of homogeneity of aptitude across conditions which it was hoped
would follow from the procedure of randomly assigning subjects.
It is also necessary to show that the aptitude measures obtained
in the present study capture more than just differences in
proficiency since these tests were administered in English, which is
not the native language of the subjects for the study. No Japanese
versions of MLAT subtests used, or of any other recognized
aptitude test, are currently publicly available (though see Murakami
1 974 who reports the results of an early attempt to develop a
Japanese aptitude test with similar subtests to the MLAT) and
MLAT has been shown to discriminate equally well within both
native-speaking English, and native-speaking Japanese populations
(Sawyer 1 993) . There are alternative aptitude tests to the MLAT,
but either they are less suitable for the subject population in the
present study (e.g., Pimsleur's Language Aptitude Battery which was
developed for use with school age adolescents, see Pimsleur 1 966)
or they have been privately developed for small scale studies (as in
Sasaki 1 993) and are less well developed than the MLAT (as is the
case with VORD, an aptitude test developed for use with military
administration time (as is the case with DLAB, see Parry & Child
1 990:53). Due to the heavy demands on their time made by the
pretesting and experimental treatment, no proficiency measures
were obtained from subjects in the present study which could have
been used to directly address the issue of the correlation of aptitude
scores and proficiency. However, three pieces of evidence support
the claim for the independence of aptitude and proficiency.
(i) Evidence from a comparable group of subjects. The
correlation of proficiency and scores on the aptitude tests was
examined for a group of subjects comparable in age, L l background,
and type and quality of previous formal instruction in English to the
subjects who took part in the experimental training sessions. This
was a group of 45 native speakers of Japanese, aged between 1 9 and
20 years, who were currently enrolled in English programs in two
Japanese universities, Tokai University in Tokyo and Tokai
University in Kyushu. This group were in Hawai' i attending a short­
term English language program at Tokai University of Honolulu.
Like the majority of young adult Japanese subjects who took part in
the present study they had all received six years of formal
instruction in English in the Japanese school system, and had
limited exposure to English in a naturalistic setting. These subjects
completed the two subtests of the MLAT, and a proficiency test,
the Michigan English Placement Test (MEPT). The MEPT
consists of four subtests of listening, reading, vocabulary and
grammatical knowledge. The correlations of these subtests and the
total MEPT scores, with the MLAT subtests and composite scores
are reported in Table 20.
Table 20. Correlations Of Aptitude And MEPT Proficiency
Subtest Scores For An Equivalent Group Of Subjects
MEPT/total Listening Grammar Vocabulary Reading
Aptitude .617* .443* .507* .526* .534*
Memory .382* .3 7 1 .256 .313 .365
Grammatical .652* .330 .617* .573 * .5 14*
Sensitivity
* p< . 0 1
=

personnel, see Parry & Child 1 990), or require excessively long

1 48 Analyses and results
As can be seen from Table 20 there is no significant correlation
of the memory subtest with any of the MEPT subtests, at the p<.O l
level, but significant correlations of the grammatical sensitivity
subtest with the grammar, vocabulary and reading subtests of the
MEPT. The combined aptitude score correlates significantly with
the total MEPT score, and all the subtests. These correlations,
although significant, are not high enough, however, to support the
claim that the aptitude test and the MEPT measure the same
abilities. There are no correlations in the high .80s or .90s which
would be necessary to support this claim (Hatch & Lazaraton
1 99 1 :44 1 ). The aptitude tests measure other abilities, apart from
those tapped by estimates of proficiency.
(ii) No significant correlation of aptitude subtest scores. As can
be seen from Table 20 the memory subtest clearly measures abilities
not captured in estimates of proficiency. Further evidence of the
independence of the aptitude measures from measures of
proficiency lies in the lack of significant correlation between the
subtest scores. Table 21 shows the correlation of memory and
grammatical sensitivity subtest scores for those subjects in the
present study, divided by condition. There are no significant
correlations at the p<.O l level for the twenty six subjects in each
condition, providing evidence that the subtest scores measure
different abilities, regardless of current level of proficiency.
Table 2 1 . Correlations Of Memory And Grammatical
Sensitivity S ubtest Scores For Subjects In Each Condition
Implicit Incidental Rule-search Instructed
.457 . 1 59 .335 .101
No correlations are significant at the p<.01 level.
(iii) No even distribution of correlations across conditions.
There is no pattern of correlations of aptitude scores with
performance on easy and hard rules across learning conditions as
shown in Table 22, and discussed in the following section. It appears
from this Table that the aptitude tests measure abilities that are
differentially important to learning under the different task
Analyses and results 1 49
conditions. If scores on the tests merely reflected proficiency
differences, then, following the procedure of random assignment t o
conditions, which ensures that subjects representing a similar range
of proficiencies are assigned to each condition, there should have
been similar correlations across conditions.
6.4.J Analyses for Hypothesis 6
Hypothesis 6 states that measures of individual differences in
language learning aptitude will be related to performance under the
rule-search and instructed conditions, but unrelated to performance
under the implicit and incidental conditions. To address this
hypothesis scores on the three aptitude measures were correlated
with accuracy and reaction time for easy and hard rules under all
conditions. Results of these correlations, corrected for attenuation,
are reported in Tables 22 and 23 .
6.4.1.1 Accuracy and aptitude
For the implicit condition there is a significant correlation of
grammatical sensitivity and accuracy on the easy (r .693, p<.0 1 )
=
and on the hard rule (r .746, p<.0 1 ). There is also a significant
=
correlation with the composite aptitude score for easy rule accuracy
(r .52 1 , p<.0 1 ) and hard rule accuracy (r .520, p<.0 1 ). Scores
= =
on the memory subtest do not correlate significantly with easy o r
hard rule accuracy. In contrast, fo r the incidental condition there is
no significant correlation of easy and hard rule accuracy with
performance on any of the aptitude measures at the p<.O 1 level.
For the rule-search condition there is a significant correlation of
aptitude and accuracy on the easy rule (r = .5 56, p<.0 1 ). The
correlation of memory and aptitude with hard rule accuracy is not
significant at the p<.0 1 level, but is significant at the p<.05 level.
There is no significant correlation of memory and performance on
the easy rule. Grammatical sensitivity correlates significantly with
performance on the easy rule (r .604, p<.0 1 ), but not the hard
=
rule. For the instructed condition, as predicted, grammatical
sensitivity correlates significantly with hard rule accuracy (r . 5 6 3 ,
=
p<.0 1 ), and easy rule accuracy (r .540, p<. 0 1 ) . The correlation o f
=
memory with performance o n the easy rule (r .489), and the hard
=
rule (r .46 1 ) is not significant at the p<.O l level, but is significant
=
at the p<.05 level. There is a significant correlation of easy (r =
. 626, p<.0 1 ) and hard rule performance (r = .620, p<.0 1 ) with the
combined score for aptitude.
150 Analyses and results Analyses and results 151

Table 22. Correlations Of Aptitude Subtests And Accuracy 6.4.1.2 Reaction time and aptitude
On Easy And Hard Rules Under Each Condition There are no significant correlations of reaction time and
performance under any condition, apart from rule-search
Condition/Rule Grammatical Sensitivity Memory Aptitude performance on the easy rule which correlates significantly with
grammatical sensitivity (r = -.502, p<.0 1 ). For the rule-search
Implicit condition the aptitude score correlates with easy rule reaction time
Easy rule .693 * * .299 . 52 1 * * at the p<.05 level of significance (r = - .4 78) .
Hard rule .746 * * .250 . 520* *
Incidental
6.4.1.3 Summary of the evidence for Hypothesis 6
Easy rule .353 .309 .389
Hard rule .285 . 1 36 .234
There is partial support for Hypothesis 6 from the results of
Rule-search correlations for those trained under the incidental and instructed
Easy rule .604 * * .4 1 9 .556** conditions. As Hypothesis 6 predicts, performance in the transfer
Hard rule .374 .5 1 1 * . 504* task is unrelated to performance on the language learning aptitude
Instructed tests for those subjects trained under the incidental condition. Also
Easy rule . 540 * * .489* .626 * * in line with this hypothesis is the finding that performance by those
Hard rule .563 * * .46 1 * .620 * * trained in the instructed condition correlates significantly with
aptitude and grammatical sensitivity. Correlations for memory and
** = p<.0 1 * = p<. 0 5 performance on both rules also approaches significance for the
instructed condition. However, performance on both easy and hard
rules by those trained under the implicit condition correlates with
scores on the grammatical sensitivity test. This is contrary to
Table 23. Correlations Of Aptitude Subtests And Reaction Hypothesis 6. Performance on the easy rule by those trained under
Time On Easy And Hard Rules Under Each Condition the rule-search condition correlates significantly with grammatical
sensitivity, but there is no significant correlation with performance
Condition/Rule Grammatical Sensitivity Memory Aptitude on the hard rule.

Implicit 6.5 Summary of research hypotheses and results

Easy rule -.385 -. 1 2 5 .016
Hard rule -. 1 1 9 . 1 24 -.282 • Hypothesis 1 is partially supported. There is no significant
Incidental difference between the implicit and incidental conditions either in
Easy rule -. 1 95 . 1 36 .006
accuracy or reaction time of responses to all sentences presented in
Hard rule -.278 .067 -.094
the transfer set, but there is a significant difference between the
Rule-search
Easy rule - . 502* * -.3 1 3 -.478*
rule-search and instructed conditions in accuracy of response. The
Hard rule -.327 -.269 -.358 instructed condition is more accurate than the rule-search condition,
Instructed and they do not therefore pattern together.
Easy rule -.200 - . 1 36 - .2 1 8
Hard rule -.289 -.048 -.203 • Hypothesis 2 is not supported. The implicit and incidental
conditions are not superior to the rule-search and instructed
** = p<.O l * = p<. 0 5 conditions in accuracy or reaction time to sentences conforming t o
the hard rule in the transfer set.
1 52 Analyses and results

idental, rule­
for those trained in the inc
Hypothesis 3 is supported ts accuracy is
ditions. For these subjec
•

search and instructed con ten ces than for Chapter 7

ponses to easy rule sen
significantly higher for res is no sig nifican t
ten ces . However, there
responses to hard rule sen d rul e sen tences
response to easy and har
difference in accuracy of tim e to easy Further Analyses
plicit condition. Reaction
for those trained in the im e to ha rd rule
tly faster than reaction tim
rule sentences is significan
ined in all conditions.
sentences for subjects tra
is no significan t difference
4 is not supported. There
7. 0 Introduction
.
• Hy po the sis

in the verbalizability of eas

y and hard rules. ; �
This chapter addresses in more0det 1· 1 th ee Issues a�ising from the
previously presented analyses I o ': �t extent Is learning as a
.
5 is not supported . Variance
of performance by tho se fi
result of exposure under i� lic1t, 1�c1dental, rule-search and
• Hypothesis

trained in the im plicit and

incidental conditions is no
t significantly �.
instructed conditions different"a y sens 1ve to the grammatical and
lower than for those tra
ined in the rule-search
and instructed �;
ungrammatical status of sent� c s, an to t�e different easy and
_ .
CE�r 7;�
res of reaction
performance using measu hard rule sentence types rese du trammg? A pr�vious study
condition s. Variance in up tha n for all fu
for the rule-search gro of implicit L2 leam g is 3 ) , and studies of L 2
time is significantly lower for ma nce is
iance in accuracy of per
other conditions, and var tructed condition than for all oth er
ins
grammaticality judgment tests
as well as studies of implicit lea ��� �
hacht�r, Ty�on & Diffley 1 976),
_
smg art1fic1al grammars (e.g.,
significantly lower for the McAndrews & Moscovitch 1 985·; o ey & Brooks 1 992), have all
condition s. produced .evidence that sub�ects tend to wrong1y accept
Hypothesis 6 is partially
supported. There are sig
nificant e
ungrammatical stimuli more oft n t an th�y wrongly reject
and accuracy of perfor
ma nce in the gram�atical stimuli. Informatio : � out differences between
correlations of aptitude
•

instructed conditions, bu
t no t in the �:
cond1t10ns in responses to g mm�t1cal versus ungrammatical
implicit , rule-search and acc ura cy and
the implicit condition sentences, and in responses to fi e different types motivating the
incidental condition. In d. Ac cur acy of
significantly correlate design of the training and tr�ns �r . sets, thus �x�ends the analysis o f
grammatical sensitivity are nd itio n sho ws the l 1 1t nd exphc1t . con�itions beyond
ined in the instructed co what can be learned under
u::J:d � ; �
performance by those tra under the rule­
aptitude. For those trained the distinction between easy a rul s exammed m the previous
strongest correlation with ate s significantly ��
atical sensitivity correl chapter. (ii) How does transfer t s per ormance on sentences o f
search condition gramm �� 1- ied as ungrai:iimatical during the
time on the easy rule. the type subjects wron l 1
with accuracy and reaction pretest relate to perfor �:U �
c n s�ntences I� the transfer task?
This question is important to addr : _
given the h�1ted pretesting of
::
grammatical knowledge in the pr ent study. (111) Is awareness of
r !
rules, and the ability to verba 1ze hem " related to the measures o f
aptitude obtained from subJ1iects m this study? Th"is question · ·
1s
��::
• • • . . ·

important smce it may be th t ness is tr1g?ered either by an

ability to detect similaritie : . . en memonzed examples of
: �:�� �
sentences experienced durin t a n or by an ability to detect
he nd rlym .
grammatical relationships e u_ _ � ? struc�res o f
exa�p_les, or by a combination of the se ab1ht1es. Smee the implicit
cond1t1on requires memorizafton, one would expect to see a stronger
1 54 Further analyses
relationship between awareness and the memory component o f
aptitude for subjects reporting awareness in this condition. Since the
rule-search condition draws attention to the grammatical status of
examples, one would expect to see a stronger relationship between
grammatical sensitivity and awareness in this condition. Of course
this may not be so, and an examination of the relationship o f
aptitude to awareness may suggest that the processing strategies
leading to awareness are unrelated to those encouraged by the
different task demands.
These issues are addressed in the following order. Firstly, in
section 7 . 1 , the relationship of the factor of Grammaticality to the
different types of easy and hard rule sentences is assessed. Following
this, in section 7 .2, performance across conditions is compared with
respect to accuracy on sentences conforming to the specific types
used in the pretest of easy and hard rule awareness. Finally, in
section 7.3, the data on rule verbalizability, described in addressing
Hypothesis 4, are examined in relation to other measures of rule
awareness obtained from the debriefing questionnaire, and m
relation to the measures of aptitude used to address Hypothesis 6.
7.1 The effect of grammaticality and sentence type
Significant interactions of Rule and Grammaticality, and o f
Grammaticality and Condition, were revealed i n the analysis o f
accuracy o f response to all sentences i n section 6 . 1 . 1 (see Table 7 ) .
The analyses fo r Hypothesis 2 described i n 6. 1 .2 showed the
interaction to be attributable to performance on the hard rule,
(Table 9, Figure 1 6). There was no significant main effect for
Grammaticality in accuracy of performance on the easy rule (Table
1 0, Figure 1 7). The analyses for Hypothesis 3 showed the
interaction of Grammaticality and Condition resulted form the
performance of instructed learners who are significantly less
accurate at judging ungrammatical hard rule sentences than they are
at judging any other sentences.
Although there was no significant main effect for
Grammaticality on accuracy for implicit, incidental and rule-search
conditions there may, however, be significant effects across
conditions for responses to sentences grouped according to the type
categorization used to formulate the easy and hard rules (see 5 .3 .4
and 5.3.5 and Table 24), and significant interactions between Typ e
and Condition. I f the implicit and incidental conditions are
equivalent, as suggested by the results of analyses performed t o
address Hypothesis 1 , then they should be equivalent m
Further analyses 155
p�rf?�ance o n each sentence type. However, the Fundamental
S1m1�a�1ty Hy�othesis described in 4.3 .4, which holds learning in all
�on�1�10n,s . will be simi�ai: since it results from the variety of
not1cmg mduced by trammg task demands, suggests differences in
perfo �mance are likely since the different conditions direct
.
a�tentlon. at different aspects of the stimulus display, and this should
differ�ntial �y . affect the learning of types across conditions. T 0
examme this issue Type was entered as a factor in repeated measure
ANOVA for accuracy and reaction time to grammatical and
ungrammatical easy and hard rule sentences.
Table 24. Summary Of The Transfer Set Sentence Types
Easy Rule Grammatical Types
Type 1 (S V Location) e.g., Lisa fell onto the grass.
Type 2 (Location V S) e.g., Across the street raced Tom.
Type 3 (Location S V) e.g., Through the window Maria stared
Type 4 (S V Time) e.g., Peter woke at eight o 'clock.
Type 5 (Time S V) e.g. , On her birthday Gabi sang.
Easy Rule Ungrammatical Types
Type l(Time V Sino inversion without Location first) e.g., On Saturday
night danced Charlie.
Type 2 (V S Adv/no inversion without Location first) e.g., Golfed Debbie
in the morning.
Hard Rule Grammatical Types
Type l(S V Location and S V Location) e.g., The pen is in the box and
the
pencil is in the drawer.
Type 2 (S singular + be) e.g., Where the car is is in the driveway.
Type 3 (S plural + be) e.g., Where the apples are is in the bowl
Type 4 (S singular + lexical verb) e.g., Where Judy teaches is a; the
University.
Type 5 (S plural + lexical verb) e.g., Where Mary and Joe relax is by the
ool.
Hard Rule Ungrammatical Types
Type 1 (�o in� ersi�n with 'Where' ) e.g., Where is Bill is in the country.
Type 2 (mvariant smgular 'be') e.g., Where Carl writes are at a desk.
Type 3 (requ�red tense agreement) e.g., Where the bird is was in the sky.
Type 4 (reqmred ' Where' ) e.g., What the King lives is in a palace.
Type 5 (required 'not' for contrasting Locations) e.g., Where Mark works is
in America in Japan.
1 56 Further analyses Further analyses 157

7.1.1 Simple rule grammatical sentence types
There are five easy rule grammatical sentence types. Two sentences
of each type were contained in the transfer set. These types are
summarized in Table 24 and the transfer set sentences are illustrated
in Appendix J. Means for accuracy and reaction time to easy rule
Using the Chi-square statistic to calculate the probability o f the
distribution of accurate and inaccurate responses to the ten easy rule
grammatical sentences for the 26 subjects in each condition gives a
ratio of 1 5 1 to 1 09 as significantly above chance at the p<. 0 1 level
of significance, (JO ( 1 5 1 ) -/E ( 1 30) = 2 1 *2 1 / 1 3 0 = 3 .3 9 *2 6 . 7).
=

grammatical sentence types across conditions are illustrated in
Figures 20 and 21 and results of the ANOVA are reported below.
7. 1. 1. 1 Accuracy
There are significant main effects for Condition (F (3/1 00)
This means those scoring over 58% accurate or below 42% accurate
are performing above chance. Responses to type 2 sentences are
!he�efore low enough to indicate that those in the implicit,
mc1dental and rule-search conditions are consistently applying
wrongly held beliefs about the ungrammatical status of the
=

sentences, rather than randomly responding to them. In contrast,

1 0.650, p<.O l ), and for Type (F (4/400) = 27.208, p<.O l ). There is
performance is significantly above chance for responses to all other
a significant interaction between Type and Condition (F ( 1 2/400) =

easy grammatical sentences for subjects in all conditions. Type 2

2.3 7 1 , p<.0 1 ), which is illustrated in Figure 20. This appears to be
attributable to the fact that subjects trained in the instructed
condition perform much more accurately on type 2 sentences than
subjects trained in the other conditions. Mean accuracy for these
sentences is 70% for those in the instructed condition, 3 6% for
those in the implicit and incidental conditions, and 28% for those in
the rule-search condition.
sentences, it should be noted, were used in the pretest, together with
type 2 hard sentences, as a basis for selecting subjects for the
present study. Implications of the failure of subjects in the implicit,
incidental and rule-search conditions to accept them as grammatical
are considered in 7.2.
Q)

1
..§ 10500
;;.... i:::
g :§ 1 0000
3
0
.9 0
cd
0 Q)
.....
9500
cd
.8 0 type 1 Oil
Oil 0 9000 0 type 1
0
-� • type 2 -�
.7 a 8500 • type 2
�
.....
Oil
.6
A type 3 §
.....
Oil
8000 6 type 3
e type 4 Gi' 7500
Gi' .5 cd • type 4
cd Q)
Q) <> type 5 """' 7000
"""' 0 <> type 5
0 .4 "'
"'
§ 6500
§ Q)
Q) .3 ::;E 6000
::;E
-
.2 v 5500
v u
u Implicit Incidental Rule-search Instructed
Implicit Incidental Rule-search Instructed

condition condition

Figure 20. Mean Accuracy For Easy Rule Grammatical Figure 2 1 . Mean Reaction Time For Easy Rule Grammatical
Sentence Types Across Conditions Sentence Types Across Conditions
158 Further analyses Further analyses 1 59

7. 1. 1.2 Reaction time 7.1.2.2 Reaction time

There is a significant main effect for Type (F (4/400)
p<.0 1 ), but no significant main effec� for Con?ition, and no
.
significant interaction of Type and Condition (see Figure 2 1 ).
6.048,
= There is no significant main effect for Condition, and no significant
interaction of Type and Condition. There is a significant main
effect for Type (F ( 111 00) 27.206, p<.0 1 ). As Figure 23 shows,
=

responses to easy ungrammatical type 2 sentences are slower across

7. 1.2 Simple rule ungrammatical sentence types
. all conditions than responses to easy ungrammatical type 1
There are two easy rule ungrammatical sentence types. Five sentences.
sentences of each type were contained in the transfer set. These
types are described in 5 .6.5 and summarized in Ta�le 24. T � e
(])

transfer set sentences exemplifying these types are illustrated m ·� 1 0500

i::

Appendix J. Results of the repeated measure A�OVA for accuracy .g 1 0000

and reaction time to easy rule grammatical sentences are g
(])
.... 9500
summarized below. �
u
9000
·�
7. 1.2.1 Accuracy a 8500
0
There are no significant main effects for Condition, or Type, and §
.... 8000
type 1

no significant interaction of Type and Conditi?n . . Respons(ls t o •

Oil
c type 2

both easy rule ungrammatical sentence -�pes are si�mficantly a�ove

::l 7500
�
chance for subjects trained in all conditions (see Figure 22), with a <>I
(]) 7000
higher mean accuracy on type 2 for implicit, incidental and rule­
<.+...
0
"' 6500
search learners, and on type 1 for instructed learners. fa
(]) 6000
�
G' 5500
Q)
� . 875 ....-�-1..����l--��'--�1 u
Implicit Incidental Rule-search Instructed
g .85
ta condition
.8 .825
ti!
s .8

� .775
Oil
0 type 1 Figure 23. Mean Reaction Time For Easy Rule
§ .75 • type 2 Ungrammatical Sentence Types Across Conditions
_..,
� .725
(])
'O .7
Type 2 sentences (see Table 24), are those in which subject-verb
§
(])
.675
inversion occurs preceding adverbials of time and location (Crawled
� .65 Mick through the tunnel). It appears that these sentences pose
� .625 ..L�-.-����.-���.-����r-�..... more of a problem, as indexed by the lengthy reaction times they
Implicit Incidental Rule-search Instructed require, than type 1 ungrammatical easy rule sentences. Type 1
condition
sentences are those in which subject-verb inversion occurs following
a time phrase (On Friday night danced Charlie). Subjects correctly
Figure 22. Mean Accuracy For Easy Rule Ungrammatical classify these as ungrammatical, but have not learned the fact that
Sentence Types Across Conditions subject-verb inversion is allowed following adverbials of location, as
in type 2 grammatical easy rule sentences. As Figure 20 shows, type
1 60 Further analyses Further analyses 161

2 grammatical easy rule sentences (Into the house ran John) are
consistently wrongly rejected as ungrammatical by subjects in the
implicit, incidental and rule-search conditions where a rule of 'reject
subject-verb inversion' seems to be being applied. Subjects in the
instructed condition, however, have learned the exception to this
rule and correctly accept type 2 easy rule grammatical sentences
over 70% of the time.
7.1.3 Complex rule grammatical sentence types
There are five hard rule grammatical sentence types, as summarized
in Table 24 and illustrated in Appendix J. Mean accuracy and
reaction time to each type are illustrated in Figures 24 and 25 .
7.1.3.1 Accuracy
There is a significant main effect for Condition (F (3/1 00) = 6 . 667,
p<. 0 1 ), and a significant interaction of Type and Condition (F
( 1 2/400) = 2 .3 8 1 , p<.0 1), but no significant main effect for Type.
�ule-search condition; indicating chance performance. Performance
IS at chance on type 3 sentences for the incidental condition, and
on ty�e 4 s�ntences for the implicit condition. Performance by
those m the mstructed condition is significantly above chance for
all sentenc� types. Performance on type, 2 grammatical hard rule
�en�en�es IS below 42% for those in the rule-search condition,
mdicatmg wrongly held beliefs about the grammatical status of these
sentences. These sei:itences are the same type as the exponents o f
the . hard rule used m the pretest o f grammatical knowledge for
_
subjects m the present study (Where the book is is on the desk not
on the �oor). Rule-search subjects have maintained their wrongly
held . beliefs about the ungrammatical status of these sentences
?esp �t� e�po_sure to them during training. In contrast, subjects in the
Imphcit, mcidental and instructed conditions correctly classify them
as grammatical at above chance levels.

;>-.
g
..... .9
g .85
g
.8 0 type 1
«!
u 0 type 1
•t; .75 • type 2
a .7 • type 2
ll. type 3.
�
..... .65 A type 3
Oil • type 4
"O .6
a e type 4 <> type 5
..<::: .55
<.+...;
0 <> type 5
"'
.5
a
(!)
.45
� .4
Implicit Incidental Rule-search Instructed
Q) . 3 5
u condition
Implicit Incidental Rule-search Instructed

Figure 25. Mean Reaction Time For Hard Rule Grammatical

condition
Sentence Types Across Conditions
Figure 24. Mean Accuracy For Hard Rule Grammatical
Sentence Types Across Conditions
7. 1.3.2 Reaction time
There is a significant main effect for Type (F (4/400) 2 2 . 3 67,
As can be seen from Figure 24, performance on two of the sentence
=

p<.0 1 ), and an almosfs ignificant effect for Condition (F (3/1 00) =

types (types 3 and 5) is between 5 8% and 42% for those in the
1 62 Further analyses
Further analyses 1 63

3 . 675, p = . 0 1 47). There is no significant interaction of Type and

trained in the incidental condition perform at chance on types 1 , 4 ,
Condition. Figure 25 illustrates the mean reaction time to hard rule
and 2 , and at a level of 3 8% accuracy on type 3 sentences, which
arammatical sentence types across conditions. Mean reaction time
violate the tense agreement condition. Subjects in the implicit
fs faster for type 4 grammatical hard rule sentences for those in the
condition, in contrast, perform at above chance levels on four of
incidental, rule-search and instructed conditions (e.g., Where the
the hard ungrammatical rule sentence types, and at a chance level
President lives is in Washington not in Chicago), and longer for the
only on type 4 sentences. Subjects in all conditions correctly
type 2 sentences discussed above, for subjects trained in these three
classify type 5 sentences (Where Mark works is in America in
conditions.
Japan), which violates the condition that adjoined location phrases
must be contrasted by 'not' , as ungrammatical at significantly above
7.1.4 Complex rule ungrammatical sentence types
. . chance levels.
There are five hard rule ungrammatical sentence types, descnbed m
5 .3 .5, summarized in Table 24, and illustrated in Appendix J. The
results of the repeated measure ANOVA for accuracy and reaction .8
G'
time are described below. c<l
3
{) .7
�
7.1.4.1 Accuracy (;!
There is no significant main effect for Condition, but a s� gn�ficant �
{)
.6 O type 1
main effect for Type (F (4/400) = 1 0. 1 0 1 , p<. 0 1 ), and a s1gmficant 8
• type 2
interaction of Type and Condition (F ( 12/400) = 2 .489, p<.0 1 ). As
discussed in the analyses performed to address Hypotheses 1 , 2 . and '
;:s
.5
b. type 3
3 there is an interaction of Grammaticality and Rule for subjects -0
• type 4
a
t;ained in the instructed condition. Performance on ungrammatical ..i::
.4
4-< 0 type 5
hard rule sentences is significantly worse for instructed subjects than "'
0

performance on ungrammatical easy rule sentences. Figure 26 shows §

Q)
.3

this to be largely due to performance on two types of �

ungrammatical hard rule sentences. Inaccurate responses to . type 2 Q) .2
. u
and type 4 ungrammatical hard rule sentences are s1gmficantly
_
above the level attributable to chance. Instructed subjects correctly Implicit Incidental Rule-search Instructed

classify only 23% of type 4 sentences as ungrammatical (What t� e condition

King lives is in a palace not in an apartment), and correctly classify
only 34% of type 2 sentences as ungrammatical (Wher� the soldiers
fight are in Europe not in Hawai' i). These s�ntences v1ola�e two of
the conditions on the hard rule taught to mstructed subjects, the
condition that the form of the verb 'be' following the main verb is
invariantly singular (type 2) and the condition that pseudo-clefts of
location begin with 'where' , and not any other wh-word (type 4). In
addition, performance on type 3 ungrammatical hard ��ntences
(Where the bird is was in the sky), which vi� late the cond1t1on that
verbs must agree in tense, is at chance for mstructed subjec _ !s, who
correctly classify only 55% of these sentences as ungrammatical.
Subjects trained in the rule-search condition perform at ch �nce
on ungrammatical hard rule sentence types 1, 2, 3 and 4. Subjects
Figure 26. Mean Accuracy For Hard Rule Ungrammatical
Sentence Types Across Conditions
7.1.4.2 Reaction time
There is no significant main effect for Condition, and no significant
interaction of Type and Condition but a significant main effect for
Type (F (4/400) = 1 8.026, p<.0 1 ). Figure 27 shows that mean
responses to type 2 sentences are longer for all conditions, and
responses to type 1 sentences are shorter, with a mean difference of
over three seconds in response time across conditions between the
types. As with responses to easy rule ungrammatical sentences
 1 64 Further analyses
Further analyses 1 65

