Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Relationships between cognitive and linguistic processes and second language production in French immersion… Bournot-Trites, Monique 1996

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


831-ubc_1996-090450.pdf [ 10.32MB ]
JSON: 831-1.0054591.json
JSON-LD: 831-1.0054591-ld.json
RDF/XML (Pretty): 831-1.0054591-rdf.xml
RDF/JSON: 831-1.0054591-rdf.json
Turtle: 831-1.0054591-turtle.txt
N-Triples: 831-1.0054591-rdf-ntriples.txt
Original Record: 831-1.0054591-source.json
Full Text

Full Text

Relationships between Cognitive and Linguistic Processes and Second Language Production in French Immersion Programmes By Monique Gabrielle Marie Bournot-Trites M.A., The University of British Columbia, 1986 B.A., The University of British Columbia, 1983 A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES Department of Educational Psychology and Special Education We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA November 1995 © Monique Gabrielle Marie Bournot-Trites, 1995 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of E J ^ ^ - ^