(Fi�ure 23) and to hard rule grammatical sentences (Figure 2 5 ) The effects of these different types on accuracy of performance in
subjects i n all conditions maintain similar rank orders o f reaction each condition, illustrated in Figures 2 1 , 25 and 27, is summarized in
time to sentence types. Table 25, which shows differences with respect to chance
performance on types across conditions. To examine whether there
v
s
was a significant difference in the frequency of accurat� and
·;:;
� 1 2000 inaccurate responses to each of the two sentences conformmg to
.9 easy grammatical types and hard grammatical and hard
t) .
ro
....
v 1 1 000 ungrammatical types by subjects in each condition, two way C� i­
�
(,)
squares were performed for accurate and inaccurate responses usmg
·:g 1 0000
0 type 1
McNemar's test for correlated frequencies. These show no
s
significant difference at the p<.0 1 level for fr�quencies of ac�urate
§.... 9000
• type 2
and inaccurate responses to sentences conformmg to all types m . the
!>I)
lJ,.
§ type 3 implicit, rule-search and instructed conditions, th�ugh there is a
"O
a 8000 • type 4 significant difference for responses to easy gr�mmatical type 1 , and
hard grammatical type 1 , and hard ungrammatica! types 2 and 5 for
..c
..... () type S -
"'
0 .
7000
those subjects trained in the incidental condit10n. Responses to
§
v these types have therefore been excluded from the summary of � he
::E
OJ
6000
relationship between type and chance perform��ce. An e'.'-planah? n
u of this finding for those in the incidental condition, consistent with
Implicit Incidental Rule-search Instructed . .
the Fundamental Similarity Hypothesis, is offered m section 8 . 3 .3
condition
of the following chapter. .
As Table 25 shows performance on type 2 easy grammatical
Figure 27. Mean Reaction Time For Hard Rule
sentences is significantly above chance for those in the instructed
Ungrammatical Sentence Types Across Conditions
condition who correctly accept them as grammatical, and for those
in the implicit, incidental and rule-search conditions who .wrongly
reject them as ungrammatical. This indicates s�bjects m these
7. 1.5 Summary of the effects ofgrammaticality and sentence type
conditions have maintained wrongly held beliefs about t? e
There is a consistently significant main effect for Type in measures
ungrammaticality of type 2 grammatical easy rule sentences despite
of reaction time to grammatical and ungrammatical sentences.
exposure to them during training. Table 25 also shows that
There is no significant interaction of Type and Condition for .
performance on hard gram.mati��l rule �e.ntence types differs across
reaction time, each condition tending to preserve the same rank
conditions. Those in the implicit condition perform at chance o n
order of response time to sentence types. There is no main effect
type 4 sentences. Subjects trained i n the rule-search condition are at
for Condition, or for Type, and no significant interaction between
chance on type 3 and 5 and wrongly accept type 2 sentences at
Type and . Condition for accuracy of response to easy rule .
significantly above chance levels, indicating wrongly held behefs
ungrammatical sentence types. Accuracy of response to each type is _
about their grammatical status. Accuracy is above chance . on these
above 5 8% in all conditions and therefore significantly above
types for those in the instructed condition. However, mstructed
chance. Differences between the conditions do emerge, though,
subjects are inaccurate, at above chance levels, on type 2 and type 4
from the accuracy data for responses to easy grammatical types,
hard ungrammatical sentences and perform at chance on type 3
and hard grammatical and ungrammatical types.
Ther� is a consistent main effect for Type, and a significant sentences. The effects of instruction on the hard rule do not appear
. to have contributed to an ability to correctly judge these three types
mteract10n of Type and Condition for easy rule grammatical
sentences, and hard rule grammatical and ungrammatical sentences. of sentence to be ungrammatical. Explanations of this finding are
offered in the following chapter.
 Further analyses 1 67
1 66 Further analyses

Table 25. (Continued) Percentage Correct Responses To Easy

Table 25. The Relationship Between Chance/Non-chance And Hard Rule Sentence Types Presented In The Transfer
Performance, Transfer Set Sentence Type And Condition Task

Sentence Type Chance/Non-chance Performance Sentence Type Chance/Non-chance performance

Easy Rule Grammatical Types Imp Inc Rule-s Inst Hard Rule Ungrammatical Types Imp Inc Rule-s Inst

Type 1 (S V Location) A x A A Type l(no inversion with Where) A c c A

e.g., Lisa/ell onto the grass. e.g., Where is Bill is in the country.
Type 2 (Location V S) I I A Type 2 (invariant singular 'be') A x c I
e.g., Across the street raced Tom. e.g., Where Carl writes are at a desk.
Type 3 (Location S V) A A A A Type 3 (required tense agreement) A c c
e.g., Through the window Maria stared e.g., Where the bird is was in the sky.
Type 4 (S V Time) A A A A Type 4 (required 'Where') c c c I
e.g., Peter woke at eight o 'clock e.g., What the King lives is in a palace.
Type 5 (Time S V) A A A A Type 5 (required 'not' for contrasting
e.g., On her birthday Gabi sang. locations) A x A A
e.g., Where Mark works is in America
Easy Rule Ungrammatical Types Imp Inc Rule-s Inst in Japan.

Type 1 (Time V Sino inversion Key: C = performance at chance

without location first) A A
e.g., On Saturday night danced Charlie.
Type 2 (V S Adv/no inversion
without location first) A A
e.g., Golfed Debbie in the morning.
Hard Rule Grammatical Types
Type 1( S V Location/S V Location)
e.g., The pen is in the box and the
pencil is in the drawer.
Type 2 (S singular + be)
e.g., Where the car is is in the driveway.
Type 3 (S plural + be)
e.g., Where the apples are is in the bowl.
Type 4 (S singular + lexical verb)
e.g., Where Judy teaches is at the
University.
Type
5 ( S plural + lexical verb)
e.g., Where Mary and Joe relax is by
the pool.
A A A = non-chance performance and accurate
I = non-chance performance, but inaccurate
X = excluded from this analysis
A A
Performance on hard ungrammatical sentence types is at chance
Imp Inc Rule-s Inst
for types 1 , 2, 3 and 4 in the rule-search condition, and at chance
for types 1 , and 4 in the incidental condition, and inaccurate at
A x A A
above chance levels for performance on type 3 sentences, again
indicating wrongly held beliefs about the grammatical status of these
A A I A sentences. Subjects trained in the implicit condition are above
chance on all hard ungrammatical sentence types, with the
A A c A exception of performance on type 4 sentences, which are
problematic for subjects in all conditions. Interestingly, the
C A A A performance of those in the implicit condition, using the measures
of chance and non-chance performance to sentence types described
here, appears to support Reber's claims about the relationship
A c c A
between learning condition and rule complexity. Performance on
more hard rule sentence types is at above chance levels for those in
the implicit condition than for those in any other condition.
Performance by those in the instructed condition is at above chance
168 Further analyses Further analyses 1 69

levels for all easy rule sentence types, and is therefore better than
0 hard type
performance by those in the other conditions. Of course, these
patterns of results must be interpreted in the light of the c • easy type
significantly higher overall accuracy of performance of the �(.) .9
(.)
instructed condition on the easy rule, and their superior overall o:!
.8
<l)
mean accuracy on the hard rule, as reported earlier in the analyses
2(.)
for Hypotheses 1 , 2 and 3 . These patterns of results are discussed .7

further in Chapter 8. gCJ) .6

7.2 Simple and complex rule pretest structures
The analyses in section 7 . 1 revealed that subjects trained in the
implicit, incidental and rule-search conditions performed
inaccurately, at above chance levels, on type 2 grammatical easy
sentences. A judgment of ungrammatical to a sentence of this type,
together with a judgment of ungrammatical to a sentence
conforming to the type 2 hard grammatical type, was used to select
subjects for the present study following the procedure described in
section 5 .3 .2. More extensive pretesting of subjects' prior
knowledge of the structures conforming to the easy and hard rules
was avoided for reasons given in sections 1 .6 and 5 .3 .2 .
The relationship between performance o n these structures
across conditions is illustrated in Figure 28. In the case of both easy
and hard pretest structures performance by the instructed condition
shows the highest accuracy and performance by the rule-search
condition shows the lowest accuracy. For all conditions mean
accuracy to the easy pretest structure is lower than mean accuracy
for the hard pretest structure. It thus appears that the easy pretest
structure is harder than the hard pretest structure.
To examine the relationship between performance on easy and
hard pretest structures for subjects in each condition paired one-way
t-tests were performed. These show a significant difference between
performance on the easy and hard structures for subjects trained
under the implicit condition (df 5 1 , t == 3 .438, p<.0 1 ), and for
subjects trained under the incidental condition (df 5 1 , t = 4.009,
p<. 0 1 ) . There is no significant difference between performance on
the two types of sentence for subjects trained under the rule-search
condition (df 5 1 , t = 1 .287, p>. 0 1 ), or for subjects trained under the
instructed condition (d/ 5 1 , t = 1 .399, p>.O l ).
....
CJ)
.
<l)
.... . .5
<l)
....
0..
'-+-< .4
0
CJ)
.3
§
<l)
.2
:;E
� Implicit Incidental Rule-search Instructed
u
condition
Figure 28. Mean Accuracy On Transfer Set Sentences
Conforming To The Types Used In The Pretest For Subjects
In All Conditions
Like the findings reported in section 7 . 1 above, this result must
be interpreted in the context of the analyses performed to address
the main hypotheses in Chapter 6, particularly the finding that
performance on sentences conforming to the the easy rule is
significantly more accurate than performance on sentences
conforming to the the hard rule for the incidental, rule-search and
instructed conditions, and significantly faster for subjects in all
conditions. This result does demonstrate, though, that the maj ority
of subjects in the implicit, incidental and rule-search conditions
have not modified their beliefs about the ungrammatical status o f
easy grammatical type 2 sentences, and so, to this extent, have no t
learned an important condition on the easy rule. The accurate
performance of those in the implicit and incidental conditions o n
hard grammatical type 2 sentences does not imply that they have
learned the rules regulating these structures, as the overall results for
accuracy on the hard rule reported in Chapter 6 show. They do
suggest that the particular configuration of two forms of the verb
'be' in sentences of this type are noticeable, and therefore
memorable. It is probably this knowledge of allowable bigram
pamngs, and memory for previous examples that guides the
decisions about grammaticality of those in the implicit and
1 70 Further analyses
incidental condition. It is not clear why this should not also be so
for those in the rule-search condition, but it appears to be the case
that rule-search learners are relying on more than memory for
bigrams and previously encountered instances in making their
responses, and that the instruction to search for rules promotes a
processing strategy in which accumulated memories for co-occurring
words plays little part. This issue is further discussed in Chapter 8.
7. 3 Awareness and aptitude
Three of the questions in the debriefing questionnaire were used t o
code subjects as noticing rules, having looked for rules and as able t o
verbalize the rules. These questions were 'Did you notice any rules
of grammar when you did the first two exercises on the computer?' ;
'Were you looking for grammar rules when you did the first two
exercises on the computer?' and 'Can you say what the rules are?'
The data presented to address Hypothesis 4 in Chapter 6 consisted
only of responses to the question 'Can you say what the rules are?'
which was used to code subjects as able to verbalize the easy rule, the
hard rule, both rules, or not able to verbalize rules. In this section I
examine responses to the three questions in relation to differences
induced by performance under the different training conditions.
This is important to address since it provides information about
whether the treatments were actually delivered. Large numbers o f
subjects in the implicit and incidental conditions claiming to have
noticed and to have been looking for rules, for example, would
imply that these conditions induced a mode of processing similar t o
that encouraged by the task demands of the rule-search condition. I
also examine the relationship between responses to these questions
and measures of aptitude used to address Hypothesis 6 in an attempt
to clarify what abilities contribute to awareness at the levels of
noticing, rule-search and rule verbalizability under each condition.
7.3.l Measures of awareness and the effects of learning condition
To examine the relationship between the measures of awareness and
learning condition three two-way Chi-square analyses of frequencies
of responses to each of three questions described above were
performed, using Condition as a grouping factor.
7.3.1. 1 Noticing across conditions
The results of the two-way Chi-square for awareness across learning
conditions shows no significant difference between frequencies of
subjects reporting they had noticed rules in the four learning
Further analyses 171
conditions (X2 = 5 . 597, p = . 1 3 3 0). Frequencies of subjects
reporting they noticed grammatical rules is lowest for those in the
implicit condition, and highest for those in the instructed condition
(see Table 26).
Table 26. Observed Frequencies Of Subjects Reporting They
Noticed Rules Across Conditions
Implicit Incidental Rule-search Instructed
Noticed rules 11 17 16 19 63
Did not notice rules 15 9 10 7 41
26 26 26 26 Totals
Nonetheless, the fact that seven subjects in the instructed
condition report that they did not notice rules is unusual considering
they had been taught them. Of course, subjects must interpret the
question about noticing rules in the light of the particular training
condition they were assigned to, and it may be that these seven
subjects are reporting the fact that they experienced problems in
applying the rules they had been taught to examples during training,
rather than lack of awareness of rules.
7.3.1.2 Looking for rules across conditions
The results of the two-way Chi-square for rule-search across
conditions show a significant difference between the frequencies of
subjects reporting that they were looking fo r rules in the different
training conditions (X2 = 1 3 .325, p<.0 1 ). The observed frequencies
for the Chi-square analysis are reported in Table 27. The numbers
of subjects reporting that they were looking for rules is highest in
the rule-search and instructed conditions, and lowest in the implicit
condition. This is to be expected. However, as with the result
reported above it is unusual that six subjects in the rule-search
condition report themselves not to have been looking for rules,
since this was the requirement of their training condition. Again it is
possible that these subjects could have been reporting themselves t o
have been unsuccessful in finding the rules, rather than not looking
for them. It is also possible, of course, that these subjects were not
following training instructions, due to fatigue or boredom.
1 72 Further analyses Further analyses 1 73

Significant numbers of subjects in the incidental condition also While the ratio of subjects reporting noticing and reporting that
report themselves to have been looking for rules. The total o f they were looking for rules is almost exactly the same, (4 1 /63 and
fifteen i s slightly less than the total o f seventeen who reported 40/64 favoring the 'noticed rules' and ' looking for rules' groups),
themselves as having noticed rules. The significant effect for the the ratio of subjects able to verbalize rules to those unable to
Chi-square is largely due to the smaller numbers of subjects in the verbalize is much lower, a ratio of 3 4 to 70. As discussed in the
implicit condition who report themselves looking for rules (post­ analyses for Hypothesis 4, the number of subjects able to verbalize
hoc cell contribution 3 .26 and -3 .26), compared to the larger is largest in the instructed condition. Although there is no necessary
numbers in the explicit rule-search and instructed conditions (post­ equation between being taught rules, using them, and being able t o
hoc cell contributions 1 .86 and - 1 .86). Even so, the fact that nine verbalize them or recall them, it i s unusual that larger numbers o f
subjects in the implicit condition reported that they were looking subjects in the instructed condition did not attempt to verbalize
for rules raises questions about the validity of the claim that rules. The procedure for coding subjects as able to verbalize that was
learning under implicit conditions, at least as operationalized in this used in the present study did not require that complete summaries of
study, is an 'unconscious' process of induction that takes place rules be given, only that subjects were able to report some aspect o f
largely outside of awareness of the structure of the stimulus domain. the stimulus domain relevant to each o f the rules and make some
comment on it. Perhaps the low numbers of instructed subjects
attempting to verbalize is again attributable to fatigue, since the
Table 27. Observed Frequencies Of Subjects Reporting They debriefing questionnaire was administered at the end of a lengthy
Were Looking For Rules Across Conditions experimental session. The methodology of using written reports as
measures of awareness is also likely to be less sensitive than other
Implicit Incidental Rule-search Instructed procedures, like oral interview, or the requirement to give
Looked for rules 9 15 20 20 64 instructions to a yoked group of subjects, as used in Matthews et al.
Did not look for rules 17 11 6 6 40 ( 1 989).
26 26 26 26 Totals
7.3.2 Measures of awareness and aptitude
To assess the relationship between the awareness data reported
7.3.1.3 Verbalizability across conditions above and measures of language learning aptitude, unpaired t-tests of
The results of the two-way Chi-square for verbalizability across scores on the memory and grammatical sensitivity subtests by
learning conditions shows no significant difference between subjects in noticing/not noticing, looking for rules/not looking for
frequencies of subjects able to verbalize rules in the four learning rules, and able to verbalize/unable to verbalize groups, divided by
conditions (X2 = 7 . 1 66, p = .0668). The observed frequencies are condition, were performed. Results of the t-tests are reported below.
reported in Table 28.
7.3.2.1 Awareness and memory
There is no overall significant effect for differences in memory as
Table 2 8 . Observed Frequencies Of Subjects Who Were Able measured by the paired-associates aptitude task between those who
To Verbalize Rules Across Conditions noticed and did not notice rules. There is an effect approaching
significance for a relationship between memory and rule noticing
Implicit Incidental Rule-search Instructed for those trained in the instructed condition (df 24, t = -2.65 8, p =
Verbalized rules 6 7 7 14 34
.0 1 3 8). There is a mean memory score of 1 0 for those who noticed
Did not verbalize rules 20 19 19 12 70 rules and a mean score of 1 5 for those who did not notice.
26 26 26 26 Totals There is no overall significant effect for differences between
looking for rules/not looking for rules groups on this measure and
no significant effect for groups trained under any condition.There is
1 74 Further analyses
no overall significant effect for differences between groups who can
and cannot verbalize the rules on this measure and no significant
effect for groups trained under any condition.
7.3.2.2 Awareness and grammatical sensitivity
There is no significant overall difference in the grammatical
sensitivity of groups at the level of noticing but a significant effect
for differences between those who noticed rules and those who did
not in the rule-search condition (dj 24, t = -3 .453, p<.O 1 ) . Those
who noticed rules in the rule-search condition have a mean
grammatical sensitivity score of 25 compared to a mean of 1 6 for
those who did not notice rules.
There is no significant overall difference in the grammatical
sensitivity of those who looked for rules and those who did not, but
there is an an effect approaching significance for those in the
implicit condition (dj 24, t = -2.683 , p = .0 1 3 0). Subjects who
looked for rules in this condition have a mean grammatical
sensitivity score of 26 compared to a mean of 1 8 for subjects who
report themselves not to have looked for rules.
Finally, there is no significant overall difference in the
grammatical sensitivity scores of those who were able to verbalize
rules versus those who were not, but a significant effect for a
difference in grammatical sensitivity scores of those able to
verbalize and unable to verbalize who were trained in the implicit
condition (df 24, t = -3 . 1 73 , p<. 0 1 ). There is a mean grammatical
sensitivity score of 29 for the able to verbalize rules group versus 1 8
for the unable to verbalize rules group.
7.3.3 Summary ofthefindings for awareness
No significant difference was found between the frequencies of
subjects reporting themselves to have noticed rules in each
condition (see Table 26). To this extent, claims for a causal
difference between the learning conditions and the learning
processes facilitated in this study attributable to the role of
conscious awareness are questionable. This measure of awareness
-the self-reported act of noticing-does not neatly distinguish
those in the rule-search and instructed conditions who are
consciously aware of the rule-governed nature of the stimulus
domain to be learned, and who thus learn 'explicitly,' from those in
the implicit and incidental conditions who are not conscious of the
Further analyses 1 75
The large numbers of subjects in the incidental condition
reporting they had noticed rules leads to the conclusion that this is a
variety of conscious learning, comparable to that in the rule-search
and instructed conditions, and differs from them only in terms of
the nature of the interaction of awareness and noticing with the
specific characteristics of the different task demands. Similarly, the
1 1 subjects reporting they had noticed rules in the implicit
condition raises questions about the claim that the instruction to
attend and memorize induces a variety of learning different from
that occurring in the explicit conditions. This finding for the
implicit and incidental conditions probably reflects a similar
propensity for second language learners to consciously notice rules
of the second language in classrooms during supposedly meaning­
focused activities, and raises doubts over claims that meaning­
focused treatments of themselves necessarily preclude learners from
consciously attending to and noticing the form of input (see Beretta
1 989; Schmidt & Frota 1 986; Slimani 1 992 for discussion).
Nonetheless, a larger number of implicit learners claimed not to
have noticed rules, compared to learners in other conditions. The
poor performance of the implicit condition on overall measures of
accuracy reported in the analyses for Hypotheses 1 , 2 and 3 is likely
to be attributable to just this differential in awareness. Less learning
occurred because fewer numbers of subjects were attending to the
rule-governed nature of the stimulus domain.
There was a significant difference between the numbers of
subjects reporting they were looking for rules in each condition (see
Table 27). This is attributable to the contrast between the large
numbers looking for rules in the rule-search and instructed
conditions, compared to the smaller numbers in the implicit
condition, as the previously mentioned post-hoe cell contributions
for these cells illustrate. This result is to be expected, given the
nature of the differences in task instructions. Nonetheless, large
numbers of incidental and implicit subjects report themselves to be
engaged in active rule-search, again casting doubt on the assertion
that the different conditions induce different varieties of learning,
attributable to differences along the dimension of conscious
awareness of structure. There was no significant difference between
the conditions in terms of the ability to verbalize rules, though
twice as many subjects were able to verbalize rules in the instructed
condition compared to other conditions (see Table 28).
What cognitive abilities contribute to awareness, given the
demands of particular task conditions? There were differences
1 76 Further analyses
between learners who noticed and did not notice rules in the
instructed condition that approach significance for memory (p =
. O 1 3 ), but no significant effect for differences in grammatical
sensitivity on awareness for these subjects. It would have been
unusual if strong effects for a relationship between aptitude and
noticing rules had been found for subjects in the instructed
condition, given that the task demands explicitly made them aware
of the structure of the stimulus display. However, as discussed above,
the 7 subjects reporting themselves not to have noticed rules in this
condition may have been interpreting the question to mean, 'Could
you apply the rule to examples successfully?' or, 'Do you
understand the rules you were taught?' or simply, ' Do you
remember the rules?' In this light the role of memory for those in
the instructed condition could be explained, since subjects claiming
not to have noticed rules are those with poorer memory scores, and
consequently they may have been unable to remember and so apply
the rules. No effects for differences in memory ability contributing
to the ability to notice rules was found for subjects in the implicit
condition, despite the fact that this condition required
memorization. There was a significant effect for differences
between those who noticed and did not notice rules in the rule­
search condition with regard to differences in measures of
grammatical sensitivity, and this is to be expected given the nature
of the task demands which require rule-search subjects to find rules,
since subjects in this condition probably rely heavily on prior
training in making metalinguistic judgments, and the procedures o f
formal grammar learning which are likely to b e linked to the degree
of grammatical sensitivity to the structural characteristics of
language.
There were no significant differences in measures of aptitude in
groups that were divided by the factor of looking for rules.
However, there was an effect approaching significance for
differences in grammatical sensitivity for subjects in the implicit
condition (p = .0 1 3). Those subjects with higher grammatical
sensitivity were more likely to engage in active rule-search. Finally,
the implicit condition again shows the only significant effect for
grammatical sensitivity on the ability to verbalize rules (p<.01). I t
should be noted that these results for the importance o f
grammatical sensitivity to measures o f awareness for those in the
implicit condition support the previous results reported for
correlations between success in performance on the transfer tas.k
and scores on the grammatical sensitivity component of the
Further analyses 1 77
aptitude test reported in the analyses for Hypothesis 6 in Chapter 6.
It should also be noted that there are no correlations of measures
of aptitude with the reported awareness of subjects in the incidental
condition, a result which again supports the earlier reported findings
for no significant correlation between aptitude and successful
performance in the transfer task for this group. In contrast, for
subjects trained under the implicit condition, differences in
grammatical sensitivity are linked to degree of awareness along the
dimension of looking for rules and verbalizing rules, and also fo
successful performance on easy and hard rules. For both implicit and
incidental learners, then, the variety of learning induced by the
memorization or meaning-focused tasks demands does not appear to
affect their propensity to notice or consciously search for rules.
However, only for implicit learners is awareness related to aptitude,
as measured by the grammatical sensitivity subtest. Coupled with the
results of the earlier reported correlations of accuracy and aptitude
for implicit learners, this suggests that the high aptitude of a small
number of learners in this condition contributed to their ability t o
successfully notice and learn the easy and hard rules. I n contrast, the
processing demands of the incidental learning task disengaged
aptitude and awareness.
7.4 Summary offurther analyses and results
The conclusions to be drawn from the further analyses are as
follows:
• The grammaticality of sentence stimuli affects accuracy but not
reaction time. There are significant differences in accuracy of
response to sentence types, with all subjects performing more
poorly on hard rule sentence types than on easy rule sentence
types. Subjects in the implicit condition perform at above chance
levels of accuracy on more hard rule ungrammatical sentence types
than subjects in any other condition. Grammaticality affects
accuracy of response to the extent that subjects in the instructed
condition perform more poorly on hard rule ungrammatical
sentences than on any others.
• There is a significant difference in accuracy of response to those
sentence types conforming to the structures used in the pretest for
subjects in the implicit and incidental conditions, but not for
subjects in the rule-search or instructed conditions.
1 78 Further analyses
• There is no significant difference across conditions in measures
of noticing rules, or measures of the ability to verbalize rules but a
significant difference between groups in the extent to which they
were looking for rules, with more subjects in the rule-search and
instructed conditions looking for rules than subjects in the incidental
or implicit condition.
• Measures of aptitude do not differ significantly for those in the
incidental condition or in the instructed condition with regard to
groupings based on the awareness data. There is a significant
difference for those in the rule-search condition where subjects who
attest to having looked for rules have significantly higher
grammatical sensitivity scores than subjects who did not claim to be
looking for rules. There is a significant difference for those in the
implicit condition where subjects who are able to verbalize the rules
have significantly higher grammatical sensitivity scores than
subjects who cannot verbalize the rules.
Chapter 8
The Fundamental Similarity Of
Implicit And Explicit L2 Learning
8. 0 Introduction
In this chapter conclusions are drawn concerning the role o f
awareness and attention to form in the different experimental
conditions (section 8 . 1 ), and concerning the effects of learning
under these conditions at the levels of rule, grammaticality and
sentence type (section 8 .2). It is suggested that the learning effects
observed at these levels are a consequence of the conscious
processing strategies encouraged by the task demands made during
training in each condition, as predicted by the Fundamental
Similarity Hypothesis described in section 4.3 .4, and that such
effects do not constitute evidence for separate conscious and
nonconscious learning processes. In section 8.3, explanations of the
performance of subjects in each condition are proposed which relate
the processing demands of the different training conditions t o
differential performance on sentence types i n the transfer task.
8. 1 Awareness and task demands
The claim that implicit and incidental learning differ from learning
following the instruction to search for rules, or following explicit
instruction in the rules, along a dimension of awareness of the
structure of the stimulus domain, has not been shown to be true in
the present study. As the further analyses of responses to the
debriefing questionnaire reported in section 7 .3 showed, there was
no significant difference between the conditions in measures of
awareness of structure. In response to the question 'Did you notice
any rules of English grammar in the first two exercises you did o n
the computer?' large numbers o f subjects in all conditions replied
that they had. To this extent, none of the effects for learning in
any of the conditions can be attributed to a nonconscious process o f
abstraction. Neither was there a significant difference between the
groups in terms of their ability to verbalize rules, though
1 80 Fundamental similarity of implicit and explicit SLA
significantly greater numbers of subjects in the rule-search and
instructed conditions claimed to have been actively looking for rules
when compared to subjects in the implicit and incidental conditions,
a fact which is to be expected considering the task instructions in
the explicit conditions. These results, then, demonstrating
widespread awareness of structure across conditions, are support for
the Fundamental Similarity Hypothesis proposed in Chapter 4. This
holds that all L2 learning is the result of conscious awareness of
form at the point of input, which is explicitly attended following
the information processing operations described in Chapter 3 , and
that the effects of learning under implicit and explicit conditions
are artifacts of the variety of conscious processing encouraged by
different task demands.
8.1.1 Varieties of awareness under different learning conditions
To the extent that the task demands in the four experimental
conditions differ, so too does the variety of awareness of structure
that the task demands facilitate. Schmidt ( 1 990), Whittlesea and
Dorken ( 1 993), and Shanks and St. John ( 1994) have all argued that
task demands are powerful determinants of what gets noticed. The
Fundamental Similarity Hypothesis described in section 4.3 .4
claimed that different learning conditions are likely to lead to a
differential reliance on strategies for accumulating knowledge of the
stimulus domain, but do not lead to differential reliance o n
conscious and nonconscious information processing. Subjects in all
conditions are similar in this respect, since they adopt conscious
strategies for meeting the demands of the particular task conditions.
This often led, across conditions in the present study, to noticing of
rules of grammar underlying the stimulus domain. However, the
different task conditions do lead to different varieties of awareness
of structure. These varieties of awareness are largely a consequence
of the processing demands of particular task conditions. The
implicit condition, with its instruction to memorize, is likely to
facilitate the development of a knowledge base in which memory
for co-occurring words plays a large part, whereas the rule-search
and instructed conditions are likely to facilitate the development o f
a knowledge base in which the inductive search for, or deductive
application of abstract configurational relations between categories
plays a large part. However, each condition will encourage a blend
of these types of processing. The effects for learning under the
implicit condition, for example, will be largely attributable to the
process of noticing co-occurring words, though conscious rule-
Fundamental similarity of implicit and explicit SLA 181
search may in some cases play a part. The evidence from the
awareness data, for example, shows that some subjects in all
conditions attested to looking for rules. Similarly, although the rule­
search and instructed conditions predispose learners to search for, or
apply rules, they do not preclude the possibility that learners will
also gather information by memorizing examples. As the awareness
data showed, despite the fact that significantly greater numbers o f
subjects in the explicit conditions claimed to have been looking for
rules, there were a number of subjects in each of these conditions
who did not report that this is what they were doing.
The varieties of learning and awareness that take place under
each learning condition are also likely to be affected by individual
differences and propensities for processing language data, in addition
to particular task demands. The data examining the relationship o f
grammatical sensitivity and memory to awareness showed a
significant difference between the grammatical sensitivity scores o f
those implicit subjects who were able to verbalize rules and those
who could not, higher grammatical sensitivity scores being
associated with greater ability to verbalize the rules. The sanie
difference in grammatical sensitivity scores for the implicit group
approached significance for the distinction between those who were
looking for rules, and those who were not. Grammatical sensitivity
scores were also significantly different for those subjects in the rule­
search condition who claimed to have noticed rules, and those who
did not. No significant differences in aptitude scores were found for
any groupings by measures of awareness for those in the incidental
or instructed conditions, however. This is to be expected for the
instructed condition, since they had been explicitly made aware of,
and taught the rules in the training sessions. The results for the
incidental condition parallel the findings of no significant
correlation with aptitude on easy or hard rule accuracy in the
transfer set.
In the incidental condition, successful learning, and the abilities
leading to awareness or noticing of rules, are unrelated to those
abilities measured by aptitude tests. This is an interesting finding,
since it suggests that processing in the incidental condition draws on
different abilities than processing in the other conditions and
partially confirms Hypothesis 6, based on Reber's and Krashen's
claims, that learning under the implicit conditions will be insensitive
to individual differences in measures of relevant cognitive abilities
like aptitude. However, the finding that aptitude correlates with
successful learning of both rules in the implicit condition contradicts
1 82 Fundamental similarity of implicit and explicit SLA
this hypothesis (for further analyses of the awareness, aptitude and
learning interactions in the present study see Robinson 1 995b).
8.1.2 The sensitivity of measures of awareness
Reber's claim that implicit learning is a process of rule abstraction
that takes place outside of conscious awareness of the structure of
the stimulus domain is based on the evidence of failure to articulate
the rules following transfer test performance that shows sensitivity
to the structure of letter strings generated by artificial grammars.
Reber has been criticized on the grounds that the typical measures
of awareness used, oral reports, are not sensitive enough to the
contents of subjects' awareness (see Brody 1 989; Dulany, Carlson &
Dewey 1 984, 1 985; Perruchet & Gallego 1 993; Shanks & St. John
1 994). Shanks and St. John have also claimed that these measures
do not meet the 'information' criterion of asking subjects exactly
what they were aware of and had learned. Since it may not have
been rules, and may instead have been knowledge of bigrams or
chunks that they were aware of, such measures create the very
effect that they claim is causal-an inability to articulate the
knowledge base that guides decision making about grammaticality.
Similar criticisms could be made of the measures of awareness used
in the present study. Although many subjects claimed to have
noticed rules, few were able to articulate them, and consequently the
small numbers of subjects able to verbalize rules revealed n o
significant difference between conditions i n this respect. If,
however, subjects had been additionally asked which sentences, or
combinations of words they remembered, as well as any rules they
identified, larger numbers of subjects may have responded. The use
of debriefing recognition tasks in addition to recall tasks might have
produced more evidence of the precise components of rules and
structures that subjects were aware of in all conditions, since direct
recall of rules is a measure of awareness biased towards transfer of
the processing strategies encouraged at task in the rule-search and
instructed conditions. Additional use of recognition tasks t o
determine what subjects noticed would facilitate a similar transfer o f
the processing strategies encouraged at task in the implicit and
incidental conditions.
Nonetheless, despite these shortcomings, the measure of
awareness used for the present study suggests that large numbers of
subjects in all conditions were aware of rules in the sentences they
processed, whether they were doing this in order to memorize or in
Fundamental similarity of implicit and explicit SLA 1 83
order to understand meaning, as well as following instructions to
apply rules or instructions to rule-search. Just as in the implicit
condition, so in the incidental condition, which Krashen ( 1 982,
1 985, 1 994) has speculated creates the conditions for acquisition
processes to occur, there is no evidence for a neat dissociation of
unconscious learning and articulable knowledge. Rather, the process
of attending to meaning facilitated by the incidental condition was
often accompanied by an awareness of the grammatical structure of
what was read, and in a number of cases by the attempt to find
grammatical rules, which were sometimes articulable.
The readiness of large numbers of subjects in the implicit and
incidental conditions to attest to awareness of the rule-governed
nature of the stimuli in present study parallels, to an extent, the
findings from the experimental studies of artificial grammar
learning. In those studies subjects in implicit conditions do often
claim to be aware of having learned something, but they are
predominantly unable to articulate the rules themselves. Likewise,
few subjects in the implicit or incidental conditions of the present
study were able to verbalize any aspect of the rules regulating the
sentences viewed during training. There are, however, some rather
obvious differences between the subjects for the present study and
the subjects for experimental studies like those of Reber. The L2
learners in the present study already have a knowledge base to draw
on in performing the tasks presented during training and in transfer.
They are also experienced and motivated learners of the language
presented during training. Given this experience, motivation, and
background knowledge it seems likely to be the case that presenting
unfamiliar sentences and structures, in any of the conditions, will
trigger access to, and conscious reflection on, some previous
experience of language learning. Subjects for Reber's implicit
learning studies, on the other hand, have no previous experience of
artificial grammar learning, no motivation to do it, and no
background knowledge to invoke. With this in mind it is surprising
that larger numbers of subjects in the present study did not attempt
to verbalize the rules they claimed to have been aware of.
As with experimental research in artificial grammar learning, so
L2 research into implicit learning also needs to develop sensitive
measures of awareness that properly meet the information criterion
by not unnecessarily restricting the scope of possible answers t o
questions about the objects o f awareness during learning. The
information criterion will be important to meet in cases like the
present study where large numbers of subjects attest to awareness,
1 84 Fundamental similarity of implicit and explicit SLA
since it is important to be able to distinguish the objects of
attention and hypothesis testing during training from the other
contents of background knowledge and previous learning experience
that are invoked by association with the present task, but are not
intrinsic or relevant to it. This may have been the case in the
present study if subjects reported awareness on the basis of noticing
examples of third person singular morphemes or the occurrence o f
regular and irregular past tense markers-things which were n ot
critical to learning the easy and hard rules underlying the sentences
presented during training.
8.2 Learning at different levels of the dependent variable
The first two main hypotheses for the present study concerned the
differential effectiveness of the implicit and explicit learning
conditions in learning pedagogic rules that were judged to differ in
complexity, following the procedure described in section 2 .4. These
hypotheses were motivated by the claims of Reber ( 1 989, 1 99 3 )
and Krashen ( 1 982, 1 985) that learning as a consequence o f
exposure under implicit and incidental conditions would be superior
to learning under rule-search and instructed conditions where the
stimulus domain was complex. By replicating Reber's implicit and
explicit rule-search conditions and matching them with incidental
and instructed conditions, the study also examined the extent of the
generalizability of claims made by Reber, and results obtained using
his methodology, to conditions invoked by Krashen in discussion
about the role of consciousness in SLA.
There was no support for Hypothesis 2, which claimed that the
implicit conditions would be superior in measures of speed and
accuracy of response to structures conforming to the hard rule,
when compared to performance by subjects in the rule-search and
instructed conditions. There was only partial support for
Hypothesis 1 , which claimed implicit and incidental conditions
would pattern together in performance on all sentences, as would
rule-search and instructed subjects. Although the implicit and
incidental conditions performed similarly on both rules, there was a
significant difference in accuracy of performance on the hard rule
between instructed and rule-search subjects. Instructed subjects were
significantly more accurate than rule-search subjects. There was also
a significant difference in accuracy of response to easy rule
sentences for instructed subjects versus subjects trained in all other
conditions. Again, instructed subjects were significantly more
accurate, with a mean accuracy rate of over 84%. There are thus
Fundamental similarity of implicit and explicit SLA 1 85
clear advantages for instruction on easy rules relative to learning
under the other conditions operationalized in the present study, and
clear disadvantages for those attempting to consciously search for
hard rules, relative to those who have been instructed in them .
These findings contradict the hypotheses based on Reber and
Krashen, and also the more recent hypothesis of Hulstijn and
DeGraaff ( 1 994: 1 03 ) that "explicit instruction has more effect in
the case of complex rules than in the case of simple rules."
Easy rule sentences were clearly easier to learn than hard rule
sentences for all subjects using measures of reaction time, since they
elicited significantly faster responses than hard rule sentences in all
conditions. They were also easier for the incidental, rule-search and
instructed conditions using measures of accuracy, since these
conditions were significantly more accurate on easy rule sentences.
Therefore Hypothesis 3, that the distinction between easy and hard
rules would be reflected in more accurate, and faster responses t o
easy rule sentences, was largely supported. However, there was n o
significant difference i n accuracy of response to easy and hard
sentences for the implicit condition. Subjects in this condition,
despite having the lowest mean accuracy of response to the easy
rule, did perform above chance on both rules, and therefore showed
evidence of knowing something about the easy and hard rules.
However, they did not display the sensitivity to the complexity o f
the stimulus domain that learners · in the other conditions did.
The effects just described were the result of comparisons of
performance on all easy and hard rule sentences across all
conditions. In order to examine in more detail the differential
effects of learning under the four conditions, the accuracy and
reaction time data were further examined in Chapter 7. There the
effects of the factors of grammaticality and sentence type were
considered across conditions. The distinction, then, between the
analyses for Chapter 6, and the main hypotheses addressed there,
and the further analyses in Chapter 7, lies in the level of the
dependent variable at which the differential effects of learning under
implicit, incidental, rule-search and instructed conditions were
observed. These levels are distinguished below before the results of
the further analyses performed in Chapter 7 are summarized.
8.2. 1 The level of rules
The notion of rule operationalized in the present study was one
which grouped together sets of sentences with the intention, in the
instructed condition, of drawing attention to allowable and non-
1 86 Fundamental similarity of implicit and explicit SLA
allowable patterns of covariance in sentence structure. So, with the
easy rule, sentences that illustrated the possibilities of subject-verb
inversion following the fronting of adverbials of two different types
were grouped together. For the hard rule sentences were grouped
together to illustrate the different forms that pseudo-clefts of
location could take. At a level of abstraction from the facts of
individual sentences, then, knowing the rule requires access t o
general principles of configurationality, like word order, and notions
of agreement, tense and number, and an ability to identify parts of
structures that these principles relate to, and to make decisions
about whether they conform to the relations between the general
principles. Knowing the rule implies an ability to make judgments
based on these principles across the entire set of sentences
exemplifying the rule, regardless of whether they are novel or have
been previously experienced. The analyses for the main hypothesis
compared performance on all easy and hard rule sentences across
learning conditions, using measures of accuracy and reaction time,
and identified an advantage for the instructed condition over the
others in accuracy of performance on the easy rule, but no clear
advantage for any single condition in accuracy of performance o n
the hard rule, and n o advantage fo r any single condition o n either of
the rules with respect to reaction time. As stated above, these
findings clearly refute the hypothesized advantage for the implicit
and incidental conditions on hard rules, and demonstrate the
positive effects of instruction on learning easy rules.
The further analyses in Chapter 7, however, showed effects for
differences in the status of grammatical and ungrammatical
sentences, and for individual sentence types conforming to easy and
hard rule sentences. No group performed at above chance on all
sentence types, and to this extent no group could be said to have
learned both rules successfully. In fact, using the criterion of above
chance performance on all sentence types illustrating a rule, only
the instructed group learned a rule successfully, that rule being the
easy rule. Greater length of exposure may have increased the
amount of learning of the hard rule, but the amount of exposure in
the present study, two trials, or runs, of 40 sentences, was
comparable to that in the implicit learning studies summarized in
Table 3, section 4. 1 . 1 .
8.2.2 The level of grammaticality
All sentences presented during training were grammatical. The
training sessions were thus examples of learning from positive
Fundamental similarity of implicit and explicit SLA 1 87
evidence of what is possible in the language, without the benefit of
negative evidence of what is not possible. As in the artificial
grammar learning experiments by Reber ( 1 989) and others,
described in section 4 . 1 , learning was assessed in a transfer task
during which subjects had to classify equal numbers of grammatical
and ungrammatical sentences. There was a significant effect for the
interaction of Grammaticality and Rule for the instructed condition.
Instructed subjects performed much less accurately on hard rule
ungrammatical sentences than on other sentences. There was n o
such effect fo r Grammaticality for the instructed condition o n the
easy rule, or for any other condition on either rule.
8.2.3 The level of type
As the further analyses for Chapter 7 showed, there were differences
across all conditions with respect to performance on individual
types of sentences, using measures of chance and non-chance
performance. The level of sentence type was manipulated t o
conform to the particular conditions on rules that were explained in
the instructed condition. The significant interaction of Rule and
Gammaticality in measures of accuracy for the instructed condition
was clearly shown to relate to above chance inaccurate performance
on two ungrammatical hard rule sentence types, as illustrated in
Table 24 (section 7. 1 ). The instructed condition performed at
chance on a third ungrammatical hard rule sentence type. Rule­
search and incidental subjects performed at chance on four of the
hard rule ungrammatical sentence types. However, subjects in the
implicit condition performed above chance on four of the hard rule
ungrammaticartypes, and to this extent performed better than the
instructed, or any other condition on the hard rule. The analyses of
mean accuracy and reaction time across all easy and hard rule
sentence types did not reveal this effect, since there were n o
significant differences between the implicit, incidental or instructed
conditions in performance on hard rule sentence types, though
there was a significant difference between the superior instructed
condition and the rule-search condition.
In the sections below, the issue of performance on hard rule
ungrammatical sentence types is examined in relation to the
specific processing demands of the different learning conditions, and
the nature of the grammaticality violations contained in each type.
This is an important issue to address, since performance on hard
ungrammatical sentence types contradicts the effects shown for
instruction in performance on the easy rule sentence types, and o n
1 88 Fundamental similarity of implicit and explicit SLA Fundamental similarity of implicit and explicit SLA 1 89

hard grammatical sentence types. Explanations are also considered Table 29. Percentage Correct Responses To Easy And Hard
for why the instructed condition differed from the other conditions Rule Sentence Types Presented In The Transfer Task
in performance on type 2 easy grammatical types, since this effect
contributed in large part to the resulting significant advantage of the Sentence Type % Correct
instructed group in performance on the easy rule. Performance in
the implicit, incidental and rule-search conditions was above chance, Easy Rule Grammatical Types Imp Inc Rule-s Inst
and inaccurate in performance on this type, while performance m
the instructed condition was above chance and accurate. Type 1 (S V Location) 86.5 82.7 84.6 92.3
e.g., Lisa fell onto the grass.
8.3 Explanations of the effects of grammaticality and type on Type 2 (Location V S) 3 6 . 5 ? 36.5? 28.8? 78.8
learning e.g., Across the street raced Tom.
Explanations of results of implicit learning studies have either Type 3 (Location S V) 6 5 .4 73 . l 6 1 .5 96.2
e.g., Through the window Maria stared
argued that the results of such learning are evidence for abstraction
Type 4 (S V Time) 67.3 73 . l 90.4 88.5
of rule-based knowledge (Reber 1 989), or knowledge based o n
e.g., Peter woke at eight o 'clock.
memory fo r fragmentary instances encountered during training Type 5 (Time S V) 69.2 76.9 76.9 92.3
(Dulany et al. 1 984; Perruchet & Pacteau 1 990; Rothkopf & e.g., On her birthday Gabi sang.
Dashen 1 995; Vokey & Brooks 1 992). Some have argued that the
effects of implicit learning are attributable to a synergy of both Easy Rule Ungrammatical Types Imp Inc Rule-s Inst
forms of knowledge (Matthews et al. 1 989). This debate was
reviewed in Chapter 4. Like Shanks and St. John ( 1 994), Whittlesea Type 1 (Time V Sino inversion
and Dorken ( 1 993) and McAndrews and Moscovitch ( 1 985) I have without location first) 64 .6 73.8 67.7 84.6
argued there is no reason, a priori, to associate the products o f e.g., On Saturday night danced Charlie.
learning under any condition with a knowledge base composed o f Type 2 (V S Adv/no inversion
either entirely rule-based o r entirely memory-based knowledge and without location first) 70.8 78.5 73.8 74.6
that the processing demands of tasks are likely to encourage the e.g., Golfed Debbie in the morning.
development of a knowledge base composed of a blend of these two
sources of knowledge. However, task conditions will predispose Hard Rule Grammatical Types Imp Inc Rule-s Inst
learners to rely more heavily on one than on the other.
Type 1( S V Location/S V Location) 7 1 .2 7 1 .2 80.8 80.8
e.g., The pen is i n the box and the
8.3.1 Learning in the instructed condition
pencil is in the drawer.
Training in the instructed condition requires subjects to apply rules Type 2 (S singular + be) 6 5 .4 75 40.4? 88.5
consciously to a display of positive examples. In the transfer phase e.g., Where the car is is in the driveway.
subjects are required to use this knowledge to discriminate Type 3 (S plural + be) 63.5 6 1 .5 5 1 . 9* 88.5
grammatical from ungrammatical strings. The training phase thus e.g., Where the apples are is in the bowl.
encourages them to look for positive evidence of structural Type 4 (S singular + lexical verb) 50* 6 1 .5 6 1 .5 84.6
configurations that conform to the rules they have been taught. e.g., Where Judy teaches is at the
Sentences presented during transfer in all conditions contained equal University.
numbers of grammatical and ungrammatical examples corresponding Type 5 ( S plural + lexical verb) 6 1 .5 55.8* 57.7* 76.9
to the easy and hard rule sentence types presented during training. e.g., Where Mary and Joe relax is by
the pool.
These types are summarized in Table 29, and Appendices G and J
contain the full set of training and transfer task sentences.
1 90 Fundamental similarity of implicit and explicit SLA Fundamental similarity of implicit and explicit SLA 191

Table 29. (Continued) Percentage Correct Responses To Easy singular (e.g., Where Carl writes are at a desk not o n the table).
And Hard Rule Sentence Types Presented In The Transfer Thirdly there were sentences violating the constraint that the verbs
Task used must agree in tense (e.g., Where the bird is was in the sky).
Fourthly there were sentences violating the constraint that
Sentence Type % Correct sentences contrasting locations using this construction should begin
with 'where'(e.g., What the King lives is in an apartment not in a
Hard Rule Ungrammatical Types Imp Inc Rule-s Inst palace). Finally there were sentences of the type (Where Mark
works is in America in Japan), which violates the constraint that
Type l(no inversion with Where) 6 1 .5 5 3 .8* 53 .8* 7 1 .2 the contrasting location phrase must be joined by 'not.'
e.g., Where is Bill is in the country.
Type 2 (invariant singular 'be') 5 9 .6 44 .2* 5 3 .8* 3 4 .6? 8.3.1.1 Confirmation bias and grammaticality
e.g., Where Carl writes are at a desk. Subjects in the instructed condition performed accurately at above
Type 3 (required tense agreement) 63.5 38.5? 48. 1 * 53.8*
chance levels on all types apart from three hard ungrammatical
e.g., Where the bird is was in the sky.
types. Performance on type 3 sentences was at chance, and
Type 4 (required 'Where') 48 . 1 * 4 8 . 1 * 5 5 . 8 * 23 . 1 ?
e.g., What the King lives is in a palace.
performance on types 2 and 4 was above chance but inaccurate,
Type 5 (required 'not' for contrasting indicating wrongly held beliefs about these sentences (see Table 29).
locations) 69.2 76.9 65.4 67.3 Why should performance on these ungrammatical types be poorer
e.g., Where Mark works is in America in the instructed condition than performance on ungrammatical
in Japan. easy rule sentence types, which was significantly above chance and
accurate? One explanation lies in the additional processing and
Key: Imp Implicit condition *
= performance at chance
= reasoning demands imposed by making judgments using knowledge
.
Inc Incidental condition ? performance above chance but maccurate
= = of the hard rule relative to making judgments based on knowledge of
Rule-s Rule-search condition
=
the easy rule.
Inst Instructed condition
=
To make a judgment of grammaticality one can either scan the
sentences to find confirming evidence that it is grammatical, or
scan the sentence to find disconfirming evidence o f
In the transfer phase, ten grammatical and ten ungrammatical ungrammaticality. The latter i s the proper procedure, since transfer
sentences easy rule sentences were presented. The ten set sentences in the present study always contained a number of
ungrammatical examples were five tokens of two types of senten� e correctly applied grammatical subrules which could have been
that violated the rule presented during training (see Figure 5 m mistakenly taken as confirming evidence of grammaticality in cases
section 2.4.3): easy rule ungrammatical type 1 sentences in w?i�h where the sentence also contains a grammaticality violation. In the
subject-verb inversion occurred sentence initially (Golfed Debbie � n case of the easy rule it is relatively easy to check disconfirming
the morning), and easy rule ungrammatical type 2 sentences m evidence. If there is subject-verb inversion, check to see if
.
which subject-verb inversion occurred following fronted adverbmls adverbials of location are at the front. If they aren't, only then is it
of time (On Saturday night danced Charlie). ungrammatical. Of course, instructed subjects were made aware o f
For the hard rule (see Figure 6 in section 2.4.3) there were five this rule during training. It i s difficult to discover it through rule­
types of ungrammatical sentences, eaeh vi� l�ting �ne of the search, or to otherwise induce it, as the poor accuracy of
.
conditions on the hard rule taught durmg trammg. Firstly, there performance of subjects in the other conditions in the present study
were sentences violating the constraint on subject-verb inversion demonstrates. To correctly judge an ungrammatical hard rule
following ' Where' (e.g., Where is Bill is in the house not in the sentence, however, subjects must scan the sentence in order to
garden). Secondly there were sentences violating the c�ns.traint that identify whether any of the five possible violations occur. This may
.
the form of the verb ' be' following the first verb 1s mvanantly be done serially, by checking to see if each of the possible violations
1 92 Fundamental similarity of implicit and explicit SLA
has occurred, or in parallel. There is no simple rule of thumb t o
guide this search (see Faerch 1 986). Here the search fo r
disconfirming evidence is potentially more effortful than i n the
ungrammatical easy rule sentences, since on occasions it may be the
last of the five violations that is checked which leads t o
disconfirmation of the hypothesis o f grammaticality. There could
also be parallel search with the same result if some violations are
more salient than others.
Given the effortfulness of this task, then, the results for poor
performance on hard rule ungrammatical sentences may reflect
'confirmation bias' in the instructed group, i.e., the tendency t o
seek confirmation o f a hypothesis by looking fo r confirming
examples rather than by looking for disconfirming examples (see
Evans et al. 1 993 ; Klayman & Ha 1 987; Wason 1 960). In this case
noticing that the first or second, or the most salient, of the possible
hard rule grammaticality violations has not been made could lead
subjects to wrongly assume that none of the others had been made.
This difference in confirmation, disconfirmation bias could explain
the difference in accuracy to ungrammatical easy rule and
ungrammatical hard rule sentences for the instructed condition. The
above chance performance on type I and type 5 hard
ungrammatical violations suggests that these violations were
checked first. If they were not violated then the sentence was
wrongly judged grammatical. This suggests a depth of search of two
grammaticality violations for the subjects in the instructed
condition. Violations in type 2 and type 3 hard ungrammatical
sentences, which contain a potentially confusing combination o f
verb tense agreement and number agreement violations, do n o t
appear to have been consistently checked by those i n the instructed
condition. Neither were type 4 hard ungrammatical violations,
which posed problems across all conditions, consistently checked by
instructed learners, due possibly to the fact that background
knowledge of the acceptability of sentence initial wh-words like
'what' interfered with the application of the rule regarding this
violation and led learners to overgeneralize the rule.
8.3.2 Learning in the implicit condition
In contrast to the instructed condition, processing in the implicit
condition encourages subjects to search for allowable co-occurrences
of words. Noticing allowable co-occurrences of words that relate to
specific hard rule violations may thus have facilitated performance
by implicit subjects. This could account for superior performance on
Fundamental similarity of implicit and explicit SLA 1 93
hard ungrammatical type 2 and 3 sentences, which are
ungrammatical by virtue of the constraints on the allowable co­
occurrence of verbs which differ in tense, and of verbs which differ
in number marking (see Table 29). A similar explanation could be
given for above chance performance on hard ungrammatical type I
sentences, since in these sentences the two verbs are separated by a
subject nominal. The co-occurrences of verb forms in hard
ungrammatical sentence types I , 2 and 3 therefore contradicts the
bigram and trigram knowledge picked up during the memorization
task. Like all other subjects, those in the implicit condition
performed at significantly above chance accuracy on type 5
sentences . Type 5 sentences are noticeably semantically ill-formed,
_
and possibly they are rejected on this basis by subjects in the
implicit, incidental and rule-search conditions. They were only at
chance on type 4 sentences. Type 4 sentences violate no bigram
knowledge that could have been picked up during training, and could
have been wrongly accepted for the same reason given above for
those in the instructed condition. Alternatively, and in line with the
explanation given for above chance performance on types 1, 2 and
3, these sentences could have been wrongly accepted by those in the
implicit condition because the bigram 'is in,' which these sentences
contain, does occur in the grammatical sentences, and subjects could
have wrongly based their judgement of grammaticality on this
knowledge.
Although the explanation of performance in the implicit
condition offered above is similar to that of Hayes and Broadbent
( 1 988:25 1 ), who describe implicit learning as the "passive
aggregation of information about the co-occurrence o f
environmental events and feature£," and similar also to that offered
by Perruchet and Pacteau ( 1 990; cf. Perruchet, Gallego & Savy
1 990; Vokey & Brooks 1 992) for performance in artificial
grammar learning experiments, there is an important modification
necessary for the purposes of explaining implicit learning of the
data presented in this study. Unlike the artificial grammar
experiments, where the same letters occurred in the same bigram
pairings in the training and transfer phase, the same words did not
co-occur during training and transfer in this experiment.
· Transfer
sentences were novel examples of the same type presented during
training. While the verb ' is' and its forms, which are critical t o
identifying the violations in type 1 , 2 and 3 hard rule
ungrammatical sentences, were necessarily rep.eated, the other
lexical verbs were not repeated, nor were the subjects of sentences
1 94 Fundamental similarity of implicit and explicit SLA
which separate the verbs in type 1 violations. Thus, memory for
the pairs of co-occurring verbs presented in training would need to
be accessed together with knowledge of differences in plural and
singular forms and past and present tense forms which moderated
their relationship in the training set sentences in order to make the
judgment of ungrammaticality on transfer set sentences of types 1 ,
2 and 3 . Activation of prior knowledge of grammatical subrules
therefore likely played a part in the decisions about grammaticality
made in the implicit condition, which were, however, triggered by
noticing restrictions on co-occurring words, rather than by a search
for more global rules. It could have been this grammatical
knowledge that those claiming awareness in the implicit condition
had in mind. In order to confirm this, of course, more sensitive
measures of the information the subjects were aware of would have
to have been made, as acknowledged in section 8. 1 .2
The explanation offered above for differences in performance
by implicit and instructed subjects on hard ungrammatical sentence
types suggests a distinction between the learning resulting from
exposure in each condition which is attributable to differences in the
variety of conscious processing encouraged by the specific training
task demands. Although subjects in the implicit condition did
perform at above chance levels on more ungrammatical hard rule
sentence types than the instructed subjects, the mean performance
for accuracy on all hard ungrammatical sentences for this group is
still low, about 60%. While implicit learning due to memorization
could have occurred following the explanation given above, it is
thus likely to have been piecemeal associative knowledge of
allowable co-occurrences of words, supported by access to
knowledge of plurality and tense. To this extent the memorization
strategies resulting in this knowledge base are likely to have been
insufficient for the acquisition of knowledge of the allowable
discontinuities in word order which characterize the easy rule
grammatical type 2 structures, since these take place over a wider
window of associative learning and attention than two or three
words. This rule also requires that subjects making correct judgments
of acceptability of easy rule grammatical type 2 sentences learn to
distinguish adverbials of location, which allow subject-verb inversion
when fronted, from adverbials of time, which do not. Noticing this
type of information is unlikely to result from the memorization
strategy, and so learning of this constraint on the easy rule did not
take place in the implicit condition.
Fundamental similarity of implicit and explicit SLA 195
8.3.3 Learning in the incidental condition
While any learning that occurred in the implicit condition may
have resulted largely from the memorization strategy encouraged
during training, learning in the incidental condition is likely to have
been constrained by the necessity to read for meaning and impose
sets of propositional relations on sentences. To this extent the
focus of attention would have been on identifying the semantic
roles of subjects and objects in the sentences in order to establish
propositional meaning, and not on the features of tense and number
agreement discussed above. Performance for the incidental group is
above chance, as for subjects in all other conditions, on type 5 hard
rule ungrammatical sentences which are semantically ill-formed by
virtue of missing the negative conjunctive coordinator 'not' which
is important to understanding the propositional value of these
sentences. However, there is a significant difference in accuracy of
response to the two sentences illustrating this type for subjects in
the incidental condition, and this is discussed below. Performance is
above chance but inaccurate on hard rule ungrammatical type 2 and
3 sentences, suggesting that the tense and number agreement
violations in these sentences are overlooked, and are possibly
redundant to the attribution of propositional sense, which is the
basis of decisions of well-formedness encouraged by transfer of the
processing requirements encouraged during training. Performance is
at chance on hard ungrammatical rule types 1 and 4.
Subjects in the incidental group perform differently from subjects
in other conditions in two respects, both of which suggest that they
base their decisions about grammaticality on more than grammatical
knowledge, and that they are influenced, to a greater extent than
subjects in the other conditions, by semantic acceptability and
lexical aspects of transfer sentences. Firstly, there is the finding,
reported as evidence supporting Hypothesis 6, that aptitude scores
for subjects in this condition, unlike those for all other conditions,
show no significant correlation with success on easy or hard rule
sentences. The abilities that contribute to successful learning in this
condition are not those captured in estimates of grammatical
sensitivity, or memory based on paired-associate learning. The
learning that results from performance in this condition thus
appears to be independent of traditional estimates of the ability t o
focus on form, or remember examples. This is i n contrast t o
learning in the rule-search and instructed conditions, which is
significantly correlated with measures of both of these abilities, and
 1 96 Fundamental similarity of implicit and explicit SLA
to learning in the implicit condition which is significantly correlated
_ be evidence of transfer of processing strategies encouraged by the
with measures of grammatical sensitivity.
S�condly, subj_ects in this condition, unlike subjects in all other
.
cond�t10n�, a�e, m a number of cases, not sensitive to the type
class1ficatlon imposed on sentences in the transfer set. As reported
in Chapter 7, section 7. 1 .5 , McNemar tests were performed t o
calculate the frequency o f accurate responses to the two sentence
tokens exemplifying each of the five easy grammatical types, and
each of the five hard grammatical and five hard ungrammatical
types for subjects in each condition. The results of these
calcu�ations show that for the_ incidental condition alone responses
to pairs of sentences representmg types were significantly different
In particular, the frequencies of accurate responses to sentence�
exemplifying four of the transfer set sentence types were
significar_itly different (p<. 0 1 ). These four types were easy
grammatical type l (sentence 1 a mean accuracy of .962, versus
sentence 2, a mean accuracy of .692); hard grammatical type 1
(sentence 1 a mean accuracy of .462, versus sentence 2, a mean
accuracy of .962); hard ungrammatical type 2 (sentence 1 a mean
accuracy of . 6 1 3, versus sentence 2, a mean accuracy of .259); and
hard ungrammatical type 5 (sentence 1 a mean accuracy of . 6 1 5,
versus sentence 2, a mean accuracy of .923). The two sentences
exemplifyi�g each of these ty�es are illustrated in Appendix J,
. rule grammatical type 2 sentences, there are no responses to hard
together with sentences exemphfymg all other types used in the
transfer set. For example, as Appendix J shows the first of the hard
. would support this conclusion. Subjects trained in the rule-searc.h
'
ungrammatical type 5 sentences is less noticeably semantically ill-
formed (Where the horse stands is in the field in the barn) than the
�econd (Wh�:e Mark work� �s in America in Japan). While subjects
m all cond1t1ons are sensitive to the ungrammaticality of these
sentences, and are above chance and accurate in responding t o
t�em, onl� i n the incidental condition is there a significant
difference m acc�racy of response to the two sentences. It is clearly
not the gram�at1cal status of the sentences that distinguishes them,
but !he�_r relative de�:ee of semantic acceptability, and it is this that
_
the mc1dental cond1tion-m contrast to the implicit memorization
condition and the rule-focused explicit conditions-a- re most
sensitive to, since this was the aspect of the stimulus sentences that
incidental subj ects w_ere instructed to attend to during training. Just
as for subjects m the other conditions, therefore the
�
grammaticality judgments of incidentally trained subjects ap ear t o
Fundamental similarity of implicit and explicit SLA 1 97
specific demands of their training condition task.
8.3.4 Learning in the rule-search condition
As in Reber's studies, those in the rule-search condition performed
poorly on hard rule sentence types, being above chance and accurate
only in responses to type 1 and type 4 grammatical hard rule
sentences, and type 5 ungrammatical hard rule sentences (see Table
29). To this extent they performed more poorly than those in the
implicit condition, who were below chance on only two hard rule
sentence types, though the results of overall hard rule performance,
reported in Chapter 6, show no significant difference between the
groups. Unlike the findings from Reber's studies, those in the rul�­
search condition do not perform significantly better than those m
the implicit condition on easy rule sentences. The explanation
given for poor performance by rule-search subjects on complex
stimuli given by Reber is that subjects in this condition search for
and find rules, but these are based on false hypotheses which are n o t
commensurate with the structure o f the complex stimulus domain.
The findings from the present study, however, do not suggest the
rule-search subjects had found and consistently misapplied any rules,
either easy or hard. With the exception of performance on hard
rule sentence types that are both above chance and inaccurate which
condition perform worse on hard rule sentences than subjects in any
other condition, being above chance and accurate only in response
to hard rule grammatical types 1 and 4, and hard rule ungrammatical
type 5 sentences . It must be noted, however, that this training
condition differed from the othe:·s in the present study in not being
given feedback about the accuracy of their responses during training.
Such feedback was, instead, limited to questions asking if they were
maintaining rule-search. This option was chosen due to the fact that
asking specific questions about the structure of training examples ,
and giving feedback on the answers, would have preempted the
subjects' own rule-search decision-making processes, and guided
their rule-search. In order to ensure comparab ility with Reber' s rule­
search condition this alternative was avoided. However, this m ay
have negatively affected their performance relative to the other
conditions, who all received some form of instructional feedback
appropriate to their training conditions, albeit feedback m the
implicit and incidental conditions that was not targeted at the
1 98 Fundamental similarity of implicit and explicit SLA
formal characteristics of the sentences regulated by the easy and
hard rules.
8.4 Summary
The explanation of behavioural differences in performance o n
transfer set sentence types offered above i s compatible "'.ith
.
Schmidt's 'noticing' hypothesis, and the model of noticmg
proposed in 3 .2.6. Differential perfor� ance on tra?sfer set
sentence types is consistent with the claim that all subjects were
transferring the consciously mediated processing demands of the
different training tasks to the transfer task (cf. B laxton 1 989; Graf
& Ryan 1 990; Hamann 1 990; Roediger et al. 1 989; s�adler 1 99 5 ),
. research into the differences between implicit and explicit L2
and that variations in the attentional demands of trammg tasks led
to variation in the extent of learning across conditions-not access
to separate conscious and unconscious learning systems (cf. K�ashen
1 982, 1 985, 1 994; Paradis 1 994; Zobl 1 992, 1 995). The evidence
of performance on hard rule sentences does not supp�rt Kr�shen
and Reber's claims that implicit and incidental learnmg will be
superior to explicit rule-search or instructed learni�g where . the
stimulus domain is complex. Although clearly supenor to subj�cts
trained in other conditions on the easy rule and on grammatical
examples of hard rule sentences, the poor performa!lce of instructed
learners on ungrammatical hard rule sentences pomts to the need
for negative evidence of ungrammaticality in ?�der t� prevent
overgeneralization of instruction based on positi�e eviden.ce o f
grammaticality alone (cf. White 1 99 1 ). Further, differences i� the
extent of awareness across conditions also reflect the attent10nal
demands of the training tasks. However, even in the implicit and
incidental conditions-which do not explicitly focus subjects'
attention on form-large numbers of subjects claim to have noticed
rules. In the case of implicit learners, those who noticed and were
able to verbalize rules were those who had significantly higher
aptitude as measured by the grammatical sens �tivity subtest of the
.
MLAT. Coupled with the finding of high correlations o f
grammatical sensitivity and accuracy o n easy and hard �les for
implicit learners, these results suggest that those wh? did �earn
.
effectively in the implicit condition, were those with high aptitude
and who therefore were more likely to · notice, and so learn, the
patterns of structural covariation described by the pedagogic rules.
Chapter 9
Summary And Conclusions
9. 0 Introduction
In this chapter the results of the present study are related to other
learning and conclusions are drawn regarding the comparability o f
Reber's claims about implicit learning to Krashen 's claims about
acquisition. Finally, after describing the implications and limitatio ns
of the present study, the main findings are summarized.
9.1 Implicit and explicit L2 learning
The research questions addressed in this study are of basic
importance to SLA theory, as well as to prescriptions for L2
pedagogy based on its findings. As stated in Chapter 1 there is
considerable debate in L2 pedagogy about the relative merits of a
focus on form, a focus on meaning, or a combination of both (see
e.g., Hulstijn 1 990a; Krashen 1 982, 1 985, 1 993 , 1 994; VanPatten
1 990; VanPatten & Cadierno 1 993). Arguments have been put
forward that instruction is beneficial to subsequent L2 develop ment,
and that its role in facilitating attention to form is causal (Ellis
1 990, 1 993a, 1 994b; Long 1 988, 1 99 1 , 1 994; Schmidt 1 990,
1 995). Effect of instruction studies examining the relationship
between instruction, attention to form, and learning have typically
been conducted under classroom conditions (e.g., Gass 1 982; Harley
1 989, 1 993 ; Leeman et al. 1 995; Lyster 1 994; Spada & Lightbown
1 993 ; Yip 1 994). Given the nature of such research it is unclear
what learning processes underlie the aggregate advantage of
instructed groups, and what the significant interactions of learning
conditions, as distinct from instructional treatment, and the formal
characteristics of language are. Experimental studies of the effect of
instruction are necessary to identify such processes, and to thereby
address Krashen 's claim ( 1 985) that the aggregate advantages of
instruction summarized by Long ( 1 983 , 1 988) are due not t o
conscious selective attention to form facilitated by instructional
200 Summary and conclusions
treatments, but to the fact they represent acquisition-rich
environments in which comprehensible input is available in
sufficient quantities to promote incidental learning or acquisition.
Given, also, that a large part of the debate about the effectiveness
of formal instruction has revolved around the question of whether
to teach grammar rules (Krashen 1 992; Lightbown & Pienemann
1993; Odlin 1 994; Rutherford & Sharwood Smith 1 988), it is
particularly necessary to supplement quasi-experimental
demonstrations of the effectiveness of instruction in pedagogic rules
(e.g., Alanen 1 995; Doughty 1 99 1 ; Eckman et al. 1 988; Gass 1 98:2;
Lightbown & Spada 1 990; Lightbown, Spada & Ranta 1 99 1 ; Lyster
1 994; Master 1 994) with evidence from studies using pure
experimental designs, as in N.Ellis (1993), and the present study.
Second language acquisition theory has often invoked implicit
knowledge as one of a number of explanatory constructs in models
of instructed L2 development (see Bialystok 1 98 1 ; Ellis 1 990,
1 993a, 1 994a, 1 994b, 1 994c, 1 995; Krashen 1985, 1 994; Towell &
Hawkins 1 994) without addressing in empirical detail how such
knowledge results from conditions of learning like those examined
in the present study. The debate about the representation o f
implicit knowledge, and the extent o f its activation and
development during implicit learning, reviewed in Chapter 4, is
important to address in the context of SLA if clear prescriptions
about the role of implicit knowledge in syllabus design and L2
methodology are to be given empirical substance. Recent L2
research has attempted to address these issues using experimental
methodologies, and some have addressed the question of the
relationship between implicit learning and rule complexity
examined in the present study.
9.1.1 Second language learning and rule complexity
The easy and hard pedagogic rules presented in the present study
were identified by the process of empirically validating the
distinction using ratings of the rules, and the structures they
contained, by experienced L2 teachers. As discussed in Chapter 1 ,
other approaches to distinguishing the complexity of structures t o
be learned by L2 learners have been motivated by linguistic notions
of accessibility and entailability, within the framework of Universal
Grammar, and by the notion of generalizability based on research
into typological markedness (see sections 1 . 1 .2-4 ). If these criteria
are adopted for distinguishing structures of different orders o f
complexity in studies o f implicit and explicit learning by L 2
Summary and conclusions 20 1
learners, then it will be important to provide a parallel estimate o f
the difficulty o f the pedagogic explanation o f these structures, as in
the present study, since the pedagogic explanation itself must be
appropriate to the level of structural detail it describes (see Figure 1 ,
section 2 . 1 .3).
The issue of rule complexity and learning under implicit and
explicit conditions has recently been addressed by other L2
researchers. Hulstijn and DeGraaff ( 1 994) have motivated a series
of hypotheses about the relationship between implicit and explicit
learning and structural complexity. They claim explicit learning will
be more effective than implicit learning in the case of properties o f
language falling outside the scope o f Universal Grammar. Explicit
instruction, they claim, will be more effective than implicit learning
where the targets of instruction are structures conforming to a
parameter which is a subset of the setting for the L l . For the
reasons given in Chapter 2, the present study avoided motivating
the distinction between easy and hard rules using Universal
Grammar. However, Hulstijn and DeGraaff clearly see the influence
of Universal Grammar as separable from the issue of rule
complexity, since this is addressed in a further hypothesis, th at
explicit instruction will have more effect in the case of complex
pedagogic rules than in the case of simple rules. Complexity is
described by Hulstijn and DeGraaff in 'cognitive' terms, and is
illustrated by two hypothetical examples of rules of plural
suffixation in two languages which differ in the number o f
derivational steps they require. Their notion o f complexity, then, is
related to the number of steps needed to decide on whether a form
applies in a structural context. They distinguish this from
' linguistic' descriptions of complexity (presumably those
motivating the first two Universal Grammar-based hypotheses), and
evidence of complexity inferred from studies of development
showing some structures to be early, and some to be late acquired.
Their hypothesis, of course, is the opposite one to the hypothesis
about the relationship of complexity to learning condition in the
present study, motivated by the claims of Krashen and Reber, and it
is not supported. The evidence from this study does not suggest th at
explicit instruction is more effective on the hard rule than learning
under implicit or incidental conditions. It also shows clear
advantages for the instructed condition in learning the easy rule
relative to the other conditions. The fact that the simple rule used
in the present study is reducible to a two-step rule of thumb, t o
illustrate the two conditions which would violate it were they not
202 Summary and conclusions
met (see section 8 .3 . 1 ), in contrast to the hard rule which has a
larger number of potential violations, suggests that Hulstijn and
DeGraaff would also be satisfied with the characterization of the
complexity differential for the two rules used in the present study.
Hulstijn and DeGraaff have not reported any findings based on their
hypotheses to date.
A small-scale pilot study, using only six subjects (DeKeyser
1 994) has also addressed the issue of the relationship between rule
complexity and implicit learning. The first hypothesis for
DeKeyser's study was identical to the second hypothesis for the
present study, and the opposite of the prediction made by Hulstijn
and DeGraaff: implicit learning is better than explicit learning for
complex rules. The attribution of DeKeyser's second hypothesis,
that explicit learning is better than implicit learning for simple
rules, to Krashen and Reber is not quite accurate. In the case o f
Krashen ( 1 982, 1 985) the qualification that the advantages for
explicit learning will only be short-lived, must be added, since
Krashen has argued that despite short-term rate advantages for
learning on easy rules, implicit learning and the acquired system
eventually catches up with the development of the learned system .
Similarly, it appears to follow from Reber's claim that implicit
learning is always more robust than explicit learning, and develops
in advance of it (see Reber 1 993 and section 4. 1 ), that there could
be no more than equivalency in performance on simple rules under
implicit and explicit conditions. It is true, though, that Reber et al.
( 1 980) did demonstrate the superiority of explicit instruction,
followed by structured exposure to relevant examples, over
instruction following exposure to examples, and over implicit
learning following instructions to memorize. Given the nature o f
Reber's experimental treatments it was not possible to assess the
duration of this learning effect. However, Reber et al. add the
further caveat that even for instruction on simple rules to be
effective when followed by structured exposure, attention must be
consistently directed at those aspects of the stimulus display which
are maximally coordinate with underlying rule. Explicit instruction
on simple rules, that is, can be less effective than implicit learning if
the rule itself is inefficient in directing learners' attention to the
critical aspects of the structure to be learned (see Figure 1 , section
2 . 1 . 3 ).
DeKeyser's second hypothesis was also modified by the
additional claim that the simple rules must be categorical in nature
(rules that always apply when their structural description is met). In
Summary and conclusions 203
contrast, DeKeyser claimed implicit learning would be equal to or
better than explicit learning in the case of prototypical rules, which
are probabilistic but not categorical. Like the present study
DeKeyser also hypothesized that the key dependent variables,
categorical and prototypical rules, would differ in terms of their
learnability. Prototypical rules, he claimed, would be harder to learn.
Assessment of the learning that took place in 25 twenty-minute
learning trials for 3 implicit and 3 explicit learners was done by
means of a grammatical classification task on counterbalanced
grammatical and ungrammatical sentences produced by the rules.
The results of the pilot study are similar to the results of the
present study. DeKeyser found that explicit learners performed
better on simple rules, and that implicit and explicit learners
performed equivalently, but poorly, on hard prototypical rules.
These findings were confirmed by the results of a larger scale study
using the same design (DeKeyser 1 995).
9.1.2 Generalizing from Reber's claims about implicit learning to
Krashen 's claims about acquisition
(i) The unconscious and conscious conditions. The present study
matched learning conditions motivated by Reber's experimental
methodology with conditions implied by Krashen's distinction
between conscious learning and unconscious acquisition. The
analyses for Hypotheses 1 and 2 support the claim for
comparability of performance by those in the implicit and
incidental conditions, but not the claim for comparability o f
performance by those in the rule-search and instructed conditions.
There are no significant differences between implicit and incidental
learners in speed and accuracy of performance on transfer set
exemplars of the easy and hard rules. However, instructed learners
are significantly more accurate than rule-search learners in
performance on easy and hard rule structures, and there is an effect
approaching significance for the greater speed of the rule-search
learners relative to the instructed learners on both rules.
The results for the instructed group parallel the findings of Ellis's
( 1 993) finding for superior performance by a structured group, who
were given rule instruction and exposure to relevant examples,
relative to performance by subjects instructed in rules with minimal
exposure to examples, or to subjects receiving exposure alone. T o
the extent that superior learning by the instructed group in the
present study, and by Ellis' structured group, can be attributed t o
204 Summary and conclusions
task conditions which focus attention on relevant aspects of the
stimulus domain, and so facilitate noticing, they also support the
findings of other studies that have shown superior effects for
learning as a result of structured exposure via ' input-enhancement'
(Sharwood Smith 1 99 1 ), e.g., the use of multiple-choice margin
glosses for underlined words in Watanabe ( 1 992); the use o f
italicization i n Alanen ( 1 995); the use o f colour-coded highlighting
in Doughty ( 1 99 1 ); the use of balded upper-case lettering ' in Shook
( 1 994); and the use of instructions to attend to form and meaning
in Hulstijn ( 1 990a). Likewise, Ellis found evidence for implicit
learning by the random exposure group in terms of above chance
performance on a transfer task requiring grammaticality judgements
to novel examples of structures presented during training, but
concludes, "it is certainly laboriously slow and it cannot compare
with the superior levels that can be encouraged by structured
exposure to relevant exemplars" ( 1 993 :3 09). This also appears t o
b e true o f the evidence for implicit learning in the present study.
(ii) Reber 's implicit and rule-search conditions. A comparison of
the findings for Reber's implicit and rule-search conditions also
reveal similar findings to those reported by Reber ( 1 989, 1 993) in
studies of artificial grammar learning. Although the analyses for the
main hypotheses revealed no significant differences between these
conditions in speed and accuracy of response on easy and hard rules,
the further analyses of chance and above chance performance on
sentence types reported in Chapter 7 and above (see Table 29 in
section 8 .3 . 1 ) do suggest differences in line with Reber's claims. As
Figure 14 shows (see section 6. 1 . 1 ) implicit learners have a higher
mean accuracy of response to hard rule sentences than rule-search
learners, while rule-search learners have a higher mean accuracy of
response to easy rule sentences. As Table 29 shows, implicit
learners perform better than any other condition on hard rule
sentences, using measures of above chance performance to sentence
types. The rule-search learners are worst, being accurate and above
chance only on responses to three of the hard rule sentence types,
confirming Reber' s claim that conscious rule-search can be effective
where the stimulus domain is simple, but will be less effective than
implicit learning where it is complex.
(iii) Krashen 's incidental and instructed conditions. A
comparison of the relative effectiveness of incidental and instructed
conditions on easy and hard rules, however, does not support
Summary and conclusions 205
Krashen' s (198 1 , 1 982, 1 985) claims. Instructed learners are
significantly more accurate than incidental learners in performance
on easy rules, and the analyses of chance and above chance
performance on hard rule sentence types also reveals superior
learning of the hard rule by the instructed condition, since they are
above chance and accurate in response to seven of the ten types,
compared to above chance and accurate performance on only five
hard rule sentence types by those in the incidental conditioIJ.
Finally, although the main analyses for the study reveal implicit and
incidental learners to be performing similarly, there is some
behavioural evidence of differences between the groups. Implicit
learners, as noted above, perform accurately and above chance on
more hard rule sentence types than any other condition. The
performance of implicit learners on easy and hard rules also
correlates significantly with measures of aptitude, as reported in
Chapter 6. In contrast, there is no significant correlation o f
performance by the incidental condition with aptitude o n either rule
(see Robinson l 995b ). Responses in the incidental condition also
appear sensitive to lexical and propositional information which
leads to significantly different responses to tokens of four of the
sentence types, whereas there are no significant differences in
response to tokens of the same type by subjects in other conditions.
9.2 Limitations of the study and directions for future research
The findings of the present study must be interpreted in the light o f
its limitations and these are listed below.
(i) The generalizability of the experimental learning conditions.
Minor adjustments to instructional variables can cause potentially
major changes in learning effects. As the review of implicit learning
studies in Chapter 4 showed (see Table 3 , section 4. 1 . 1 ), there is
some inconsistency across the studies of artificial grammar learning
in cognitive psychology, unrelated to manipulations of variables o f
interest, which, given the small effect sizes typical in such studies
(see Table 4, section 4. 1 . 1 ), can obscure the interpretation of
individual results and limit claims for generalizability made on the
basis of groups of results from supposedly similar studies. The
present study has motivated the research hypotheses with specific
reference to current work in cognitive psychology and examined
the generalizability of these claims to the field of SLA research.
This has the advantage of locating current SLA research in relation
to issues explored in detail in another field, whose findings can
206 Summary and conclusions
thereby be more clearly related to SLA data and deepen its
interpretation. Nonetheless, it could be argued that the implicit,
incidental and rule-search conditions in the present study are less
ecologically valid, and so produce results which are less generalizable
to L2 pedagogy, than conditions operationalized in Alanen ( 1 995),
Doughty ( 1 99 1 ) and Watanabe ( 1 992) which simulate learning
conditions more familiar from contemporary L2 pedagogy. Given,
also, the differences in operationalizing learning conditions in those
studies, and the present study, further research is needed to confirm
the generalizability of the apparently similar findings regarding the
relative effectiveness of implicit and explicit learning.
(ii) The generalizability of the easy/hard rule distinction. The
motivation for distinguishing rules and structures of different orders
of complexity in the studies of Hulstijn and DeGraaff ( 1 994), and
DeKeyser ( 1 994), reviewed above, also differs from the procedure
adopted in the present study. To this extent the generalizability o f
claims based o n the complexity differential are limited to
distinctions between pedagogic rules of different orders o f
complexity based on a consensus agreement among experienced
teachers about the relative difficulty of the structures and the rule
explanations following the procedure described in section 2 .4 .
Further confirmatory research regarding the differential
effectiveness of learning conditions is needed which adopts other
motivations for distinguishing rule complexity, described in sections
1 . 1 . 1-4. Despite the problems of control, though, there are
considerable advantages in terms of external validity if actual
language samples are used as the dependent variables, as in the
present study, and in Doughty ( 1 988, 1 99 1 ), Alanen ( 1 995),
Watanabe ( 1 992) and Ellis ( 1 993 ), rather than artificial or semi­
artificial examples as in DeGraaff ( 1 995), Hulstijn ( 1 990a) and
DeKeyser ( 1 994, 1 995).
(iii) Amount of exposure. The findings of the present study are
based on limited exposure to examples during training. It is possible
that clearer effects for memory-based learning in the implicit
condition, and learning in the incidental condition, would be
obtained if more examples had been presented, particularly given
Ellis' ( 1 993) characterization of implicit learning as 'slow' and
Krashen's ( 1 985, 1 994) characterization of acquisition as a process
taking place over longer periods of time than instructed learning,
and requiring large amounts of comprehensible input. Increasing the
Summary and conclusions 207
number of examples presented during trammg might also have
affected performance under the explicit rule�based conditions. It is
possible that with greater exposure, a transition in strategies for
making grammaticality judgments, from rule-based to instance-based
memory retrieval, might occur in the explicit conditions. This has
been proposed by Logan ( 1 988a; Logan & Klapp 1 99 1 ; Logan &
Stadler 1 99 1 ) in relation to rule-based learning of alphabet
arithmetic problems, and has been examined in relation to L2 rule
learning under explicit conditions by Robinson and Ha ( 1 993), who
found some evidence for the hypothesized strategy switch. This
transition may also be affected by the complexity of the stimulus
domain, with instruction on complex stimuli being relatively
impervious to memory-based processing, in contrast to instruction
on simple stimuli, or vice versa.
(iv) The effects of positive and negative evidence. The training
conditions for the present study involved the presentation o f
positive examples o f grammatical sentences. Negative evidence o f
what i s not possible was not presented during training. I t i s possible,
however, that the significantly poorer performance of the
instructed group on hard rule ungrammatical sentences, relative t o
performance o n hard rule grammatical sentences, would have been
improved if ungrammatical examples had been presented and judged
during training. There was no significant interaction o f
grammaticality and accuracy or speed o f performance o n either rule
for subjects trained in other conditions, or on easy rules for those in
the instructed condition. The present study thus provides evidence
of the limitations of instruction in hard rules when this is
accompanied by positive evidence of what is possible alone. Further
research is needed into the reasoning processes of subjects in
instructed conditions. Little evidence for implicit learning was
shown in the present study, though the explanation given in section
8.3.2 suggests that such learning is possible from positive evidence,
but is likely to be most effective in the case of patterns of co­
occurrence occurring within a small window of attention, and may
be limited to two or three features. Rules regulating structures of this
level of complexity and perceptual detail may be the most
susceptible to implicit learning.
(v) Duration of learning effects. The duration of learning effects
is also a question for further research. The present study measured
only short-term learning apparent in a transfer task immediatly

208 Summary and conclusions
following training. Consequently it is unclear whether the short­
term gains attributable to instruction on easy rules are also evident
over longer periods of time, or whether they are short-lived relative
!o !he smaller i!1�remental gains from learning under implicit and
mc1den�al conditions. Reber has claimed that implicit learning is
robust m the face of both clinical disorders and aging processes (see
Abrams & Reber 1 98 8). Consequently its effects may be retained
over longer intervals than is the case with explicit learning.
(vi) Limitations of the transfer test of learning. As in Reber s
;
�tudies, the presen� study used responses to a grammaticality
Judgment test as an mdex of learning. However, additional evidence
?f avail�bility . for productive use is necessary to supplement
mfor� ation gamed by such tests. The transfer test also required
consc10us attempts to make grammaticality judgements, and to this
extent favours transfer of the processing strategies encouraged by
the rul.e� search and instructed task conditions. It is possible that
recogmt10n tests, such as those used in the implicit memory tests
described in section 4.2.3, would be more sensitive to the
information obtained during training under implicit and incidental
conditions.
(vii) The scope of the knowledge base involved in implicit and
explicit learning. The present study has dealt only with grammatical
knowledge presented and tested in the visual modality. Given the
�cope of.the knowledge and skill base involved in L2 development,
mformat10n about the relative effectiveness of implicit and explicit
learning of pragmatic, phonological and lexical aspects of L2
knowledge is needed, as is information about cross-modal transfer o f
information learned under different conditions. There has been little
research i ?to modality effects in artificial grammar learning, but
what studies there are (e.g., Whittlesea & Dorken 1 993) have
shown that where implicit learning occurs it is largely modality
specific. This is a particularly important issue for SLA research t o
address.
9.3 Summary offindings
The main findings of the present study can be summarized as
follows:
1. The � asy Rule. Instructed subjects are significantly more
accurate m response to easy rule sentences than subjects trained in
Summary and conclusions 209
implicit, incidental and rule-search conditions. Only subjects trained
in the instructed condition are above chance and accurate in
response to all easy rule sentence types.
2. The Hard Rule. Instructed subjects are significantly m ore
accurate in their response to hard rule sentences than those trained
in the rule-search condition. There is no significant difference in
accuracy of response to hard rule sentences for those trained in
instructed, implicit and incidental conditions.
3. Gram maticality. The positive effect of instruction, apparent
from the superior accuracy of instructed subjects on easy rule
sentences and hard rule grammatical sentences relative to subjects
trained in other conditions, is not reflected in accuracy o f
performance on hard rule ungrammatical sentences. Instructed
subjects perform poorly on these types. Those trained in the
implicit condition are above chance in accuracy of response t o
more hard rule ungrammatical sentence types than those trained in
incidental, rule-search or instructed conditions.
4. Awareness. There are no significant differences in measures of
having noticed rules, or ability to verbalize rules for those trained in
instructed, implicit, incidental and rule-search conditions.
Significantly greater numbers of instructed and rule-search subjects
than implicit and incidental subjects report that they were looking
for rules during training.
5. Aptitude. Aptitude correlates significantly with accuracy of
response for subjects trained in instructed, implicit and rule-search
conditions. There are no significant correlations of aptitude with
accuracy of response to easy or hard rules for those trained in the
incidental condition.
6. Variance. There is significantly less variance in accuracy for
instructed subjects versus all other subjects, and significantly less
variance in reaction time for rule-search subjects versus all others.
 Appe n dix A

GRAMMATICALITY JUDGMENT TEST

USED TO IDENTIFY UNFAMILIAR
STRUCTURES AS INPUT TO THE RATING
SCALE SURVEY OF RULE COMPLEXITY

Choose one answer for each sentence:

e.g., Mary is a teacher.
grammatical ungrammatical not sure

1 . It's Mary who eats cake.

grammatical ungrammatical not sure

2 . John eats bread and Mary eats cake.

grammatical ungrammatical not sure

3 . I placed book that on the table.

grammatical ungrammatical not sure

4 . Who did you suggest I apply to?

grammatical ungrammatical not sure

5 . That there cow gives a lot of milk.

grammatical ungrammatical not sure

6 . Who Mary is happy to see?

grammatical ungrammatical not sure

7 . What Mary does is eat cake.

grammatical ungrammatical not sure

8 . The cheese is in the bag and the bread is in the basket.

grammatical ungrammatical not sure
212 Appendix A Appendix A 213

9 . Who did John give the book to? 2 3 . Who did you suggest that should apply?
grammatical ungrammatical not sure grammatical ungrammatical not sure
1 0 . To whom did John give the book? 24. I saw the dog barked.
grammatical ungrammatical not sure grammatical ungrammatical not sure
1 1 . Where the cheese is is in the bag not in the basket. 2 5 . Is what Mary does good?
grammatical ungrammatical not sure grammatical ungrammatical not sure
12. What Mary does is good. 26. He jogs in the morning.
grammatical ungrammatical not sure grammatical ungrammatical not sure
13. Mary John likes very much. 27. I saw the dog that you fed.
grammatical ungrammatical not sure grammatical ungrammatical not sure
14. A copy of that book exists. 2 8 . Who did you suggest should apply?
grammatical ungrammatical not sure grammatical ungrammatical not sure
1 5 . To John yesterday I gave the book that I read. 2 9 . This cow here gives a lot of milk.
grammatical ungrammatical not sure grammatical ungrammatical not sure
1 6 . In the morning he jogs. 3 0 . John hit him so I was angry.
grammatical ungrammatical not sure grammatical ungrammatical not sure
1 7. On Tuesday he jogs always. 3 1 . Who did you suggest that I apply to ?
grammatical ungrammatical not sure grammatical ungrammatical not sure
1 8. I have never seen him. 3 2 . It' s cake Mary that eats.
grammatical ungrammatical not sure grammatical ungrammatical not sure
1 9 . Into the house ran John. 33. Never have I seen him.
grammatical ungrammatical not sure grammatical ungrammatical not sure
2 0 . I saw the dog that barked. 34. John ran into the house.
grammatical ungrammatical not sure grammatical ungrammatical not sure
21. From school Mary John took. 3 5 . That cow there gives a lot of milk.
grammatical ungrammatical not sure grammatical ungrammatical not sure
22. I saw the dog you fed. 3 6 . There exists a copy of that book.
grammatical ungrammatical not sure grammatical ungrammatical not sure
 214 Appendix A

3 7 . I sent the book that I read yesterday to John.

grammatical ungrammatical not sure
Appendix B
3 8 . This here cow gives a lot of milk.
grammatical ungrammatical not sure
RATING SCALE SURVEY OF RULE
3 9 . John' s hitting him angered me. COMPLEXITY
grammatical ungrammatical not sure

4 0 . Into the house John ran. Could you please give the following information?
grammatical ungrammatical not sure Name. _________

Native language·------=-=------;----------
4 1 . John gave the book to Mary. No. of months/years ESL teaching experience. ______
_

grammatical ungrammatical not sure No of months/years experience teaching other languages. ___

No. of languages learned, and degree of proficiency in each.

42. John eats bread and Mary eats cake.
grammatical ungrammatical not sure -
- - -
Degreesand/or professional qualifications earned orcurrently
43 . I sent to John the book that I read yesterday. studying toward s.__________ __ __

grammatical ungrammatical not sure

_ __ _

Please look through the following eight rules of English. Rate t�em
in terms of their difficulty for a second language learner of Enghsh
using the following rating scales. A rating of -3 means very easy. A
rating of + 3 means very difficult.

Rule 1 -3 -2 -1 0 +l +2 +3

Rule 2 -3 -2 -l 0 +l +2 +3

Rule 3 -3 -2 -1 0 +l +2 +3

Rule 4 -3 -2 -1 0 +l +2 +3

Rule 5 -3 -2 -1 0 +l +2 +3

Rule 6 -3 -2 -1 0 +l +2 +3

Rule 7 -3 -2 -1 0 +l +2 +3

Rule 8 -3 -2 -1 0 +1 +2 +3
216 Appendix B
RULE 1
Some sentences contain a subject a verb and a location phrase or a
time phrase. Here are some examples:
Joan (subject) crashed (verb) into the wall. (location phrase)
John (subject) slept (verb) in the morning. (time phrase)
Both location and time phrases can be moved to the front of the
sentence:
Into the wall Joan crashed.
On Tuesday morning John slept.
The verb can also come before the subject, but only when location
phrases are at the front of the sentence:
Into the wall (location phrase) crashed (verb) Joan. (subject)
* On Tuesday morning (time phrase) slept (verb) John. (subject)
The * means this sentence is ungrammatical.
RULE 2
Some sentences consist of a subject, a verb and a time phrase. Here
are some examples.
He (subject) jogs (verb) in the morning. (time phrase)
John studies in the evening.
Time phrases can be moved to the front of the sentence:
In the morning (time phrase) he (subject) jogs. (verb)
In the evening John studies.
Appendix B 217
RULE 3
Some sentences describe an action and a response to that action.
Here are some examples.
John hit Eric. (action) Bill was angry. (response)
Mary told a joke. (action) I was amused. (response)
These sentences can be combined, as in the following examples:
John 's hitting Eric angered Bill.
Mary 's telling a joke amused me.
To make sentences like these make the subject of the first (action)
sentence possessive and change the verb to the 'ing' form. The
whole of the first (action) sentence then becomes the subject. For
example:
John 's hitting Eric (subject) . . .
Mary 's telling a joke. . .
Delete the subject and verb of the second (response) sentence and
change the adjective into a verb (angry > angered) that carries
tense. The deleted subject of the second sentence is then changed to
the object.
. . . angered (verb) Bill. (object)
. . . . amused (verb) me. (object)
218 Appendix B
RULE 4
Some sentences tell about the locations of two things e.g.,
Alice stands on the right and Judy stands on the left.
We can change this sentence to focus on one of the locations, e.g.,
Where Alice stands is on the right.
To make sentences like these, first, choose the subject you want to
emphasize, 'Alice', then place 'where' in front of it:
Where Alice . . .
Next, follow the subject with the verb, 'stands' :
Where Alice (subject) stands (verb) . . .
Note that the verb cannot come before the subject:
* Where stands (verb) Alice (subject) ...
The * means this sentence is ungrammatical.
Next add a singular form of the verb 'be' which agrees in tense
followed by the phrase describing the location of the subject:
Where Alice stands is on the right.
Where Alice stood was on the right.
If the verb does not agree in tense, the sentence is ungrammatical:
* Where Alice stands was on the right.
For special emphasis, the second location can be contrasted with the
focus location by joining the locations using 'not' . This 'not' is
required. Without it the sentence is ungrammatical.
Where Alice stands is on the right not on the left.
* Where Alice stands is on the right the left.
Appendix B 219
RULE S
Some sentences describe transfer of possession from one person to
another. For example:
John gave the book to Mary.
It is possible to turn this sentence into a question if the person
receiving the object is unknown.
Who did John give the book to?
To make sentences like these first choose a wh-word that agrees
with the unknown indirect object you are questioning. This goes at
the front of the sentence. For people the wh-word is always'who' :
John gave the book to Mary.
* What did John give the book to?
The * means this sentence is ungrammatical.
Next add a form of the verb 'do' that takes the tense of the
sentence, and follow this with the subject and the main verb. Note
that the main verb is in the infinitive form, it does not carry tense:
Who did John (subject) give (verb) . . .
* Who did John (subject) gave (verb) . . .
Place the object after the verb:
Who did John (subject) give (verb) the book. (object)
Finally add the preposition from the first sentence:
Who did John give the book to?
220 Appendix B Appendix B 22 1

RULE 6 RULE 7

Some sentences contrast two activities. For example: Some sentences are used to indicate a specific subject.

John eats bread and Mary eats cake. That cow gives a lot of milk.

It is possible to focus on one of these by making sentences like this:
What Mary does is eat cake.
To make sentences like these, first, choose the subject whose
activity you want to emphasize, 'Mary', then place 'what' in front
of it.
We can use an adverb of location (there) to emphasize where the
subject is. This can be placed after the subject:
That cow (subject) there (adverb) gives a lot of milk.
It is possible to put the adverb of location before the subject as well,
e.g.,

What Mary. . . That there (adverb) cow (subject) gives a lot of milk.

Next, follow the subject with a form o f the 'do' verb: To make sentences like these first identify the subject, then either
place the adverb between the noun and the verb,
What Alice (subject) does ('do' verb) . . .
That cow there gives a lot of milk.
Note that the verb cannot come before the subject:
or place it between the determiner 'that' and the noun.
* What does (verb) Mary (subject) . . .
That there cow gives a lot of milk.
The * means this sentence i s ungrammatical.

Next add a singular form of the verb 'be' which agrees in tense
followed by the phrase describing the activity:

What Mary does is ('be') eat cake.

If the verb 'be' does not agree in tense with the following verb the
sentence is ungrammatical

* What Mary does is ate cake.

222 Appendix B

RULE 8

Some sentences describe transfer of a thing from one person to Appendix C

another. For example:
COMPUTERIZED INSTRUCTIONS FOR THE
John gave the book to Mary.
IMPLICIT CONDITION FAMILIARIZATION
We can tum this sentence into a question if the person receiving PHASE OF TRAINING
the object is unknown.
Screen 1
To whom did John give the book?
You will now see some sentences in English. Try to
To make sentences like these first choose a wh-word that agrees remember them. At the end of this experiment you
with the unknown indirect object you are questioning. This goes at will be given a memory test to see if you can
the front of the sentence, after the preposition. For people the wh­ remember the sentences you saw. After you see each
word is always'whom' : sentence you will be asked a question. For example,
try to remember this sentence:
John gave the book to Mary.
* To what did John give the book? Joan crashed into the wall.

The * means this sentence is ungrammatical. Press any key to continue.

Next add a form of the verb 'do' that takes the tense of the
sentence, and follow this with the subject and the main verb. Note Screen 2
that the main verb is in the infinitive form (i.e., to give), it does
not carry tense: Were the words
To whom did John (subject) give (verb) . . . crashed wall
* To whom did John (subject) gave (verb) . . .
next to each other in the sentence?
Finally, place the object after the verb: If you think they were press 'c' for correct.
If you think they weren't press 'm' for mistake.
To whom did John (subject) give (verb) the book ? (object)

Screen 3

Correct/ Incorrect

Press any key to continue.

224 Appendix C Appendix C 225

Screen 4 Screen 7

The answer was 'm' for mistake. Correct/ Incorrect

The words

crashed wall Press any key to continue.

were not next to each other in the sentence;

Screen 8
Joan crashed into the wall.
The answer was 'c' for correct.
The words
Press any key to continue.
Alice stands

Screen 5 are next to each other in the sentence;

Now practice again. Remember this sentence: Where Alice stands is on the right not on the left.

Where Alice stands is on the right not on the left.

Press any key to continue.

Screen 6 Screen 9

Were the words You will now see some sentences. You must try t o
remember them. After each sentence you must answer
Alice stands a question about the words in the sentence. You press
'c' for correct if the answer is yes, and 'm' for
next to each other in the sentence? mistake if the answer is no.

Remember, press 'c' for correct if they were. Are you ready?
Press 'm' for mistake if they weren't.

Press any key to continue.

 Appendix D

COMPUTERIZED INSTRUCTIONS FOR THE

INCIDENTAL CONDITION FAMILIARIZATION
PHASE OF TRAINING

Screen 1

You will now see some sentences in English. Try t o

understand them. At the end o f this experiment you
will be asked to answer comprehension questions o n
some of the sentences you saw. After you see each
sentence you will be asked a question. For example,
read this sentence:

Joan crashed into the wall.

Press any key to continue.

Screen 2

Did Joan crash into a car?

If the answer is 'yes', press 'c' for correct.

If the answer is 'no', press 'm' for mistake.

Screen 3

Correct/ Incorrect

Press any key to continue.

228 Appendix D
Screen 4
The answer was 'm' for mistake.
Joan crashed into the wall.
She did not crash into a car.
Press any key to continue.
Screen 5
Now practice again. Read this sentence:
Where Alice stands is on the right not on the left.
Screen 6
Does Alice stand on the right?
Remember, press 'c' for correct if answer is yes.
Press 'm' for mistake if the answer is no.
Screen 7
Correct/ Incorrect
Press any key to continue.
Appendix D 229
Screen 8
The answer was 'c' for correct.
Where Alice stands is on the right.
She does not stand on the left.
Press any key to continue.
Screen 9
You will now see some more questions. Read them and
try to understand them. After each sentence you will
be asked a question to test your understanding.You
press 'c' for correct if the answer is yes, and 'm' for
mistake if the answer is no.
Are you ready?
Press any key to continue.
 Appendix E

COMPUTERIZED INSTRUCTIONS FOR THE

RULE-SEARCH CONDITION FAMILIARIZATION
PHASE OF TRAINING

Screen 1

You will now see some sentences in English. They are

examples that illustrate two rules of English grammar.
As you see the sentences try to find the rules. At the
end of this experiment you will be asked to say what
the rules are in your own words.

Press any key to continue.

Screen 2

After you see each sentence you will

be asked a question. For example:

Have you found the rules yet?

If the answer is yes, you must press 'c' for

correct. If the answer is no, you must press 'm'
for mistake

Press any key to continue.

Screen 3

For example, study this sentence:

Joan crashed into the wall.

232 Appendix E

Screen 4 Appendix F
Does this sentence illustrate a rule that you
know? COMPUTERIZED INSTRUCTIONS FOR THE
INSTRUCTED CONDITION FAMILIARIZATION
Pres � '�' for c?rrect, or if the answer is yes, PHASE OF TRAINING
and m for mistake, if the answer is no.
L I
Screen 1
Screen 5
Here are two sentences based on the rules you studied.
Now study this sentence:
Into the wall crashed Joan.
Where Alice stands is on the right not on the left. RULE 1

Where Alice stands is on the right not on the left.

Screen 6 RULE 2

Did this sentence contain a rule that you know?
�e!11.ember, answer 'c' if the answer is yes, and
m if the answer is no.
You will now see some questions about these
sentences. Study the questions and make sure you
understand the answers. You may look at the rules
if
you want to.

If the answer to the question is yes, press 'c' for

correct . If the answer is no, press 'm' for mistake.
Screen 7

You will now see some more sentences. You must try
Press any key to continue.
to find the rules that these sentences illustrate. After
;a�h sentence yo_u must answer a question. You press
c for �orrect if the answer is yes, and 'm' for
.
mIStake if the answer is no.

Are you ready?

Press any key to continue.

234 Appendix F
Appendix F 235

Screen 2
Screen 5
RULE 1
Correct/ Incorrect
Into the wall crashed Joan.
Press any key to continue.
Study these questions and understand the answers.

1 . Was a location phrase used in the sentence?

(answer 'c', yes) Screen 6
2 . Did the subject of the sentence come after the
verb? (answer 'c', yes) The answer was 'c' for correct.
3 . Was the phrase of time or location at the front of In the sentence:
the sentence? (answer 'c', yes)
4. Was a time phrase used in the sentence? Joan crashed into the wall.
(answer 'm', no)
'into the wall' is a location phrase.
Press any key to continue.

Press any key to continue.

Screen 3

Now study this sentence: Screen 7

RULE 2 .
Joan crashed into the wall.
Where Alice stands is on the right not on the left.

Study these questions and understand the answers.

Screen 4
I. Was a contrasting location introduced by 'not'
used? (answer 'c', yes)
Was a location phrase used in the sentence?
2. Did the subject of the sentence come after the
verb? (answer 'm', no)
3 . Was the verb 'to be' in the singular form?
Answer 'c' for correct or yes.
(answer 'c', yes)
Answer 'm' for mistake or no.
4. Did the verbs used agree in tense? (answer 'c', yes)

Press any key to continue.

236 Appendix F Appendix F 23 7

Screen 8 Screen 12

Study this sentence:
Where Alice and Judy stand is on the left.
You will now see some more sentences that are
formed by the rules you have studied. Try t o
remember the rules when you see the sentences.
Remember, when you answer the questions you must
press 'c' for correct if the answer is yes, and 'm' for
mistake, if the answer is no.

Are you ready?

Screen 9

Was the verb 'to be' in the singular form? Press any key to continue.

Press 'c' for correct if the answer is yes.

Press 'm' for mistake if the answer is no.

Screen 1 0

Correct/ Incorrect

Press any key to continue.

Screen 1 1

The answer was 'c' for correct.

In the sentence:

Where Alice and Judy stand is on the left.

the verb 'to be' is in the singular form 'is'.

Press any key to continue.

 Appendix G
THE TRAINING SET SENTENCES

Sentences presented during the training sessions

Five easy rule and five hard rule grammatical sentence types were
presented, and exemplified by four tokens each during training,
making a total of forty tokens presented during each trial of
training. These correspond to the sentence types used to illustrate
the easy and hard rules in the rating scale survey, and in the rule
formats presented to subjects in the instructed condition. For hard
sentence types the optional contrasting location phrase introduced
by 'not' was present in two tokens of each of the four types t o
which it is possible to attach it, and absent from the other two
tokens.

Easy rule sentences

easy grammatical type 1

i) Paul drove across the city.
ii) Mary sank under the sea.
iii) Jill walked over the bridge.
iv)John ran into the house.

easy grammatical type 2

i) Up the road came Eric.
ii) Off the horse fell Amy.
iii)Down the hill slid Jack.
iv) Onto the boat jumped Sue.

easy grammatical type 3

i) Round the track Bill raced.
ii) Through the forest Jane went.
iii)Past the river Helen rode.
iv )Out of the shop Jim rushed.

easy grammatical type 4

i) Lori ate at five'o clock.
ii) Ron left on Thursday night.
240 Appendix G

! ii)Roland returned at midnight.

1v)Peggy slept on Tuesday morning.
Appendix H
�asy grammatical type 5
1) In the afternoon Elaine arrived.
ii) At the weekend Randy exercised. TRAINING SET SENTENCES AND FOLLOW-UP
!ii) On Christmas day Lily phoned. QUESTIONS VIEWED IN EACH CONDITION
1v)In the morning Claire jogged.
Tokens of each easy and hard rule sentence type are followed by the
Hard rule sentences questions which followed them in each condition. (Imp = implicit;
Inst = instructed; Rule-s = rule-search; Inc = incidental condition).
hard grammatical type 1
!! The girl pl_ay� in the park and the boy plays in the yard. easy grammatical · type 1
��! The letter 1s m the bag and the book is on the chair.
� n)John works in the shop and Mary works in the hospital . i) Paul drove across the city.
1v) The milk is in the glass and the coffee is in the jar. Imp Were the words Paul across next to each other in the
sentence?
hard grammatical type 2 Inst Was a time phrase used in the sentence?
i) Where Helen is is in New York. Rule-s Did you notice any rules?
!!! Where the phone is is in the bedroom. Inc Was Paul travelling by sea?
� n) Where Sue is is in the car not in the boat.
iv) Where the cheese is is on the plate not in the basket. ii) Mary sank under the sea.
Imp Were the words sank sea next to each other in the
hard grammatical type 3 sentence?
i) Where the children are is on the beach . Inst Was a location phrase used in the sentence?
ii) Where Eric and Jill are is at home. Rule-s Does this sentence contain a rule that you know?
!ii) Where Jim and Amy a:e is in the kitchen not in the garden. Inc Was Mary at home?
1v) Where the people are is in the park not in the church.
iii) Jill walked over the bridge.
hard grammatical type 4 Imp Were the words walked over next to each other in the
i) Where the cat sleeps is on the floor. sentence?
ii) Where Bill eats is at the table. Inst Did the subject of the sentence come after the verb?
!ii) Where the President lives is in Washington not in Boston. Rule-s Did you notice the rule?
iv) Where Jane sits is on the sofa not on the stool. Inc Was Jill taking a walk?
hard grammatical type 5 iv) John ran into the house .
i) Where the birds nest is in the tree. Imp Were the words the mouse next to each other in the
!!! Where Paul and Jack surf is at Waikiki. sentence?
� n) Where Steve and Mark swim is in the pool not in the ocean. Inst Was the phrase of time or location at the front of the
Iv) Where the students meet is in the library not in the sentence?
restaurant. Rule-s Did this sentence contain a rule that you know?
Inc Was John driving a car?
 Appendix H 243
242 Appendix H

easy grammatical type 2 Imp Were the words through Jane next to each other in the
sentence?
i) Up the road came Eric. Inst Was a location phrase used in the sentence?
Imp Were the words road came next to each other in the Rule-s Have you found the rules yet?
sentence? Inc Was Jane watching a movie?
Inst Was a time phrase used in the sentence?
Rule-s Have you found any rules yet? iii) Past the river Helen rode.
Inc Was Eric travelling by plane? Imp Were the words Helen rode next to each other in the
sentence?
ii) Off the horse fell Amy. Inst Did the subject of the sentence come after the verb?
Imp Were the words horse Amy next to each other in the Rule-s Did this sentence contain a rule that you know?
sentence? Inc Did Helen fall into the river?
Inst Was a location phrase used in the sentence?
Rule-s Did this sentence contain a rule that you know? iv) Out of the shop Jim rushed.
Inc Was Amy riding a camel? Imp Were the words shop rushed next to each other in the
sentence?
iii) Down the hill slid Jack. Inst Was the phrase of time or location at the front of the
Imp Were the words Down the next to each other in the sentence?
sentence? Rule-s Are you still looking for the rules?
Inst Did the subject of the sentence come after the verb? Inc Was Jim in a hurry?
Rule-s Are you still looking for the rules?
easy grammatical type 4
Inc Was Jack in the classroom?

iv) Onto the boat jumped Sue. i) Lori ate at five' o clock.
Imp Were the words onto boat next to each other in the Imp Were the words Lori at next to each other in the
sentence? sentence ?
Inst Was the phrase of time or location at the front of the Inst Was a time phrase used in the sentence?
sentence? Rule-s Did you notice any rules?
Rule-s Does this sentence contain a rule that you know? Inc Was Lori eating her lunch?
Inc Did Sue get onto the boat?
ii) Ron left on Thursday night. .
easy grammatical type 3 Imp Were the words left night next to each other m the
sentence?
i) Round the track Bill raced. Inst Ws a location phrase used in the sentence?
Imp Were the words track raced next to each other in the Rule-s Have you found the rules yet?
sentence? Inc Did Ron stay until Friday?
Inst Was a time phrase used in the sentence?
Rule-s Did you notice any rules? iii) Roland returned at midnight.
Inc Was Bill studying at home? Imp Were the words returned at next to each other in the
sentence?
ii) Through the forest Jane went. Inst Did the subject of the sentence come after the verb?
 244 Appendix H Appendix H 245

Rule-s Did this sentence contain a rule that you know? hard grammatical type 1
Inc Was Roland asleep by midnight?
i) The girl plays in the park and the boy plays in the yard.
iv) Peggy slept on Tuesday morning. Imp Were the words plays the next to each other in the
Imp Were the words Peggy morn ing next to each other in sentence?
the sentence? Inst Did the subject of the sentence come after the verb?
Inst Was the phrase of time or location at the front of the Rule-s Did you notice any rules?
sentence? Inc Were the boy and girl playing together?
Rule-s Do you know the rules for sentences like this?
Inc Was Peggy tired? ii) The letter is in the bag and the book is on the chair.
Imp Were the words book on next to each other in the
easy grammatical type 5
sentence?
Inst Was the verb 'to be' in the singular form?
i) In the afternoon Elaine arrived. Does this sentence contain a rule that you know?
Rule-s
Imp Were the words in afternoon next to each other in the Is the letter on the chair?
Inc
sentence?
Inst Was a time phrase used in the sentence? iii) John works in the shop and Mary works in the hospital.
Rule-s Have you found the rules yet? Imp Were the words Mary works next to each other in the
Inc Did Elaine arrive at night? sentence?
Inst Did the verbs used agree in tense?
ii) At the weekend Randy exercised. Rule-s Did you notice the rule?
Imp Were the words the Randy next to each other in the Inc Is John a doctor?
sentence?
Inst Was a location phrase used in the sentence? iv) The milk is in the glass and the coffee is in the jar.
Rule-s Does this sentence contain a rule that you know? Imp Were the words coffee cup next to each other in the
Inc Did Randy exercise every day? sentence?
Inst Was a contrasting location introduced by 'not' used ?
iii) On Christmas day Lily phoned. Rule-s Did this sentence contain a rule that you know?
Imp Were the words Lily phoned next to each other in the Is there any coffee left?
Inc
sentence?
Inst Did the subject of the sentence come after the verb? hard grammatical type 2
Rule-s Have you found the rules yet?
Inc Did Lily forget to call on Christmas day? i) Where Helen is is in New York.
Imp Were the words Helen is next to each other in the
iv) In the morning Claire jogged. sentence?
Imp Were the wordson the next to each other in the sentence? Inst Was a contrasting location introduced by 'not' used?
Inst Was the phrase of time or location at the front of the Rule-s Have you found any rules yet?
sentence? Inc Is Helen in Honolulu?
Rule-s Do you know the rules for sentences like this?
Inc Did Claire exercise in the morning? ii) Where the phone is is in the bedroom.
Imp Were the words where phone next to each other in the
246 Appendix H Appendix H 24 7

sentence? iv) Where the people are is in the park not in the c.hurch.
Inst Did the subject of the sentence come after the verb? Imp Were the words are in next to each other m the sentence?
Rule-s Did this sentence contain a rule that you know? Inst Did the verbs used agree in tense?
Inc Is there a phone in the bedroom? Rule..:s Are you still looking for the rules?
Inc Are there any people in the park?
iii) Where Sue is is in the car not in the boat.
Imp Were the words is the next to each other in the sentence? hard grammatical type 4
Inst Was the verb 'to be' in the singular form?
Rule-s Are you still looking for the rules? i) Where the cat sleeps is on the floor.
Inc Is Sue in the boat? Imp Were the words sleeps on next to each other in the
sentence?
iv) Where the cheese is is on the plate not in the basket. Inst Did the verbs used agree in tense?
Imp Were the words plate basket next to each other in the Rule-s Did you notice any rules?
sentence? Inc Does the cat sleep on the floor?
Inst Did the verbs used agree in tense?
Rule-s Do you know the rules for sentences like this? ii) Where Bill eats is at the table.
Inc Is the cheese in the basket? Imp Were the words Bill is next to each other in the sentence?
Inst Was the verb 'to be' in the singular form?
hard grammatical type 3 Rule-s Have you found the rules yet?
Inc Does Bill eat off the floor?
i) Where the children are is on the beach.
Imp Were the words are on next to each other in the iii) Where the President lives is in Washington not in Boston.
sentence? Imp Were the words not Boston next to each other in the
Inst Was the verb 'to be' in the singular form ? sentence?
Rule-s Did you notice any rules? Inst Did the subject of the sentence come after the verb?
Inc Are the children in bed? Rule-s Did this sentence contain a rule that you know?
Inc Does the President live in New York?
ii) Where Eric and Jill are is at home.
Imp Were the words at home next to each other in the iv) Where Jane sits is on the sofa not on the stool. .
sentence? Imp Were the words Jane sits next to each other m the
Inst Did the subject of the sentence come after the verb? sentence?
Rule-s Have you found the rules yet? Inst Was a contrasting location introduced by 'not' used?
Inc Is Eric alone in the house? Rule-s Do you know the rules for sentences like this?
Inc Does Jane sit on the sofa?
iii) Where Jim and Amy are is in the kitchen not in the garden.
Imp Were the words garden not next to each other in the hard grammatical type 5
sentence?
Inst Was a contrasting location introduced by 'not' used? i) Where the birds nest is in the tree. .
Rule-s Did this sentence contain a rule that you know? Imp Were the words birds nest next to each other m the
Inc Are Jim and Amy swimming? sentence?
Inst Was a contrasting location introduced by 'not' used?
248 Appendix H

Rule-s Have you found the rules yet?

Inc Do the birds nest in the garage?
Appendix I
ii) Where Paul and Jack surf is at Waikiki.
Imp Were the words Paul surf next to each other in the
COMPUTERIZED INSTRUCTIONS FOR THE
sentence?
Inst Did the subject of the sentence come after the verb? FAMILIARIZATION PHASE OF THE TRANSFER
Rule-s Do you know the rules for sentences like this? TASK
Inc Do Paul and Jack surf at Ala Moana?
Screen 1
iii) Where Steve and Mark swim is in the pool not in the ocean.
Imp Were the words where Mark next to each other in the You will now see some more sentences. You must
sentence? decide if they are grammatical or ungrammatical. I f
Inst Was the verb 'to be' in the singular form? you think the sentence is grammati�al press 'c' for
Rule-s Have you found the rules yet? _
correct. If you think the sentence is ungrammatical
Inc Does Steve swim in the pool? press 'm' for mistake.

iv) Where the students meet is in the library not in the restaurant. Press any key to continue.
Imp Were the words meet library next to each other in the
sentence?
Inst Did the verbs used agree in tense? Screen 2
Rule-s Do you know the rules for sentences like this?
Inc Do the students meet in the restaurant?
For example, read the sentence bel�w.
Press 'c' if you think it is grammatical:
Press 'm' if you think it is ungrammatical.
Do this as quickly as you can.

The tall woman ate the pizza.

Screen 3

The answer was 'c' for correct.

Now try to work as quickly as you can.

When you see the sentences press 'c' if theJ'. are
grammatical and 'm' if they are ungrammatical.

Are you ready?

Press any key to continue.

 Appendix J

TRANSFER SET SENTENCES

Sentences presented during the transfer sessions

The forty transfer set sentences included two new tokens of the
easy and hard grammatical types which were presented during the
training session, plus five tokens of easy ungrammatical types 1 and
2 and two tokens of hard ungrammatical types 1-5 . The
ungrammatical sentence types were violations of the conditions on
the rules presented during the rating scale survey, and presented to
learners during the instructed condition.

easy grammatical type 1

i) Craig walked into the room.
ii) Lisa fell onto the grass.
easy grammatical type 2
i) Across the street raced Tom.
ii) Out of the door ran Kate.
easy grammatical type 3
i) Through the window Maria stared.
ii) Over the town Lucy flew.
easy grammatical type 4
i) Peter woke at eight' o clock.
ii) Wendy started on Monday morning.
easy grammatical type 5
i) On her birthday Gabi sang.
ii) In the evening Dick studied.
easy ungrammatical type 1
i) On Saturday night danced Charley.
ii) At nine thirty left Robert.
iii) In the afternoon spoke Ian.
iv) On Sunday painted Graham.
v) At night ate Judy.
easy ungrammatical type 2
i) Golfed Debbie in the morning.
ii) Died Stan on Friday night.
iii) Departed Fred at nine thirty.
iv) Crawled Mick through the tunnel.
 252 Appendix J

v) Slipped Maggy off the roof.

hard grammatical type 1
i) The pen is in the box and the pencil is in the drawer.
ii) Lisa reads in the bedroom and Mick reads in the study. Appendix K
hard grammatical type 2
i) Where L.A is is in California. SUBJECT INFORMATION
ii) Where the car is is in the driveway not in the road.
hard grammatical type 3
i) Where the apples are is in the bowl. Table 95. Details of Subj ects For The Present Study
ii) Where Ann and Ian are is at the movies not at the zoo.
hard gramm atical type 4
Subject Native Native Gender Age Condition
i) Where Judy teaches is at the univers ity.
No. Country Language
ii) Where Peter waits is by the stairs not in the cafe.
hard grammatical type 5
i) Where my parents vacation is in Europe. 1 Japan Japanese Female 23 Implicit
ii) Where Maria and Joe relax is by the pool not on the beach. 2 Japan Japanese Female 25 Incidental
hard ungram matical type 1 3 Japan Japanese Female 22 Implicit
i) Where is Bill is in the country not in the city. 4 Japan Japanese Female 34 Incidental
ii) Where cooks Jim is in the kitchen not in the bathroom. 5 Japan Japanese Female 20 Implicit
hard ungram matical type 2 6 Japan Japanese Female 26 Instructed
i) Where Carl writes are at a desk not on the floor. 7 Japan Japanese Female 31 Rule-search
ii) Where the soldiers fight are in Europe not in Hawai'i . 8 Japan Japanese Female 26 Rule-search
hard ungram matical type 3 9 China Mandarin Female 29 Incidental
i) Where the bird is was in the sky not in the sea. 10 Japan Japanese Female 21 Implicit
ii) Where the plane lands was in the airport not at the dock. 11 Japan Japanese Female 28 Instructed
hard ungram matical type 4 12 Japan Japanese Female 23 Instructed
i) What the King lives is in a palace not in an apartment. 13 Japan Japanese Female 31 Instructed
ii) What the dog is is in the yard not in the house. 14 Korea Korean Male 23 Instructed
hard ungram matical type 5 15 China Mandarin Male 22 Incidental
i) Where the horse stands is in the field in the barn. 16 Japan Japanese Female 19 Implicit
ii) Where Mark works is in America in Japan. 17 Korea Korean Female 21 Incidental
18 India Urdu Female 24 Rule-search
19 Japan Japanese Female 21 Implicit
20 Korea Korean Female 22 Instructed
21 Japan Japanese Female 21 Instructed
22 Japan Japanese Male 23 Implicit
23 Japan Japanese Female 22 Rule-search
24 Japan Japanese Female 22 Incidental
25 Japan Japanese Female 23 Instructed
26 Japan Japanese Female 22 Rule-search
27 Japan Japanese Female 23 Rule-search
28 Japan Japanese Female 24 Incidental
29 Japan Japanese Female 23 Rule-search
254 Appendix K Appendix K 255

Table 95. (Continued) Details of Subj ects For The Present Table 95. (Continued) D etails of Subjects For The Present
Study Study

Subject Native Native Gender Age Condition Subject Native Native Gender Age Condition
No. Country Language No. Country Language
30 Japan Japanese Male 23 Implicit 64 Japan Japanese Female 19 Incidental
31 Japan Japanese Female 25 Rule-search 65 Japan Japanese Female 19 Implicit
32 Japan Japanese Female 24 Incidental 66 Japan Japanese Female 19 Rule-search
33 China Mandarin Female 29 Instructed 67 Japan Japanese Female 19 Rule-search
34 China Mandarin Female 22 Implicit 68 Japan Japanese Male 19 Instructed
35 Japan Japanese Female 23 Incidental 69 Japan Japanese Male 19 Incidental
36 Japan Japanese Female 24 Instructed 70 Japan Japanese Male 19 Rule-search
37 Japan Japanese Male 24 Instructed 71 Japan Japanese Female 19 Incidental
38 Japan Japanese Male 24 Incidental 72 Japan Japanese Female 19 Incidental
39 Japan Japanese Female 25 Implicit 73 Japan Japanese Female 19 Incidental
40 Japan Japanese Male 26 Rule-search 74 Japan Japanese Female 19 Incidental
41 Japan Japanese Male 26 Rule-search 75 Japan Japanese Female 19 Rule-search
42 Japan Japanese Male 25 Implicit 76 Japan Japanese Female 19 Rule-search
43 Japan Japanese Female 25 Incidental 77 Japan Japanese Female 19 Rule-search
44 Japan Japanese Female 24 Incidental 78 Japan Japanese Female 20 Implicit
45 Japan Japanese Female 24 Incidental 79 Japan Japanese Female 19 Implicit
46 Japan Japanese Female 23 Instructed 80 Japan Japanese Female 19 Instructed
47 Japan Japanese Female 25 Instructed 81 Japan Japanese Male 19 Instructed
48 Korea Korean Male 26 Implicit 82 Japan Japanese Male 19 Instructed
49 China Mandarin Female 27 Implicit 83 Japan Japanese Male 19 Instructed
50 Japan Japanese Male 25 Implicit 84 Japan Japanese Male 19 Instructed
51 Japan Japanese Male 24 Incidental 85 Japan Japanese Female 20 Implicit
52 Japan Japanese Female 22 Incidental 86 Japan Japanese Male 19 Implicit
53 Japan Japanese Female 25 Instructed 87 Japan Japanese Male 19 Implicit
54 Japan Japanese Female 30 Implicit 88 Japan Japanese Male 19 Implicit
55 Japan Japanese Female 27 Implicit 89 Japan Japanese Female 19 Incidental
56 Korea Korean Female 24 Incidental 90 Japan Japanese Female 19 Incidental
57 Japan Japanese Male 29 Incidental 91 Japan Japanese Female 19 Incidental
58 Japan Japanese Female 19 Rule-search 92 Japan Japanese Male 19 Rule-search
59 Japan Japanese Female 19 Rule-search 93 Japan Japanese Female 19 Rule-search
60 Japan Japanese Female 19 Rule-search 94 Japan Japanese Male 19 Rule-search
61 Japan Japanese Female 19 Instructed 95 Japan Japanese Male 20 Rule-search
62 Japan Japanese Female 19 Implicit 96 Japan Japanese Male 19 Rule-search
63 Japan Japanese Female 19 Implicit 97 Japan Japanese Male 19 Rule-search
256 Appendix K

Table 95. (Continued) Details of Subj ects For The Present

Study Bibliography

Subject Native Native Gender Age Abrams, M. ( 1 987). Implicit learning in the psychiatrically
Condition
No. Country Language impaired. Unpublished doctoral dissertation, City University
of New York.
98 Japan Abrams, M., & Reber, A. ( 1 98 8). Implicit learning: Robustness in
Japanese Male 19 Rule-search
99 Japan the face of psychiatric disorders. Journal of Psycholinguistic
Japanese Female 19 Instructed
1 00 Japan Research, 1 7, 425-43 9.
Japanese Female 19 Instructed
101 Japan Ackerman, P.L. ( 1 988). Determinants of individual differences
Japanese Female 19 Instructed
1 02 Japan during skill acquisition: Cognitive abilities and information
Japanese Male 19 Instructed
1 03 China processing. Journal of Experimental Psychology: General,
Mandarin Female 36 Implicit 11 7, 2 8 8-3 1 8 .
1 04 Korea Korean Female 26 Instructed Adams, F., Aizawa, K., & Fuller, G. ( 1 992). Rules in programming
languages and networks. In J. Dinsmore (Ed.), The symbolic
and connectionist paradigms: Closing the gap, pp.49-6 7 .
Hillsdale, NJ: Lawrence Erlbaum.
Alanen, R. ( 1 995). Input enhancement and rule presentation in
second language acquisition. In R. Schmidt (Ed.), A ttention
and awareness in foreign language teaching and learning,
pp. 5 3 - 8 1 . University of Hawai' i at Manoa: Second Language
Teaching and Curriculum Center Technical Report # 9.
Alkon, D. ( 1 988). Memory traces in the brain. Cambridge:
Cambridge University Press.
Allwright, R.L. ( 1 984). Why don't learners learn what teachers
teach? In D. Singleton & D. Little (Eds.), Language learning
in formal and informal contexts, pp. 1-1 8. Dublin: IRAAL.
Anderson, J.R. ( 1 983). The architecture of cognition. Cambridge,
MA: Harvard University Press.
Ashby, F.G. ( 1 992). Multidimensional models of categorization. In
F.G. Ashby (Ed.), Multidimensional models of perception and
categorization, pp.449-485. Hillsdale, NJ: Lawrence Erlbauni.
Asher, J.J. ( 1 979). Motivating children and adults to acquire a
second language. SPEAQ Journal, 3, 87-99.
Atkinson, M. ( 1 992). Children 's syntax: An introduction to
principles and parameters theory. Cambridge, MA: Basil
Blackwell.
Atkinson, R., & Shiffrin, R. ( 1 968). Human memory: A proposed
system and its control processes. In K. Spence (Ed.), The
psychology of learning and motivation, Volume. 2, pp . 8 9-
1 95. New York: Academic Press.
258 Bibliography
Baars, B.J. ( 1 988). A cognitive theory of consciousness. Cambridge,
MA: Cambridge University Press.
Baddeley, A. ( 1 986). Working memory. Oxford: Oxford University
Press.
Baddeley, A., & Hitch, G. ( 1 974). Working memory. In G. Bower
(Ed.), Recent advances in learning and motivation, pp.47-90.
New York: Academic Press.
__ . ( 1 993 ). The recency effect: Implicit learning with explicit
retrieval? Memory & Cognition, 21, 1 46- 1 5 5 .
Baker, C.L. ( 1 989). English syntax. Cambridge, MA: MIT Press.
Balota, D.A. ( 1 983). Automatic semantic activation and episodic
memory encoding. Journal of Verbal Learning and Verbal
Behavior, 22, 88- 1 04.
Bardovi-Harlig, K. ( 1 987). Markedness and salience in second
language acquisition. Language Learning, 3 7, 3 85-407.
Beretta, A. ( 1 989). Attention to form or meaning: Error treatment
in the Bangalore Project. TESOL Quarterly, 23, 2 83-3 03 .
Besson, M., Boaz, T., Fischler, I. & Raney, G. ( 1 992). Effects o f
automatic associative activation on explicit and implicit
memory tests. Journal of Experimental Psychology:
Learning, Memory and Cognition, 18, 89-1 0 5 .
Best, J.B. ( 1 992). Cognitive psychology. (3rd edition) New York:
West Publishing.
Bialystok, E. ( 1 978). A theoretical model of second language
learning. Language Learning, 28, 69-84.
. ( 1 979). Explicit and implicit judgments of L2
-- grammaticality. Language Learning, 29, 8 1 - 1 03 .
__ . ( 1 982). On the relationship between knowing and using
linguistic forms. Applied Linguistics, 3, 25 5-262 .
-- . ( 1 988). Psycholinguistic dimensions of second language
proficiency. In W. Rutherford & W. Sharwood Smith (Eds.),
Grammar and second language teaching, pp. 1 9-4 1 . Rowley,
MA: Newbury House.
. ( 1 990). The dangers of dichotomy: A reply to Hulstijn.
-- Applied Linguistics, 1 1, 46-5 1 .
Bialystok, E., & Sharwood Smith, M. ( 1 985). Interlanguage is not a
state of mind. Applied Linguistics, 6, 1 0 1 -1 1 7 .
Bickerton, D. ( 1 99 1 ). Consciousness I and consciousness II.
University of Oregon Lectures on the Philosophy of Mind.
Unpublished manuscript, Department of Linguistics,
University of Hawai'i at Manoa.
Bibliography 259
Blaxton, T.A. ( 1 989). Investigating dissociations among memory
measures: Support for a transfer appropriate processing
framework. Journal of Experimental Psychology: Learning,
Memory and Cognition, 15, 6 5 7-68 8 .
Bley-Vroman, R . ( 1 986). Hypothesis testing in second language
acquisition theory. Language Learning, 36, 3 5 3-3 76 .
. ( 1 989). What is the logical problem of foreign language
-- learning? In S. Gass & J. Schacter (Eds.), Linguistic
perspectives on second language acquisition, pp.4 1 -6 8 . New
York: Cambridge University Press.
Bley-Vroman, R., Felix, S., & Ioup, G. ( 1 988). The accessibility of
Universal Grammar in adult language learning. Second
Language Research, 4, 1-3 2 .
Bowers, K.S., & Meichenbaum, D . (Eds.) ( 1 984). The unconscious
reconsidered. New York: John Wiley.
Braine, M.D.S. ( 1 965). The insufficiency of a finite-state model for
verbal reconstructive memory. Psychonomic Science, 2, 2 9 1 -
292 .
Braine, M.D.S., Brody, R.E., Brooks, P., Sudhalter, V., Ross, J.A.,
Catalano, L., & Fisch, S .M. ( 1 990). Exploring language
acquisition in children with a miniature artificial language:
Effects of item and pattern frequency, arbitrary subclasses and
correction. Journal of Memory and Language, 29, 5 9 1 -6 1 0.
Bresnan, J., & Kanerva, J.M. ( 1 989). Locative inversion in
Chichewa: A comparative study of factorization in grammar.
Linguistic Inquiry, 20, 1 -5 0 .
Brewer, W.F. ( 1 974). There i s no convincing evidence for operant
or classical conditioning in adult humans. In W.B. Weimer &
D.S. Palermo (Eds.), Cognition and the symbolic processes,
pp. 1-42. Hillsdale, NJ: Lawrence Erlbaum.
Broadbent, D. ( 1 954). A mathematical model for human attention
and immediate memory. Psychological Review, 64, 205 .
. ( 1 95 8). Perception and communication New York:
Pergamon.
Broadbent, D., Fitzgerald, P., & Broadbent, M.H.P. ( 1 986). Implicit
and explicit knowledge in the control of complex systems.
British Journal of Psychology, 77, 33-5 0 .
Brody, N . ( 1 989). Unconscious learning o f rules: comment o n
Reber' s analysis of implicit learning. Journal of Experimental
Psychology: General, 118, 236-23 8 .
·Bunge, M . ( 1 973). Model, method, matter. Dordrecht: Reidel.
260 Bibliography
Carlson, R.A., Khoo, B.H., Yaure, R.G., & Schneider, W. ( 1 990).
Acquisition of a problem solving skill: Levels of organization
and use of working memory. Journal of Experimental
Psychology: General, 1 1 9, 1 93-2 1 4.
Carr, T.H., & Curran, T. ( 1 994). Cognitive factors in learning about
structured sequences: Applications to syntax. Studies in
Second Language Acquisition, 1 6, 205-22 5 .
Carrell, P. ( 1 984). Schema theory and ESL reading pedagogy.
TESOL Quarterly, 18, 1 52- 1 69 .
Carroll, J.B . ( 1 973). Implications of aptitude test research and
psycholinguistic theory for foreign language teaching.
International Journal of Psycholinguistics, 2, 5- 1 4 .
__ . ( 1 98 1 ). Twenty five years o f research o n foreign language
aptitude. In K. Diller (Ed.), Individual differences and
universals in language learning aptitude, pp.83-1 1 8 . Rowley,
MA: Newbury House.
__ . ( 1 990). Cognitive abilities in foreign language aptitude:
Then and now. In C. Stansfield (Eds.), Language aptitude
reconsidered, pp. 1 1-30 . Englewood Cliffs, NJ: Prentice Hall.
__ . ( 1 993). Cognitive abilities: A factor analytic approach to
research. Oxford: Oxford University Press.
Carroll, J.B., & Sapon, S.M. ( 1 959). The modern language aptitude
test. San Antonio, TX: The Psychological Corporation.
Ceci, S.J., & Liker, J.K. ( 1 986). A day at the races: A study of IQ,
expertise and cognitive complexity. Journal of Experimental
Psychology: General, 1 15, 25 5-266.
Celce-Murcia, M., & Larsen-Freeman, D. ( 1 983). The grammar
book. Rowley, MA: Newbury House.
Chandler, S. ( 1 993). Are rules and modules really necessary for
explaining language? Journal of Psycholinguistic Research,
22, 5 93-606.
Chapelle, C., & Green, P. ( 1 992). Field dependence/independence in
second language acquisition research. Language Learning, 42,
47-8 3 .
Chapelle, C., & Roberts, C . ( 1 986). Ambiguity tolerance and field
independence as predictors of proficiency in English as a
second language. Language Learning, 36, 27-45.
Cherry, E.C. ( 1 953). Some experiments on the recognition of
speech with one and two ears. Journal of the Acoustical
Society of America, 25, 975-979.
Chi, M., Glaser, R., & Rees, E. ( 1 982). Expertise in problem
solving. In R. J. Sternberg (Ed.), Advances in the psychology
Bibliography 261
of human intelligence, Volume 1, pp.7-76 . Hillsdale, NJ:
Lawrence Erlbaum.
Chomsky, N. ( 1 959). Review of Skinner's 'Verbal Behavior. '
Language, 35, 26-5 8 .
__ . ( 1 965). Aspects of the theory of syntax. Cambridge, MA:
MIT Press.
__ . ( 1 9 8 1 ). Lectures on government and binding. Dordrecht:
Foris.
__ . ( 1 986). Knowledge of language: Its nature origin and use.
New York: Praeger.
__ . ( 1 990). Accessibility "in principle." Behavioral and Brain
Sciences, 13, 600-60 1 .
Cohen, A., Larsen-Freeman, D., & Tarone, E. ( 1 9 9 1 ). The
contribution of SLA theories and research to teaching
language. In E. Sadtono (Ed.), Language acquisition and the
second/foreign language classroom. pp. 1 3 6-1 56 . Singapore:
RELC.
Cohen, M.M., & Massaro, D. ( 1 992). On the similarity of
categorization models. In F.G. Ashby (Ed.), Multidimensional
models of perception and categorization, pp.3 95-449.
Hillsdale, NJ: Lawrence Erlbaum.
Cohen, N.J. ( 1 984). Preserved learning capacity in amnesia:
Evidence for multiple systems. In L. Squire & N. Butters
(Eds.), Neuropsychology of memory, pp. 83- 1 03 . New York :
Guildford Press.
Cook, V. ( 1 977). Cognitive processes in second language learning.
International Review of Applied Linguistics, 15, 1 -20.
__ . ( 1 979). Aspects of memory in secondary school language
learners. Interlanguage Studies Bulletin, 4, 1 6 1- 1 72.
__ . ( 1 986). Do second language learners have a cognitive
deficit?. In V. Cook (Ed.), Experimental approaches to
second language acquisition, pp. 73-79. Oxford: Pergamon
Press.
__ . ( 1 993). Linguistics and second language acquisition.
London: Longmans.
Corder, S.P. ( 1 988). Pedagogic grammars. In W. Rutherford & M.
Sharwood Smith (Eds.), Grammar and second language
teaching, pp. 1 23-145. New York: Newbury House.
Cotman, C.W., & Lynch, G.S. ( 1 989). The neurobiology of learning
and memory. Cognition, 33, 2 0 1 -24 1 .
Cowan, N. ( 1 988). Evolving conceptions of memory storage,
selective attention and their mtual constraints within the
 Bibliography 263
262 Bibliography

doctoral dissertation, University of Pennsylvania, Ann Arbor:

human information processing system. Psychological
University Microfilms International no. 8 8 1 6 1 2 1 .
Bulletin, 1 04, 1 63-1 9 1 .
. ( 1 99 1 ). Second language instruction does make a difference:
. ( 1 993). Activation, attention, and short-term memory.
__

Evidence from an empirical study of relativization. Studies in

__

Memory and Cognition, 2 1, 1 62-1 67.

Second Language Acquisition, 13, 43 1-469.
Craik, F.K., & Lockhart, R.S. ( 1 972). Levels of processing: A
Dulany, D., Carlson, R., & Dewey, G. ( 1 9 84). A case of syntactical
framework for memory research. Journal of Verbal Learning
learning and judgment: how conscious and how abstract?
and Verbal Behavior, 1 1, 671-684.
Journal of Experimental Psychology: General, 1 13, 541-5 5 5 .
Crick, F. ( 1 995). The astonishing hypothesis: The scientific search
. ( 1 985). On consciousness in syntactic learning and
for the soul. New York: Simon & Schuster.
__

judgment: A reply to Reber, Allen and Regan. Journal of

Crowder, R.G. ( 1 993). Short-term memory: Where do we stand?
Experimental Psychology: General, 1 1 4, 25-3 2 .
Memory and Cognition, 2 1, 1 42- 1 4 5 .
Dupuy, B . , & Krashen, S . ( 1 993). Incidental vocabulary acquisition
Curran, T., & Keele, S.W. ( 1 993). Attentional and nonattentional
in French as a foreign language. Applied Language Learning,
forms of sequence learning. Journal of Experimental
4, 5 5-63 .
Psychology: Learning, Memory and Cognition, 19, 1 89-202.
Eckman, F. ( 1 977). Markedness and the contrastive analysis
Daneman, M., & Carpenter, P.A. ( 1 983). Individual differences in
hypothesis. Language Learning, 2 7, 3 1 5-330.
integrating information between and within sentences.
. ( 1 985). Some theoretical and pedagogical implications o f
Journal of Experimental Psychology: Learning, Memory and
__

the markedness differential hypothesis. Studies in Second

Cognition, 9, 5 6 1-5 83 .
Language Acquisition, 7, 289-3 07.
Dawson, M.E., & Schell, A.M. ( 1 982). Electrodermal reponses to
( 1 99 1 ). The structural conformity hypothesis and the
a.tten�ed and unattended significant stimuli during dichotic
__ .

acquisition of consonant clusters in the interlanguage of ESL

hstenmg. Journal of Experimental Psychology: Human
learners. Studies in Second Language Acquisition, 13, 23-4 1 .
Perception & Performance, 8, 3 1 5-324.
Eckman, F., Bell, L. & Nelson, D . ( 1 988). On the generalization o f
Day, E., & Shapson, S. ( 1 99 1 ). Integrating formal and functional
relative clause instruction in the acquisition of English as a
approaches in language teaching in French immersion: An
second language. Applied Linguistics, 9, l -20.
experimental study. Language Learning, 41, 25-58.
Elley, W.B. ( 1 99 1 ). Acquiring literacy in a second language: The
Declerck, R. ( 1 994). The taxonomy and interpretation of clefts and
effect of book-based programs. Language Learning, 41, 3 7 5-
pseudo-clefts. Lingua, 93, 1 83-220.
411.
DeGraaff, R. ( 1 995). The experanto experiment; Acquiring the rules
of an artificial language. Unpublished manuscript' Free
Ellis, N. ( 1 993 ). Rules and instances in foreign language learning:
interactions of explicit and implicit knowledge. European
University of Amsterdam.
Journal of Cognitive Psychology, 5, 289-3 1 8.
DeKeyser, R. ( 1 994). Implicit and explicit learning of L2 grammar:
A pilot study. TESOL Quarterly, 28, 1 88-1 94. __ . ( 1 994). Implicit and explicit learning: An overview. In N .
. ( 1 995). Learning second language grammar rules: An Ellis (Ed.), Implicit and explicit learning of language, pp. 1 -
1 6. London: Academic Press.
__

experiment with a miniature linguistic system. Studies in

Second Language Acquisition, 1 7, 379-4 1 0.
Dennett, D. ( 1 99 1 ). Consciousness explained. Boston, MA : Little
Brown.
Dienes, Z. ( 1 992). Connectionist amd memory-array models of
artificial grammar learning. Cognitive Science, 1 6, 4 1-79.
Doughty, C. ( 1 988). The effect of instruction on the acquisition of
relativization in English as a second language. Unpublished
Ellis, R. ( 1 9 89). Are classroom and naturalistic acquisition the
same? A study of the acquisition of German word order rules.
Studies in Second Language Acquisition, 1 1, 305-328.
__ . ( 1 990). Instructed second language acquisition. Oxford:
Basil Blackwell.
__ . ( 1 992). Second language acquisition and language
pedagogy. Oxford: Pergamon Press.
264 Bibliography
__ . (l 993 a). The structural syllabus and second language
acquisition. TESOL Quarterly, 2 7, 9 1-1 1 3 .
__ . ( 1 993b). Second language acquisition research: How does it
help teachers? ELT Journal, 4 7, 3-1 1 .
__ . ( l 994a). Implicit knowledge and L2 pedagogy. TESOL
Quarterly, 28, 1 67-1 73 .
__ . ( 1 994b ). The study of second language acquisition. Oxford:
Oxford University Press.
__ . ( l 994c). A theory of instructed second language acquisition.
In N. Ellis (Ed.), Implicit and explicit learning of language,
pp.79-1 1 4 . London: Academic Press.
__ . ( 1 995). Interpretation tasks for grammar teaching. TESOL
Quarterly, 29, 87- 1 0 5 .
Elman, J . ( 1 988). Finding structure in time. CRL Technical Report
no. 880 1 . Center for Research on Language, University of
California, San Diego.
Ericsson, K.A., & Simon, H. ( 1 980). Verbal reports as data.
Psychological Review, 8 7, 2 1 5-25 1 .
Estes, W.K. ( 1 986). Array models for category learning. Cognitive
Psychology, 18, 500-549.
__ . ( 1 992). Mental psychophysics of categorization and
decision making. In H. Geissler, S. Link & J.T. Townsend
(Eds.), Cognition, information processing and psychophysics:
Basic issues, pp. 1 23-14 1 . Hillsdale, NJ: Lawrence Erlbaum.
Eubank, L. (Ed.) ( 1 99 1 ). Point counterpoint: Universal Grammar
in the second language: Amsterdam: John Benjamins.
Evans, J. St.B.T., Newstead, S., & Byrne, R.M. ( 1 993). Human
reasoning: The psychology of deduction. Hillsdale, NJ:
Lawrence Erlbaum.
Faerch, C. ( 1 986). Rules of thumb and other teacher formulated
rules in the foreign language classroom. In G. Kasper (Ed.),
Learning, teaching and communication in the foreign
language classroom, pp.65-79. Aarhus: Aarhus University
Press.
Faerch, C., & Kasper, G. ( 1 984). Two ways of defining
communication strategies. Language Learning, 34, 45-63 .
Felix, S. ( 1 99 1 ). The accessibility of Universal Grammar in second
language acquisition. In L.Eubank (Ed.), Point counterpoint:
Universal Grammar in the second language, p p . 89- 1 04 .
Amsterdam:John Benjamins.
Bibliography 265
Felix, S., & Weigl, W. ( 1 99 1 ). Universal Grammar in the classroom:
the effects of positive and negative evidence in the classroom.
Second Language Research, 7, 1 62-1 80.
Fletcher, P. ( 1 980). Paraphrase relationships among clefted
sentences. In G.D. Prideaux, B.L. Derwing & W.J. Baker
(Eds.), Experimental psycholinguistics, pp. 1 86-20 1 . Alberta:
University of Alberta Press.
Flynn, S. ( 1 987a). Contrast and construction in a parameter-setting
model of L2 acquisition. Language Learning, 3 7, 1 9-62 .
. ( 1 9 87b). A parameter-setting model of L2 acquisition.
-- Dordrecht: Reidel.
Fodor, J., & Crain, S. ( 1 987). Simplicity and generality of rules in
language acquisition. In B. MacWhinney (Ed.), Mechanisms of
language acquisition, pp.35-63. Hillsdale, NJ: Lawrence
Erlbaum.
Fotos, S. ( 1 993). Consciousness-raising and noticing through focus
on form: Grammar task performance versus formal
instruction. Applied Linguistics, 14, 3 8 5-407.
Fotos, S., & Ellis, R. ( 1 99 1 ). Communicating about grammar: A
task-based approach. TESOL Quarterly, 25, 87-1 1 2 .
Gass, S . ( 1 979). Language transfer and universal grammatical
relations. Language Learning, 29, 327-344 .
--. ( 1 982). From theory to practice. In W. Rutherford, & M.
Hines (Eds.), On TESOL '81, pp . 1 29- 1 3 9. Washington:
TESOL.
Gasser, M. ( 1 990). Connectionism and universals of second language
acquisition. Studies in Second Language Acquisition, 1 2,
1 79- 1 99 .
Givon, T . ( 1 985). Function, structure and language acquisition. I n
D. Slobin (Ed.), The crosslinguistic study of language
acquisition, Volume 1, pp. I 005- 1 028. Hillsdale, NJ:
Lawrence Erlbaum.
Glass, A., & Holyoak, K. ( 1 986). Cognition. New York: Prentice
Hall.
Gluck, M., & Granger, R. ( 1 993). Computational models of the
neural bases of learning and memory. Annual Review of
Neuroscience, 1 6, 667-706.
Goodluck, H. ( 1 99 1 ). Language acquisition: A linguistic
introduction. Oxford: Basil Blackwell.
Gordon, P. ( 1 990). Learnability and feedback. Developmental
Psychology, 26, 2 1 7-220.
266 Bibliography
Graf, P. ( 1 990). Life span changes in implicit and explicit memory.
Bulletin of the Psychonomic Society, 28, 3 53-3 5 8 .
Graf, P . , & Mandler, G. ( 1 984). Activation makes words more
accessible, but not necessarily more retrievable. Journal of
Verbal Learning and Verbal Behavior, 23, 5 5 3-568.
Graf, P., & Ryan, L. ( 1 990). Transfer-appropriate processing for
implicit and explicit memory. Journal of Experimental
Psychology: Learning, Memory and Cognition, 1 6, 978-992.
Green, P .S., & Hecht, K. ( 1 992). Implicit and explicit grammar: an
empirical study. Applied Linguistics, 13, 1 60-1 84 .
Green, R.E.A., & Shanks, D.R. ( 1 993). On the existence o f
independent explicit and implicit systems: A n examination o f
some evidence. Memory and Cognition, 21, 3 04-3 1 7.
Gregg, K. ( 1 984). Krashen's Monitor and Occam 's razor. Applied
Linguistics, 5, 79-1 00.
__ . ( 1 988). Epistemology without knowledge: Schwartz on
Chomsky, Fodor and Krashen. Second Language Research, 4,
66-8 0 .
Greenberg, S . ( 1 966). Universals of language typology. Stanford:
Stanford University Press.
Grossberg, S. ( 1 980). How does a brain build a cognitive code?
Psychological Review, 8 7, 1-5 1 .
__ ( 1 988). Neural networks and natural intelligence.
Cambridge, MA: MIT Press.
Hamann, S.B. ( 1 990). Level-of-processing effects in conceptually
driven implicit tasks. Journal of Experimental Psychology:
Learning, Memory and Cognition, 1 6, 970-977.
Hanson, S.J., & Burr, D.J. ( 1 990). What connectionist models learn:
Learning and representation in connectionist networks.
Behavioral and Brain Sciences, 13, 471-5 1 8.
Harley, B. ( 1 989). Functional grammar in French immersion: A
classroom experiment. Applied Linguistics, 1 0, 3 3 1-359.
__ . ( 1 993). Instructional strategies and SLA in early French
immersion. Studies in Second Language Acquisition, 1 5,
245-2 5 9 .
Harley, B . , & Swain, M. ( 1 984). The interlanguage of immersion
students and its implications for second language teaching. In
A. Davies, C. Criper & A. Howatt (Eds.), Interlanguage,
pp.29 1 -3 1 1 . Edinburgh: Edinburgh University Press.
Harrington, M., & Sawyer, M. ( 1 992). L2 working memory
capacity and L2 reading skill. Studies in Second Language
Acquisition, 1 4, 25-3 8 .
Bibliography 267
Hatch, E., & Lazaraton, A. ( 1 99 1 ). The research manual: Design
and statistics for applied linguistics. New York: Newbury
House.
Hawkins, J. ( 1 988). On generative and typological approaches to
universal grammar. Lingua, 74, 85-1 00.
Hayek, F.A. von. ( 1 962). Rules, perception and intelligibility.
Proceedings of the British Academy, 48, 3 2 1 -344.
Hayes, N., & Broadbent, D. ( 1 988). Two modes of learning for
interactive tasks. Cognition, 28, 249-276.
Hebb, D.O. ( 1 949). The organization of behavior. New York:
Wiley.
Henning, M.J. ( 1 993). A connectionist model of the second
language acquisition of negation. Unpublished M.A. thesis,
University of Hawai'i at Manoa, Department of English as a
Second Language.
Hilles, S. ( 1 986). Interlanguage and the pro-drop parameter. Second
Language Research, 2, 33-52.
Hintzman, D.L. ( 1 990). Human learning and memory: connections
and dissociations. Annual Review of Psychology, 41, 1 09-
1 39.
Holding, D.H. (Ed.)( 1 989). Human skills: Studies in human
performance. New York: John Wiley.
Holland, J.H., Holyoak, K.J., Nisbett, R.E., & Thagard, P.R. ( 1 986).
Induction: Processes of iriference, learning, and discovery.
Cambridge, MA: MIT Press.
Howard, J.H., & Ballas, J.A. ( 1 980). Syntactic and semantic factors
in the classification of non speech transient p atterns.
Perception and Psychophysics, 28, 43 1 -43 9.
Howard, J.H., Mutter, S.A., & Howard, D.V. ( 1 992). Serial pattern
learning by event observation. Journal of Experimental
Psychology: Learning, Memory and Cognition, 1 8, 1 029-
1 040.
Hulstijn, J. ( 1 989). Experiments with semi artificial input in second
language acquisition research. Scandinavian Working Papers
on Bilingualism, 8, 28-40.
__ . ( 1 990a). Implicit and incidental second language learning:
Experiments in the processing of natural and partly artificial
input. In H.W. Dechert (Ed.), Interlingual processing, pp . 5 0-
73 . Tubingen: Gunter Narr.
__ . ( 1 990b). A comparison between the information processing
and analysis/control approaches to language learning. Applied
Linguistics, 1 1, 3 0-45 .
268 Bibliography
Hulstijn, J., & Hulstijn, W. ( 1 984). Grammatical errors as a function
of processing constraints and explicit knowledge. Language
Learning, 34, 23-43 .
Hulstijn, J., & DeGraaff, R. ( 1 994). Under what conditions does
explicit knowledge facilitate the acquisition of implicit
knowledge? A research proposal. In J. Hulstijn & R. Schmidt
(Eds.), Consciousness and second language learning:
Conceptual, methodological and practical issues in language
learning and teaching. A/LA Review, 1 1, 97-1 1 2 .
Hyams, N. ( 19 87). The theory o f parameters and syntactic
development. In T. Roeper & E. Williams (Eds.), Parameter
setting, pp.3 1-49. Dordrecht: Reidel.
Jackendoff, R. ( 1 987). Consciousness and the computational mind.
New York: Academic Press.
Jacobs, B . ( 1 988). Neurobiological differentiation of primary and
secondary language acquisition. Studies in Second Language
Acquisition, 1 0, 3 03-3 3 7 .
Jacoby, L.L. ( 1 983). Remembering the data: Analyzing interactive
processes in reading. Journal of Verbal Learning and Verbal
Behavior, 22, 485-508.
Jacoby, L.L., & Dallas, M. ( 1 9 8 1 ). On the relationship between
autobiographical memory and perceptual learning. Journal of
Experimental Psychology: General, 1 1 0, 3 06-340.
Johnson, M.K., & Hasher, L. ( 1 987). Human learning and memory. ·
Annual Review of Psychology, 38, 63 1-668.
Jourdenais, 0., Ota, M., Stauffer, S., Boyson, B., & Doughty, C.
( 1 995). Does textual enhancement promote noticing? A
think-aloud protocol analysis. In R. Schmidt (Ed.), Attention
and awareness in foreign language teaching and learning,
pp. 1 32- 1 69. University of Hawai'i at Manoa: Second
Language Teaching and Curriculum Center Technical Report #
9.
Kadlec, H., & Townsend, J.T. ( 1 992). Signal detection analyses of
dimensional interactions. In F.G. Ashby (Ed.),
Multidimensional models of perception and categorization,
pp. 1 8 1-229. Hillsdale, NJ: Lawrence Erlbaum.
Kahneman, D., Slovic, P., & Tversky, A. ( 1 982). Judgments under
uncertainty: Heuristics and biases. Cambridge: Cambridge
University Press.
Keenan, E., & Comrie, B . ( 1 977). Noun phrase accessibility and
universal grammar. Linguistic Inquiry, 8, 63-99.
Bibliography 269
Keil, F .C. ( 1 989). Concepts, word meaning and cognitive
development. Cambridge, MA: MIT Press.
Kihlstrom, J.F. ( 1 984). Conscious, subconscious, unconscious: A
cognitive perspective. In K. Bowers & D. Meichenbaum
(Eds.), The unconscious reconsidered, pp . 1 49-2 1 1 . New
York: Wiley.
Kirsch, D. ( 1 99 1 ). PDP, learnability and innate knowledge of
language. UCSD Center for Research in Language Newsletter,
6, 3-1 7.
Klayman, J., & Ha, Y. ( 1 989). Hypothesis testing in rule discovery:
Strategy, structure and content. Journal of Experimental
Psychology: Learning, Memory and Cognition, 15, 5 9 6-604.
Knowlton, B ., & Squire, L. ( 1 994). The information acquired during
artificial grammar learning.Journal of Experimental
Psychology: Learning, Memory and Cognition, 20, 79-9 1 .
Kohonen, V. ( 1 992). Experiential language learning: Second
language learning as cooperative learner education. In D.
Nunan (Ed.), Collaborative language learning and teaching,
pp. 1 4-39. Cambridge: Cambridge University Press.
Krashen, S. ( 1 976). Formal and informal linguistic environments in
language acquisition and language learning. TESOL Quarterly,
1 0, 1 5 7-1 68 .
__ . ( 1 979). A response to McLaughlin, 'The monitor model:
some methodological considerations'. Language Learning,
29, 1 5 1-1 67.
__ . ( 1 98 1 a). Second language acquisition and second
language learning. Oxford: Oxford University Press.
__ . ( 1 9 8 1 b ). Aptitude and attitude in relation to second
language acquisition and learning. In K. Diller (Ed.),
Individual differences and universals in language learning
aptitude, pp. 1 55-1 75 . Rowley, MA: Newbury House.
__ . ( 1 982). Principles and practice in second language
acquisition. Oxford, New York: Pergamon.
__ . ( 1 985). The input hypothesis: Issues and implications.
Oxford, New York: Pergamon.
__ . ( 1 989). We acquire vocabulary and spelling by reading:
More evidence for the input hypothesis. Modern Language
Journal, 73, 440-464.
__ . ( 1 992). Under what circumstances, if any, should formal
grammar instruction take place? TESOL Quarterly, 26, 409-
41 1 .
270 Bibliography
__ . ( 1 993). The power of reading. Englewood, CO: Libraries
Unlimited.
__ . ( 1 994). The input hypothesis and its rivals. In N. Ellis
(Ed.), Implicit and explicit learning of language, pp.45-79.
London: Academic Press.
Krashen, S., & Seliger, H. ( 1 975). The essential contributions o f
formal and instruction in adult second language acquisition.
TESOL Quarterly, 9, 1 73-1 83 .
Krashen, S., & Terrell, T. ( 1 983). The natural approach:
Language acquisition in the classroom. California: Alemany
Press.
Krashen, S., Terrell, T., Ehrman, M., & Herzog, M. ( 1 984). A
theoretical basis for teaching the receptive skills. Foreign
Language Annals, 1 7, 26 1-275.
Kunst-Wilson, W., & Zajonc, R.B. ( 1 980). Affective discrimination
of stimuli that cannot be recognized. Science, 207, 5 5 7-55 8 .
Kyllonen, P.C., & Christal, R . ( 1 989). Cognitive modelling o f
learning abilities: A report o n LAMP. In R.F. Dillon & J.
Pellegrino (Eds.), Testing: Theoretical and applied
perspectives, pp. 1 46- 1 7 3 . New York: Praeger.
Lachter, J., & Bever, T.G. ( 1 988). The relation between linguistic
structure and associative theories of language learning-A
constructive critique of some connectionist learning models.
Cognition, 28, 1 95-247.
Lado, R. ( 1 965). Memory span as a factor in second language
learning. International Review of Applied Linguistics, 1, 1 23 -
1 30.
Langer, E.J. ( 1 978). Rethinking the role of thought in social
interaction. In J.H. Harvey, W. Ickes, & R.F. Kidd (Eds.), New
directions in attribution research, Volume 2, pp .3 5-5 5 .
Hillsdale, NJ: Lawrence Erlbaum.
Larson-Freeman, D., & Long, M.H. ( 1 99 1 ). An introduction to
second language acquisition research. New York: Longman.
Lee, C., & Gasser, M. ( 1 992). Where do underlying representations
come from?: A connectionist approach to the acquisition o f
phonological rules. In J . Dinsmore (Ed.), The symbolic and
connectionist paradigms: Closing the gap, pp. 1 79-2 0 8 .
Hillsdale, NJ: Lawrence Erlbaum.
Leeman, J., Arteagoitia, Fridman, B., & Doughty, C. ( 1 99 5 ) .
Integrating attention to form with meaning: Focus o n form in
content-based Spanish instruction. In R. Schmidt (Ed.),
Attention and awareness in foreign language teaching and
Bibliography 271
learning, pp. 1 03 - 1 3 1 . University o f Hawai'i at Manoa:
Second Language Teaching and Curriculum Center Technical
Report # 9.
Lewicki, P. ( 1 985). Nonconscious biasing effects of single instances
on subsequent judgments. Journal of Personality and Social
Psychology, 48, 5 63-574.
__ . ( 1 986). Nonconscious social information processing. New
York: Academic Press.
Lewicki, P., Hill, T., & Bizot, E. ( 1 988). Acquisition of procedural
skill about a pattern of stimuli that cannot be articulated.
Cognitive Psychology, 20, 24-37.
Light, L.L., & Singh, A. ( 1 987). Implicit and explicit memory in
young and older adults. Journal of Experimental Psychology:
Learning, Memory and Cognition, 13, 5 3 1 -54 1 .
Lightbown, P., & Pienemann, M . ( 1 993). Comments on Stephen D .
Krashen's "Teaching Issues: Formal Grammar Instruction. "
TESOL Quarterly, 2 7, 7 1 7-722.
Lightbown, P., & Spada, N. ( 1 990). Focus-on-form and corrective
feedback in communicative language teaching: Effects o n
second language learning. Studies In Second Language
Acquisition, 12, 429-448.
Lightbown, P., Spada, N., & Ranta, L. ( 1 99 1 ). The effect o f
instruction on IL-formed questions. Paper presented at the
Second Language Research Forum, UCLA, Los Angeles, USA.
Lightfoot, D. ( 1 989). Degree-0 learnability and the child's trigger
experience. Behavioral and Brain Sciences, 12, 3 2 1 -334.
. ( 1 99 1 ) . How to set parameters: Arguments from language
-- change. Cambridge, MA: MIT Press.
Logan, G.D. ( 1 988a). Toward an instance theory of automatization.
Psychological Review, 95, 492-527 .
. ( 1 988b). Automaticity, resources and memory: Theoretical
-- controversies and practical implications. Human Factors, 3 0,
5 83-5 9 8 .
--. ( 1 990). Repetition priming and automaticity: Common
underlying mechanisms? Cognitive Psychology, 22, 1-35.
Logan, G.D., & Etherton, J.L. ( 1 994). What is learned in
automatization? The role of attention in constructing an
instance. Journal of Experimental Psychology: Learning,
Memory and Cognition, 20, 1 022- 1 050.
Logan, G.D., & Klapp, S.T. ( 1 99 1 ). Automatizing alphabet
arithmetic problems 1 : Is extended practice necessary t o
 272 Bibliography
produce automaticity? Journal of Experimental Psychology
: MacWhinney, B., Leinbach, J., Taraban, R., & McDonald, J.
Learning, Memory and Cognition, 1 7, 1 79-1 95.
Logan, G.D., & Stadler, M. ( 1 99 1 ). Mechanisms of perfo
rmance
improvement in consistent mapping memory search
: Malt, B. ( 1 989). An on-line mvestlgat10n of prototyp� and
Automaticity or strategy shift? Journal of Experiment
al
Psychology: Learn ing, Memory and Cognition, 1 7, 478-4
96.
Long, M.H. ( 1 983). Does second language instruction
make a
difference? A review of research. TESOL Quarterly, 1 7, 3 5
9-
3 82.
. ( 1 985). A role for instruction in second langua
ge
__
development: Task-based language learning. In K. Hyltenstam
& M. Pienemann (Eds.), Modelling and assess
ing second
language acquisition, pp.77-99. Clevedon,
Avon :
Multilingual Matters.
__ . ( 1 988). Instructed interlanguage development. In L. Beebe
(Ed.), Issues in second language acquisition: Multip
le
perspectives, pp. 1 1 5-14 1 . New York: Newbury House.
__ . ( 1 989). Task, group and task-group interactions. University
of Hawaii Working Papers in ESL, 8, 1-26.
. ( 1 990). Maturational constraints on language devel opme
nt.
__
Studies in Second Language Acquisition, 12, 2 5 1 -286.
__ . ( 1 99 1 ). Focus on form: A design feature in language
teaching. In K. de Bot., R. Ginsberg & C. Kram sch (Eds.)
, Memory and Cognition, 15, 1 083-1 100.
Foriegn language research in cross cultural persp
ective,
pp.39-52. Amsterdam: John Benja mins.
__ . ( 1 994). The role of the linguistic environment in second
language acquisition. In W.C. Ritchie (Ed.), Handbook
of
language acquisition, Volume 2: Second language
acquisition. New York: Academic Press.
Long, M.H., & Crookes, G. ( 1 992). Three approaches to task-b
ased
syllabus design. TESOL Quarterly, 26, 1 34-1 56.
Loschky, L., & Bley-Vroman, R. ( 1 993). Grammar and task-b
ased
learning. In S. Gass & G. Crookes (Eds.), Task-based language
teaching: Pedagogical perspectives. Clevedon: Multilingual
Matters.
Lyster, R. ( 1 994). The effect of functional-analytic teaching
aspects of French immersion students' sociolinguistic
on
competence. Applied Linguistics, 15, 2 62-28 7.
Lyster, R., & Ranta, L. ( 1 995). Getting learners to notice
: of second language learning. L on don:
Negotiation ofform as negative evidence. Paper presented at
the Second Language Research Forum, Cornell University,
Ithaca, New York, USA.
Bibliography 273
( 1 989). Language learning: Cues or rules? Journal of Memory
and Language 28, 2 5 5 -2 7 7:
. .
exemplar strategies in classification. Jour��l of Experimental
Psychology, Learning, Memory and Cognztwn, 15, 539-5 5 5 .
Mandler, G . ( 1 992). Toward a theory o f consciousness. I� H .
.
Geissler, S . Link, & T . Townsend (Eds.), Cognztwn,
information processing and psychophysics, pp.43-67 ·
Hillsdale, NJ: Lawrence Erlbaum.
Marcus, G., Pinker, S., Ullman, M., Hollander, M., Rosen, T., &_ ?Cu,
.
F. ( 1 992). Overregularization in language acqulSlt1�n.
.
Monographs of the Society for Research m Child
Development, 57, 4 (Serial no. 228). .
Master, P. ( 1 994). The effect of systemati� instruction on l� arnmg
the English article system. In T. Odlm (Ed.), Perspective� on
pedagogical grammar, pp.229-252. New York: Cambndge
University Press. .
Matthews, R.C., Buss, R.R., Stanley, W.B., Blanchar�-F1�l�s, F.,
Ryeul Cho, J., & Druhan, B . ( 1 9 89). Role of imphc1t �d
.
explicit processes in learning from examples: A synerg1�tlc
effect. Journal of Experimental Psychology, Learning,
.
McAndrews, M.P., & Moscovitch, M. ( 1 985). Rule-based and
.
exemplar-based classification in artificial grammar learnmg.
Memory & Cognition, 13, 469-475.
Mccawley, J. (1 988). The Syntactic phenomena of Englzsh: . Volume
1. Chicago: University of Chicago Press.
McClelland, J., & Elman, J. ( 1 986). The TRACE model of speech
perception. Cognitive Psychology, 18, 1-86.
.
McKoon, G., Ratcliff, R., & Dell, G. ( 1 986). A critical evalu�t10n of
the semantic episodic distinction. Journal of Experimental
Psychology: Learning, Memory and Cognition, 12, 295-3 ? 6.
McLaughlin, B. ( 1 978). The monitor model: Some methodological
.
considerations. Language Learning, 28, 3 09�332. .
. ( 1 9 80). On the use of miniature arti�cial languages m
_ .
--second language research. Applied Psycholmguzstzcs, 1, 3 5 7-
3 69 .
--. ( 1 987). Theories
Edward Arnold.
__. ( 1 990a). Restructuring. Applied Linguistics, 1 1, 1 1 3-1 28.
 274 Bibliography
Bibliography 275
__. ( 1 990b). "Conscious" versus "unconscio
us" learning. TESOL
Quarterly, 24, 6 1 7-6 3 4 . Murphy, G.L., & Medin, D.L. ( 1 985 } . The role of theories in
McLaughlin, B . , Rossman, T .
. , & McLeod, B . ( 1 983 ). conceptual coherence. Psychological Review, 92, 2 89-3 1 6 .
language learning: An informa Second Naiman, N., Frolich, M., Stern, H.H., & Todesco? A . 0 978). The
tion processing perspectiv e. .
Language Learning, 33, 1 3 5-1
57. good language learner. Research in �ducation Senes 7 : The
Medin, D. � ., & Smith, E.E . Ontario Institute for Studies in Educat10n.
( 1 98 1 ). Strategies and classifi .
learnmg. Journal of Experim cation Nation, R., & McLaughlin, B. ( 1 986). Experts and novices: an
ental Psychology: Huma n
Learning and Memory, 7, 24 1-2
53 . information processing approach to the good language learner
Meike, B . ( 1 988). Mind/ah 2.
1 . Dartmouth College: Coursew problem. Applied Psycholinguistics, 7, 4 1-5 � .
De vel opm ent De partment. are Navon, D. ( 1 984). Resources: A theoretical soupstone.
Meisel, J., Clahsen, H., & Psychological Review, 91, 2 1 6-234.
Pienemann, M. ( 1 98 1 ). On
deve!op�ental stages in natura Navon, D., & Gopher, D. ( 1 980). On the economy of the human
l second language acquisition.
Studies zn Second Language Acq
uisition, 3, 1 09- 1 3 5 . processing system. Psychological Review, 86, 254-2 5 5 .
Mellow, D , & Cum i g, A .
: i_n � ( 1 994 ). Concord in interlangua Nelson, D.L. ( 1 989). Implicitly activated �nowledg� and memory.
ge: .
. Efficiency or pnmmg? Applied Linguistics, 1 5, 442-47 1 . In C. Izawa (Ed.), Current issues zn cognitive processes,
Menckle, �·!"f ·, & R in old, pp.3 69-3 87. Hillsdale, NJ: Lawrence Erlbaum.
� � E.M . ( 1 99 1 ). Comparing dire .
(Exphc19 and mdirect (Im
plicit) measures to study
ct Nelson, D.L., Canas, J., Bajo, M., & Keelean, P. ( 1 9 � 7). Comparmg
uncon�cious memory. Journal word fragment completion and cued recall with letter cues.
of Experimental Psychology, .
Learning, Memory and Cogniti Journal of Experimental Psychology, Learning, Memory and
on, 1 7, 224-23 3 .
Mohle, D., & Raupach, M . Cognition, 13, 542-5 52.
( 1 989 ). Language transfer and
procedur�l knowledge. In H. Dec Nelson, D.L., & McEvoy, C.L. ( 1 979). Encoding context and �et
hert & M. Raupach (Eds.),
Transfer zn language production
, pp. 8 1 - 1 03 . Norwood, NJ: size. Journal of Experimental Psychology: Human Learning
Ablex. and Memory, 5, 292-3 14.
Moray, N. ( 1 959 ). Attention in Nisbett, R.E., & Wilson, T.D. ( 1 977). Telling more than we � an
dichotic listening task s:Affective
cues and the influence of instruc know: Verbal reports on mental processes. Psychological
tion s. Quarterly Journal of
Experimental Psychology, 1 1, 5
6-60. Review, 84, 23 1 -2 5 9 .
Morgan, J., & Newport, E. ( 1 98 Nissen, M.J., & Bullemer, P. ( 1 987). Attentional requirement� of
1 ). The role of constituent structur .
in the induction of an artificial e learning: Evidence from performance measures. Cognitive
. language. Journal of Verbal
Learning and Verbal Behavior, 20,
67- 8 5 . Psychology, 1 9, 1-32.
Morg�n, J., �eier, R., & Newpor .
t, E . ( 1 9 87) . Structural packaging Norman, D.A. ( 1 968). Toward a theory of memory and attent10n.
m the mput to language learning Psychological Review, 84, 23 1-259.
. . : Contributions of prosodic .
and morphologica l markmg of phrases in the acq Nosofsky, R.M. ( 1 992). Exemplar-based a��roach to relatmg
language. Cognitive Psychology, uisition o f categorization, identification and recogmtion. In F :G. Ashby
1 9, 498-55 0.
( 1 989 ). Facilitating the acquisition (Ed.), Multidimensional �odels of perception and
of syntax with cross­
.
_

sentential cues to phrase structur categorization, pp.363-395 . Hillsdale, NJ: Lawrence Erlbaum.
e. Journal of Memory and
Language, 28, 3 60-3 74. Nosofsky, R.M., Clark, S .E., & �hin, H.J. ( 1989). Rules .and
Munsell, P ., & Carr, T.H . ( 1 98 . . . _
exemplars in categorization, identification and recogmt10n.
.
�
revie': o ' Second Language Acquisi
Lea�nm g by S.D .Krashen. Langua
1 ). Monitoring the monitor: A
tion and Second Language Journal of Experimental Psychology: � .
rning, Memory and
Cognition, 15, 282-3 04.
ge Learning, 31, 493 -50 2. . .
Murakami, K. ( 1 974 ). A languag
e aptitude test for the Japanese. Nosofsky, R.M., Kruschke, J., & McKinley, S. ( 1 992). Com�mi?g
System, 2, 3 1-47. exemplar-based category representations and connect10� ist
learning rules. Journal of Experimental psychology: Learning,
Memory and Cognition, 18, 2 1 1 -23 3 .
 276 Bibliography
Nunan, D. ( 1 988). Syllabus design. Oxford: Oxford Unive
rsity
Press.
. ( 1 989). Designing tasks for the communicative classroom.
__
. Camb ridge: Cambridge University Press.
Odlm, T. ( 1 9 � 6). On �he nature and use of explicit knowl
edge.
International Review of Applied Linguistics, 24, 1 23-14 4.
__ ( 1 994). (Ed.) 1:ersP_ectives on pedagogical grammar.
Oxford: Oxford Umversity Press.
O' Grady, w. 0 ? 93). Syntactic development: The acquisition
_ h. Chica _ rsity of of
Englts go: Umve Chicago Press.
O' Malley, M., & Chamot, A. ( 1 990). Learn ing strategies in second
language acquis ition. Cambridge: Cambridge University Press.
Papagno, C., Valentine, T., & Baddeley, A. ( 1 99 1 ). Phono logica
l
short-term memory and foreign-language vocabulary learning.
._Jo urnal of Memory and Language, 30, 3 3 1-347 .
Paradi s, M. ( 1 994 ). �eurolinguistic aspects of implic it and explic
_ _ it
mem?�y: implic ations for bilingualism. In N. Ellis (Ed.),
Impltc1t and explicit learning of language, pp.39 3-4
1 9.
London: Academic Press.
Parry, T., & Child, R. ( 1 990). Field testing a language aptitud
e
me�sure: the cas� of VORD. In C. Stansfield (Ed.), Language
aptitude reconsidered, pp.3 1-6 1 . Washington: Center
for
Applied Linguistics.
Patkowski, M.S. ( 1 980). The sensitive period for the acquisition
of
syntax . Language Learning, 30, 449-4 72.
Pavesi , M. ( 1 986). Markedness, discoursal modes, and relativ
e . .
clause formation in a formal and an informal context. Studies
In Second Language Acquisition, 8, 3 8-55 .
Pekala, R.J. ( 1 99 1 ). Quantifying consciousness: A n empir
ical
approach. New York: Plenum Press. ·
Perruchet, P., & Amorim, M.A. ( 1 992). Conscious knowledge and
c� ang: s �_ n performance in sequence learning: Evidence against
dissoc iation. Journal of Experimental Psychology: Learn ing,
Memory and Cognition, 18, 78 5-800 .
Perruchet, P., & Gallego, G. ( 1 993). Association between conscious
knowledge and performance in normal subjects: A reply t o
Cohen and Curran ( 1 993), and Willingham, Greeley and
Bardo�e ( 1 993). Journal of Experimental Psychology:
Learnmg, Memory and Cognition, 1 9, 1 43 8-1 444. ·
Perruchet, P., ?allego, J., & Savy, I. ( 1 990). A critical reappraisal
of the evidence for unconscious abstraction of determ inistic
Bibliography 277
rules in complex experimental situations. Cognitive
Psychology, 22, 493-5 16.
. .
Perruchet, P., & Pacteau, C. ( 1 990). Synthetic grammar learnmg:
Implicit rule abstraction or fragmentary knowledge? Journal
of Experimental Psychology: General, 1 19, 264-275.
Pica, T. ( 1 983). Adult acquisition of English as a second language
under different conditions of exposure. Language Learnmg, _
33, 465-497.
. ( 1 985). The selective impact of classroom instruction o n
-- second language acquisition. Applied Linguistics, 6, 2 1 4-222.
--. ( 1 994). Questions from the language classroom: A research
perspective. TESOL Quarterly, 28, 49-80.
Pimsleur, P. ( 1 966). The Pimsleur language aptitude battery. New
York: Harcourt Brace Jovanovich .
Pienemann, M. ( 1 992). Teachability theory. University of Sydney:
Language Acquisition Research Centre.
.
Pienemann, M., Johnston, M., & Brindley, G. ( 1 988). Developmg
an acquisition-based procedure for second language assessment.
Studies in Second Language Acquisition, 1 0, 2 1 7-243 .
Pinker, S. ( 1 989). Learnability and cognition: The acquisition of
argument structure. Cambridge, MA: MIT Press.
__ . ( 1 99 1 ). Rules of language. Science, 253, 530-? 35.
. ( 1 994). The language instinct. London: Pengum Boo�s .
.
__
Pinker, S., & Prince, A. ( 1 988). On language and connectiomsm:
analysis of a parallel distributed processing model of language
acquisition. Cognition, 28, 73- 1 93.
Pitts, M., White, H., & Krashen, S. ( 1 989). Acqmnng second
languagevocabulary through reading: A replication of the
.
Clockwork Orange study using second language acqmrers.
Reading in a Foreign Language, 5, 27 1-275.
Polanyi, M. ( 1 95 8). Personal knowledge: Towards a post-critical
philosophy. Chicago: University of Chicago Press.
Prabhu, N.S. ( 1 987). Second language pedagogy. Oxford: Oxford
University Press.
Preston, D. ( 1 989). Sociolinguistics and second language
acquisition. Oxford: Basil Blackwell.
. .
Prince, E. ( 1 978). A comparison of wh-clefts and it-clefts m
discourse. Language, 54, 8 83-907.
Raaijmakers, J.G.W., & Shiffrin, R.M. ( 1 992). Models of memory.
Annual Review of Psychology, 43, 205-234.
.
Radford, A. ( 1 990). Syntactic theory and the acquisition of English
syntax. Oxford: Basil Blackwell.
 Bibliography 279
278 Bibliography

contingencies and instructional control, pp.36-5 1 .

New York:
Rankin, J. (1 990). A case for closedmindedness: Accuracy and
complexity in performance on open and closed second Plenum Press.
of mem ory .
langu�[fe tasks. Unpublished manuscript, University o f Richardson-Klavehn, A., & Bjork, R. (1 988). Measures
Hawm 1 at Manoa, Department o f English as a Second Annual Review of Psychology 39, 475- 543 .
dural skill
Language. Robbins, S. ( 1 992). A neurobiological model of proce
5.
Ramat, A.G. ( 1 992). Grammaticalization processes in the area o f acquisition. Issues in Applied Linguistics, 3, 23 5-26
in the units of
temporal and modal relations. Studies in Second Language Robinson, P. ( 1 986). Constituency or dependency er' s
the learn
Acquisition, 1 4, 297-3 22. language acquisition? An approach to describing
1 0, 4 1 7-
Reber, A.S. (1 967). Implicit learning of artificial grammars. Journal analysis of formulae. Lingvisticae Investigationes,
of Verbal Learning and Verbal Behavior, 6, 8 5 5-863 . 437.
learning.
. (1 969). Transfer of syntactic structure in synthetic . ( 1 9 89). Procedural vocabulary and language
--
Journal of Pragmatics, 13, 523-546.
__

languages. Journal of Experimental Psychology, 81, 1 1 5-1 1 9. data:

. ( 1 990). Metaphors for the description of acquisition
--: (1 976 !. Implicit learning of synthetic languages: The role of --
From constituency 'trees ' to dependency
' frame s' .
mstructional set. Journal of Experimental Psychology: 292 .
Human Learning and Memory, 2, 88-94. International Review of Applied Linguistics, 28, 273-
tiation and
. ( 1 989). Implicit learning and tacit knowledge. Journal of . ( 1 992) . Discourse semantics, interlanguage nego
1,
--the lexicon. University of Hawai 'i Working Papers ESL, 1 in
__

Experimental Psychology: General, 1 1 8, 2 1 9-23 5 .

. (1 990). On the primacy of the implicit: Comment on 5 1 -90 .
cit/explicit
. ( 1 993a). Problems of knowledge and the impli
__

Perruchet and Pacteau. Journal of Experimental Psychology: -- i 'i Working

General, 1 1 9, 3 40-342. distinction in SLA theory. University of Hawa
. ( 1 992). The cognitive unconscious: An evolutionary Papers in ESL, 12, 90-1 3 8 .
for the
. ( 1 993b ). Review of David Nunan's 'Designing tasks
__

perspective. Consciousness and Cognition, J , 93- 1 3 3 . , 14, 442- 446 .

. ( 1 993 ). Implicit learning and tacit knowledge. Oxford: --communicative classroom ' . Applied Lingulangu istics
age acqu isitio n
. ( 1 994a ). Implicit knowledge, second
__

Clarendon Press. -- 1 60-1 66.

Reber� A.S., & �llen, R. ( 1 978) . . Analogy and abstraction strategies and syllabus construction. TESOL Quarterly, 28,
sses: Noun
m sy�t�etic grammar learnmg: A functionalist interpretation. . ( 1994 b). Universals of word formation proce
an as a second
Cognztzon, 6, 1 89-22 1 .
·

--incorporation in the acquisition of Samo

Reber, A., Kas� im, S., Lewis, s ., & Cantor, G. ( 1 980). On the language. Language Learning, 44, 569-6 1 5 .
locative
. .
relati�msh1p between 1mphcit , and explicit modes in the . ( 1 994c ). Modeling the acquisition of the
isitio of verb
n
learnmg of a complex rule structure. Journal of Experimental --alternation: Rules and associations in the acqu University o f
Psychology, 6, 492-502. argument structure. Unpublished manuscript,
a Second
Reber, A.s .,, & Millw�rd, R.B. ( 1 97 1 ) . Event tracking in probability Hawai' i at Manoa, Department of English as
learnmg. American Journal of Psychology, 84, 85-99. Language.
narrative
Reber, A.�., Walke�field, F .,, � Hernstadt, F. ( 1 99 1 ). Implicit and . ( 1 995a ). Task comp lexity and second language
--
.
exphc�t learnmg: Ind1v1dual differences and IQ. Journal of discourse. Language Learning, 45, 99-1 40.
similarity
Experimental Psychology: Learning, Memory and Cognition' . ( 1 995b ). Aptitude, awareness and the fundamental
learn ing. In R.
1 � 8 8 8-89 6 . --of implicit and explicit second language n lang uage
Reed, S., Ackinclos�, �.�., & Voss, A. ( 1 990). Selecting analogous Schmidt (Ed.) , Attention and awareness in foreig at
of Haw ai'i
problems: S1m1lanty versus inclusiveness. Memory & teaching and learn ing, pp.3 1 1-36 2. University
culum Cent er
Cognition, 18, 83- 98. Manoa: Second Language Teaching and Curri
Reese, H. ( 1 989). Cognitive and behavioristic views of rule use. I n Technical Report # 9 .
S.C. Hays (Ed.), Rule-governed behavior: Cognition,
 280 Bibliography
Bibliography 281

. (1 995 c). Attention, memory and

the 'noticing' hypoth esis .
__

Language Learning, 45, 283 -3 3 1 application o f experience. Journal ?f Experimental

. .
__ . ( 1 995 d). Implicit learning, Psychology: Learning, Memory and Cognztwn, 21, 469-4 8 2 .
memory and knowledge:
Conceptual and methodological Rumelhart, D.E., & McClelland, J. ( 1 986). O n learning the past
issues for experimental
second language research.
Manuscript submitted for tenses of English verbs. In J. Mcclelland, D. Rumelhart, & the
PDP Research Group (Eds.), Parallel distributed processing: _
publication.
. ( 1 996 ) Le�n_iing i ple and com Explorations in the microstructures of cognition, Volume 2:
: �� plex second language rules
__

under 1mphc1t, mc1dental, rule Psychological and biological models, pp.2 1 6-2 7 1 .
-search and instructed
. conditions. Studies in Second Language Acquisition, 18, ( 1 ) Cambridge, MA: MIT Press.
. . . .
Robmson, P., & Ha, M . ( 1 993 ). Russo, R., & Parkin, A. ( 1 993). Age differences m 1mphc1t
Instance theory and second
language rule learning under explicit memory: More apparent than real. Memory & Cognition, 2 1 ,
conditions. Studies in
Second Language Acquisition, 1 5, 73-8 0 . . . .
4 1 3-43 8.
Robinson, P., & Ross, S. ( 1 996 ). Rutherford, W. ( 1986). Grammatical theory and L2 acqms1t1on: A
The development of task-based
testing in English for Academic Pur brief overview. Second Language Research, 2, 1-1 5 .
poses programs. Applied
Linguistics, 1 7 (2). . ( 1 987). Second language grammar: Learning and
Robinson, P., Ting, S., & Urwin, teaching. London, New York: Longman.
J. ( 1 995 ). Investigating second --

language task complexity. RELC Jou Rutherford, W., & Sharwood Smith, W. (Eds.) ( 1 988). Grammar
. rnal 26 (2).
Roediger, H.L ., & Blaxton, T. ( 1 987 and second language teaching. Rowley, MA: Newbury House.
). Effects of varying mod ality,
surface features and retention inte Sasaki, M. ( 1 993). Relationships among se�ond . language
rval on priming in word­
frag ment completion. Memory & Cogniti proficiency, foreign language aptitude and mtelhgence: A
. on, 15, 3 79-3 88 . .
Roed1g�r, H L Weldon, M.S . & Cha structural equation modeling approach. Language Learning,
: ·; llis, B .H. ( 1 989 ). Explaining
d1ssoc1at10ns between implicit and 43, 3 1 3-344.
explicit measures of
rete!1tion: A processing account. Sato, E., & Jacobs, B. ( 1 992). From input to intake: towards a
In H.L . Roediger, & F .I.
Cra1k (Eds.), Varieties of memory brain-based perspective on selective attention. Issues In
and consciousness, pp. 3-
4 1 . Hillsdale, NJ: Lawrence Erlbaum Applied Linguistics, 3, 267-292.
. .
Rosansky, E.J. ( 1 975 ). The critical Sawyer, M. ( 1 993). The modern language aptitude test: A Ras eh
period for the acquisition o f .
language: Som e cognitive develop analysis of its behavior with native an� non-natz��
mental con siderati ons . . . ,
Working Papers on Bilingualism, 6, populations. Unpublished manuscript, Umvers1ty of Hawm I
1 1 5-1 2 1 .
Rosch, E . ( 1 973 ). Natural categor at Manoa, Department of English as a Second Language.
ies. Cognitive Psychology 4
3 2 8-3 5 0 . ' ' Schacter, D.L. ( 1987). Implicit memory: History and current status.
Rosenb latt, F. ( 1 958 ). The Percep Journal of Experimental Psychology: Learning, Memory and
tron : A probabilistic mod el for
information storage and organiza Cognition, 13, 501-5 1 8 . .
tion m the brain.
Psychological Review, 65, 386-40
8. __ . ( 1 989). On the relation between �emory an� consc10usness:
Ross, B .H. ( 1 989 ). Remindings in dissociable interactions and consc10us expenence. In H.L.
learning and instruct ion. In S.
Vosniadou & A. Ortony (Eds.), Sim Roediger, & F.I. Craik (Eds.), Varieties of memory and
ilarity and analogical
reason ing, pp.43 8-4 69. Cambrid consciousness, pp.35 5-3 89. Hillsdale, NJ: Lawre�c� Erlbaum.
ge: Cambridge Univers ity
Pre ss. Schacter, D.L., Chiu, P., & Ochsner, K. ( 1 993). Imp licit memory :
Ros s, s..a: ., & Kenned , P.T . ( 1 990 A selective review. Annual Review of Neuroscience, 1 6, 1 5 9-
� ). Generalizing from the use o f
earher exam ples m problem solving. 1 82 .
Journal of Experimental
Psychology: Learning, Memory and Schacter, J . ( 1 990). The issue o f completeness in second language
Cognition, 1 6, 42- 5 5 .
Rothkopf, E.Z ., & Dashen, M.L acquisition. Second Language Research, 6, 93-124 . . .
. ( 1 995 ). Particularization: _
Inductive speeding of rule-governe Schachter, J., Tyson, A., & Diffley, F. ( 1 976). Learner mtmt10ns of
d decisions by narrow
grammaticality. Language Learning, 26, 67-76.
282 Bibliography
Schmidt, R. ( 1 988). The potential of parallel distributed processing
for SLA theory and research. University of Hawai 'i Working
Papers in ESL, 7, 5 5-66.
__ . ( 1 990). The role of consciousness in second language
learning. Applied Linguistics, 1 1, 129- 1 5 8 .
__ . ( 1 992). Psychological mechanisms underlying second
language fluency. Studies in Second Language Acquisition,
1 4, 3 5 7-3 87.
__ . ( 1 993a). Awareness and second language acquisition. Annual
Review of Applied Linguistics (1993), 1 3, 206-226 .
( 1 993b). Consciousness, learning and interlanguage
pragmatics. In G. Kasper & S. Blum-Kulka (Eds.),
Interlanguage pragmatics, pp.2 1-42. New York: Oxford
University Press.
__ . ( 1 994a). Deconstructing consciousness in search of useful
definitions for applied linguistics. In J. Hulstijn & R. Schmidt
(Eds.), Consciousness and second language learning:
Conceptual, methodological and practical issues in language
teaching and learning. A/LA Review, 1 1, 1 1-26.
__ . ( 1 994b). Implicit learning and the cognitive unconscious: Of
artificial grammars and SLA. In N. Ellis (Ed.), Implicit and
explicit learning of language, pp . 1 65-209. London:
Academic Press.
__ . ( 1 995). Consciousness and foreign language: A tutorial o n
the role o f attention and awareness in learning.. In R. Schmidt
(Ed.), Attention and awareness in foreign language teaching
and learning, pp. 1 2-52. University of Hawai'i at Manoa:
Second Language Teaching and Curriculum Center Technical
Report # 9.
Schmidt, R., & Frota, S. ( 1 986). Developing basic conversational
ability in a second language: A case study of an adult learner of
Portuguese. In R. Day (Ed.), Talking to learn: Conversation in
second language learning, pp.23 7-326. Rowley, MA:
Newbury House.
Schneider, W. ( 1 993). Varieties of working memory as seen in
biology and in connectionist/control architectures. Memory
and Cognition, 21, 1 84- 1 92 .
Schneider, W., & Detweiler, M . ( 1 988). The role o f practice in
dual-task performance: Toward workload modeling in a
connectionist-control architecture. Human Factors, 30, 5 3 9-
566.
Bibliography 283
Schwartz, B.D. ( 1 986). The epistomological status of second
language . acquisition. Second Language Research, 2, 1 2 0-
1 59 .
. ( 1 988). A reply to Gregg: In defence of theory building.
-- Second Language Research, 4, 1 5 7- 1 73 .
--. ( 1 992). Testing between UG-based and problem-solving
models of L2 Acquisition: Developmental sequence data.
Language Acquisition, 2, 1-19.
. (1 993). On explicit and negative data effecting and
-- affecting competence and linguistic behavior. Studies in
Second Language Acquisition, 1 5, 1 47-1 63 .
Schwartz, B.D., & Gubala-Ryzak, M. ( 1 992). Learnability and
grammar reorganization in L2A: Against negative evidence
causing the unlearning of verb movement. Second Language
Research, 8, 1-38.
Schwartz, B., Rosse, R., & Deutsch, S. ( 1 993). Limits of the
processing view in accounting for dissociations among
memory measures in a clinical population. Memory &
Cognition, 21, 63-72.
Scovel, T. ( 1 988). A time to speak. Rowley, MA: Newbury House.
Searle, J. ( 1 990). Consciousness, explanatory inversion and
cognitive science. Behavioural and Brain Sciences, 13, 5 8 5-
596.
__ . ( 1 992). The rediscovery of the mind. Cambridge, MA: M I T
Press.
Seliger, H. ( 1 975). Inductive method and deductive method in
language teaching: A re-examination. International Review of
Applied Linguistics, 13, 1-1 8 .
. ( 1 979). On the nature and function of language rules in
-- language teaching. TESOL Quarterly, 13, 359-396.
Servan-Schreiber, E., & Anderson, J.R. ( 1 990). Learning artificial
grammars with competitive chunking. Journal of
Experimental Psychology: Learning, Memory and Cognition,
1 6, 5 92-60 8 .
Shanks, D.R., & St. John, M.F. ( 1994). Characteristics o f dissociable
human systems. Behavioral and Brain Sciences, 1 7, 3 6 7-44 7.
Shannon, C.E. ( 1 948). A mathematical theory of communication.
Bell System Technical Journal, 27, 3 79-243 .
Sharwood Smith, M. ( 1 980). Consciousness-raising and the second
language learner. Applied Linguistics, 2, 1 59-1 68.
284 Bibliography
__ . ( 1 99 1 ). Speaking to many minds: On the relevance of
different types of language information for the L2 learner.
Second Language Research, 7, 1 1 8- 1 3 2 .
__ . ( 1 993 ) . Input enhancement in instructed SLA. Studies in
Second Language Acquisition, 1 5, 1 65-1 79.
__ . ( 1 994). The unruly world of language. In N. Ellis (Ed.),
implicit and explicit learning of language, pp. 3 4-43 .
London: Academic Press.
Shiffrin, R.M. ( 1 993). Short-term memory: A brief commentary.
Memory & Cognition, 21, 1 93-197.
Shiffrin, R.M., & Schneider, W. ( 1 977). Controlled and automatic
human information processing: II. Perceptual learning,
automatic attending, and a general theory. Psychological
Review, 84, 1 27-1 90.
Shirai, Y. ( 1 992). Conditions on transfer: A connectionist
approach. Issues in Applied Linguistics, 3, 9 1- 1 20.
Shook, D.J. ( 1 994). FL/L2 reading, grammatical information and
the input-to-intake phenomenon. Applied Language
Learning, 5, (2), 5 7-93 .
Skehan, P. ( 1 986). The role of foreign language aptitude in a model
of school learning. Language Testing, 3, 1 88-22 1 .
__ . ( 1 987). Cluster analysis and the identification of learner
types. In V. Cook (Ed.), Experimental approaches to second
language acquisition, pp. 8 1-94. Oxford: Pergamon Press.
__ . ( 1 989). Individual differences in second language learning.
London: Edward Arnold.
Skousan, R. ( 1 990). Analogic modeling of language structure.
Dordrecht: Reidel.
Slimani, E. ( 1 992). Evaluation of classroom interaction. In J.C.
Alderson & A. Beretta (Eds.), Evaluating second language
education, pp. 1 97-220. Cambridge: Cambridge University
Press.
Slobin, D . ( 1 979). Psycholinguistics. (2nd edition) Glenview, IL:
Scott Foresman .
__ . ( 1 985). Crosslinguistic evidence for the language making
capacity. In D. Slobin (Ed.), The crosslinguistic study of
language acquisition, Volume 2, pp. 1 1 5 7- 1 249. Hillsdale,
NJ: Lawrence Erlbaum.
Smolensky, P. ( 1 988). On the proper treatment of connectionism.
Behavioral and Brain Sciences, 1 1, 1-74.
Snodgrass, J.G. ( 1 989). How many memory systems are there
really?: Some evidence from the picture fragment completion
Bibliography 285
task. In C. Izawa (Ed.), Current issues in cognitive processes,
pp. 1 3 5- 1 74. Hillsdale, NJ: Lawrence Erlbaum.
Spada, N., & Lightbown, P. ( 1 993). Instruction and the
development of questions in L2 classrooms. Studies in Second
Language Acquisition, 15, 205-224.
Squire, L. ( 1 987). Memory and brain. New York: Oxford
University Press.
. ( 1 992). Declarative and nondeclarative memory: Multiple
-- brain systems supporting learning and memory. Journal of
Cognitive Neuroscience, 4, 232-243 .
Squire, L., Amaral, D., & Press, G. ( 1 990). Magnetic resonance
_ .
imaging measurements of htppocampal format10n and
mammillary nuclei distinguish medial temporal lobe and
diencephalic amnesia. Journal od Neuroscience, 1 0, 3 1 0 6-
3 1 l 7.
Squire, L., & Cohen, N. ( 1 9 84). Human memory and amnesia. In
J.McGaugh, G. Lynch, & N. Weinberger (Eds.), Proceedings
of the conference on the neurobiology of learning and
memory, pp.3-64. New York: Guildford Press.
Squire, L., & McKee, R. ( 1 993). Declarative a�d nondeclarati":e
memory in opposition: When prior events mfluence amnesic
patients more than normal subjects. Memory & Cognition, 21,
424-43 0.
Squire, L., & Zola-Morgan, S. ( 1 99 1 ). The medial temporal lobe
memory system. Science, 253, 1 3 80-1 3 86.
Stadler, M. ( 1 995). Role of attention in implicit learning. Journal
of Experimental Psychology; Learning, Memory and
Cognition, 21 , 674-685.
Stansfield, C. (Ed.) ( 1 990). Language aptitude reconsidered.
Washington: Center for Applied Linguistics.
Stem, L. ( 1 985). The structures and strategies of human memory.
Homewood, IL: Dorsey Press.
Stevick, E. ( 1 976). Memory, meaning and method. Rowley, MA :
Newbury House.
Strayer, D.L., & Kramer, A. ( 1 990). Attentional requirements o f
controlled and automatic processing. Journal of Experimental
Psychology: Learning, Memory and Cognition, 1 6, 67-82.
Sweller, J. ( 1 988). Cognitive load during problem solving: Effects on
learning. Cognitive Science, 12, 257-2 8 5 .
Sweller, J., Chandler, P., Tierney, P . , & Cooper, M . ( 1 990).
Cognitive load as a factor in the structuring of technical
286 Bibliography
material. Journal of Experimental Psychology: General, 1 1 9,
1 76- 1 92 .
Tarone, E . ( 1 985). Variability in interlanguage use: A study o f style
shifting in morphology and syntax. Language Learning, 3 5,
3 73-403 .
__ . ( 1 988). Variation in interlanguage. London: Edward
Arnold.
Taylor, M.M., Lindsay, P.H., & Forbes, S.M. ( 1 967).
Quantification of shared capacity processing in auditory and
visual discrimination. Acta Psychologica, 2 7, 223-229.
Terrell, T. ( 1 9 82). The natural approach to language teaching: an
update. Modern Language Journal, 65, 24 1-247.
Thompson, R.F. ( 1 986). The neurobiology of learning and
memory. Science, 233, 94 1-947.
Tomlin, R., & Villa, V. ( 1 994). Attention in cognitive science and
SLA. Studies in Second Language Acquisition, 1 6, 1 85-204.
Towell, R. & Hawkins, R. ( 1 994 ). Approaches to second language
acquisition. Clevedon, Avon: Multilingual Matters.
Trahey, M., & White. L. ( 1 993). Positive evidence and preemption
in the second language classroom. Studies in Second
Language Acquisition, 15, 1 8 1 -204.
Tulving, E. ( 1 984). Precis of 'Elements of episodic memory. '
Behavioral and Brain Sciences, 7, 223-238.
__ . ( 1 985). How many memory systems are there? A merican
Psychologist, 4, 3 8 5-3 98.
__ . (1 986). What kind of hypothesis is the distinction between
episodic and semantic memory? Journal of Experimental
. Psychology: Learning, Memory and Cognition, 12, 307-3 1 1 .
Tulving, E., Schacter, P., & Stark, H. ( 1 982). Priming effects in
word fragment completion are independent of recognition
memory. Journal of Experimental Psychology: Learning,
Memory and Cognition, 8, 336-342.
Turner, C.W., & Fischler, LS. ( 1 993 ). Speeded tests of implicit
knowledge. Journal of Experimental Psychology: Learning,
Memory and Cognition, 19, 1 1 65- 1 1 77.
Vallar, G., & Shallice, T. ( 1 990). Cognitive impairments of short­
term memory. Cambridge: Cambridge University Press.
VanPatten, B. ( 1 988). How juries get hung: Problems with the
evidence for a focus on form in teaching. Language Learning,
38, 243-260.
__ . ( 1 990). Attending to form or meaning in the input. Studies
in Second Language Acquisition, 12, 287-3 0 1 .
Bibliography 287
VanPatten, B., & Cadierno, T. ( 1 993). Explicit instruction and
input processing. Studies in Second Language Acquisition,
1 5, 225-243 .
Velmans, M. ( 1 99 1 ). Is human information processing conscious?
Behavioural and Brain Sciences, 14, 65 1-726.
Vokey, J., & Brooks, L. ( 1 992). Salience of item knowledge in
learning artificial grammars. Journal of Experimental
Psychology: Learning, Memory and Cognition, 1 8, 3 2 8-344.
Voss, J.F. ( 1 986). Informal reasoning and subject matter knowledge
in the solving of economics problems by naive and novice
individuals. Cognition and Instruction, 3, 269-302.
Voss, J.F., Perkins, D., & Segal, J. ( 1 99 1 ). Informal reasoning and
education. Hillsdale, NJ: Lawrence Erlbaum.
Wason, P.C. ( 1 960). On the failure to eliminate hypotheses in a
conceptual task. Quarterly Journal of Experimental
Pschology, 12, 1 29- 1 40.
Watanabe, L. ( 1 980). Selective listening and attention. Japanese
Psychological Review, 23, 3 3 5-3 54.
Watanabe, Y. ( 1 992). Effects of increased processing on incidental
learning offoriegn language vocabulary. Unpublished M.A.
thesis, University of Hawai'i at Manoa, Department o f
English as a Second Language.
Wellman, H.M. ( 1 990). The child 's theory of mind. Cambridge,
MA: MIT Press.
Wesche, M. ( 1 9 8 1 ). Language aptitude measures in streaming:
matching students with measures and diagnosis of learning
problems. In K. Diller (Ed.), Individual differences and
universals in language learning aptitude, pp . 1 1 9- 1 5 4.
Rowley, MA: Newbury House.
Westney, P. ( 1 994). Rules and pedagogic grammar. In T. Odlin
(Ed.), Perspectives on pedagogical grammar, pp. 72-96. New
York: Cambridge University Press.
White, L. ( 1 987). Markedness and second language acquisition. The
question of transfer. Studies In Second Language Acquisition,
9, 26 1-285.
. (l 989a). Universal grammar and second language
-- acquisition. Amsterdam: John Benjamins.
. ( 1 989b). Linguistic universals, markedness and learnability:
-- Reconciling different approaches. Second Language
Research, 5, 1 27-1 40.
__ . ( 1 990). Second language acquisition and universal grammar.
Studies in Second Language Acquisition, 12, 1 2 1 - 1 3 3 .
 288 Bibliography

__ . ( 1 9 9 1 ). Adverb placement in second language acquisition:

Some effects of positive and negative evidence in the
classro om. Second Language Research, 7, 1 22-1 6 1 . S ubject Index
__ . ( 1 992). On triggering data in L2 acquisition: A reply t o
Schwartz and Gubala-Ryzak . Second Language Research, 8,
1 2 0- 1 3 7 .
White, L., Spada, N., Lightbown, P.M., & Ranta, L. ( 1 99 1 ). Input
enhancement and L2 question format ion. Applied Linguistics,
12, 4 1 6-432. Anderson's ACT* Theory of rule-based models, 99-
White, R. ( 1 988). The ELT curriculum: Design, management and skill acquisition, 52, 66, 1 0 1 , 207
innovation. Oxford : Basil Blackwell. 72-73 prototypes and, 99-1 00,
Whittlesea, B.W.A ., & Dorken, M.D. ( 1 993). Incidentally, things aptitude, 66, 77-79, 1 1 1 , 1 1 5- 202-203
in general are particul arly determined: An episodic-processing 1 1 6, 1 45-1 5 1 , 1 73-1 77 connectionism, 24, 5 3 , 5 8, 1 00-
account of implicit learning. Journal of Experimental and learning under 101
Psychology: General, 122, 227-24 8. different conditions, 1 70- consciousness, 1 , 9-1 4, 5 1-79
Wickens, C.D. ( 1 980). The structure of attentional resources. In R. 173, 1 77 as awareness, 9-1 0, 2 1 ,
Nickerson (Ed.), Attention and performance VIII, pp .23 9- and awareness under 5 1 -75, 1 70-1 77, 1 79-1 84
257. Hillsdale, NJ: Lawrence Erlbaum. different conditions, 1 73- as explicit knowledge, 1 0-
( 1 984). Processing resources in attentio n. In R. 1 77 1 1 , 26-27, 1 02-1 04
Parasuraman & D. Davies (Eds.), Varieties of attention, articulable knowledge, 1 0, 34- as intention, 9-1 0, 2 1
pp.63-1 02. New York: Academic Press. 3 5 , 9 1 , 1 4 1-143, 1 73-174, cross-modal transfer and
__ . ( 1 989). Attention and skilled performance. In D. Holding 1 80-1 82 learning, 208
(Ed.), Human skills, pp. 7 1 - 1 0 5 . New York, London: John artificial grammars, 6, 14, 1 5 ,
Wiley. 30, 82-87, 1 53 , 1 83-1 84, detection, 9, 53-54, 5 8-75
Yip, V. (1 994). Grammatical consciousness-raising and learnabi lity. 1 93 developmental stage, 3, 3 4, 49,
In T. Odlin (Ed.), Perspectives on pedagogical gramm ar, attention, 9-1 2 , 26, 5 1-75 , 78- 200
pp. 1 23-1 78. New York: Cambridge University Press. 79, 1 98, 202
Zalewski, J.P. ( 1 993). Number/person errors in an information as a filter, 53-54 effect of instruction studies, 5-
processing perspective: Implications for form-focused as capacity, 54-57 6, 1 99-200
instruction. TESOL Quarterly, 2 7, 69 1-704. short-term memory and,
Zobl, H. (1 980). The formal and developmental selectivity of L 1 57-67, 74-75 focus on form, 1, 2-5, 3 1, 1 99-
influence on L2 acquisition. Language Learning, 3 0, 43-57. automaticity, 1 0-12, 66--68, 69- 200
. ( 1 983). Markedness and the projection problem. Language
·
70, 74, 207 Fundamental Difference
Learning, 33, 292-3 1 3 . Hypothesis of adult second
. ( 1 985). Grammars in search o f input and intake. In S . Gass Bangalore Procedural Syllabus, language learning, 1 04-
-- & C. Madden (Eds.), Input in second language acquisition, 9, 19, 25-26 1 05
pp.329-344. Rowley, MA: Newbury House. Bialystok's Analysis/Control Fundamental Similarity
. ( 1 992). Sources of linguistic knowledge and uniformity o f Model of SLA, 1 1-12, 25 Hypothesis of
-- non-native performance. Studies in Second Language implicit/explicit second
A cquisition, 1 4, 3 8 7-403. categorization, 99-102 language learning,
__. ( 1 995). Converging evidence for the acquisition/learning exemplar-based models, 1 03-1 05, 1 55, 1 75-1 98
distinction. Applied Linguistics, 1 6, 3 5-5 7. 1 0 1-1 02, 207

290 Subject index
hypothesis-testing,
and confirmation bias, I 03,
1 9 1- 1 92
and overgeneralization,
1 92, 1 98
implicit knowledge, 1 0-12, 24-
26, 32, 33, 3 5 , 98- 1 03,
1 92- 197, 200
implicit learning, 1 0-12, 1 3-14,
29, 3 0, 8 1-92, 125-1 40,
1 53-1 79, 1 92-1 97, 1 99-
20 1 , 208
implicit memory, 65, 68-75,
92-98, 208
information processing,
and the implicit/explicit
distinction, 5 1-75, 79
individual differences in
resources and, 75-79
input-enhancement, I , 3 1 , 207
Kahneman's Capacity Model of
attention, 54-56
Krashen's Monitor Model of
SLA, I , 7-1 0, 26, 3 1-33,
64, 1 08-1 09, 1 84-1 85,
204-205
learning conditions,
implicit, 6, 1 7- 1 9, 88, 1 1 7,
1 92-1 94, 204
incidental, 6, 9, 1 6-19,
1 1 7- 1 1 8, 1 82-1 83, 195-
1 96, 204
instructed, 6, 1 6- 1 9, 2 1 ,
89, 1 1 8-1 1 9, 1 90-1 92, 204
rule-search, 2 1 , 88, 1 1 8,
1 97- 1 98, 204
Logan's Instance Theory of
automaticity, 1 0-1 1 , 66-
68, 69-70, 74, 207
Subject index 29 1
of gerundivization, 3 9-41
markedness, 3, 5, 33-34, 35, 5 5, of subjacency, 40-41
200 of third-person agreement,
maturational constraints, 1 3 , 3 1, 3 8
1 04-1 05 o f wh-question formation,
McLaughlin's Information
32, 33-34, 3 9-4 1
Processing Model of SLA,
1 0, 1 1-12, 52, 66
memory systems, 52-53, 64-78 task-based language teaching,
20, 3 1 , 53, 5 7
Natural Approach, 8-9, 1 9, 78 teacher knowledge, 36-38, 42-
negative evidence, 3, 4, 6, 207 44
noticing, 1 -2, 9-1 0, 27, 5 1 , 5 7-
Universal Grammar,
64, 77-79, 1 05, 1 4 1 , 1 55 ,
1 70-1 77, 1 82-1 83, 1 98- accessibility of, 4, 29, 33,
1 99 200-20 1
Noun Phrase Accessibility and explicit learning, 20 1
Hierarchy, 6, 1 7 and implicit knowledge,
1 0, 1 00-1 02
perceptual salience, 33-34, 3 5 and parameter-setting, 3,
physicalism, 24 24, 1 0 1 , 200
positive evidence, 4, 6, 207 and triggering, 3, 24, 1 0 1 ,
200
priming, 1 , 68-69, 78, 95, 96
rules, Wickens' Multiple Resource
and connectionism, 24, 32, Model of attention, 52-59,
1 00-1 02 76
and rule understanding, 9,
1 0, 25, 26-27
and rules of thumb, 27, 49,
20 1
complexity of second
language, 1 , 20, 28-49,
108- 1 1 0, 1 2 1-123, 200-
202
familiarity of pedagogic,
3 8-42
in linguistic theory, 23-24,
32-33, 36, 1 00-1 02
in second language
pedagogy, 23-49, 200
of article use, 3 1 , 32, 3 5
of focus, 39-4 1 , 44-49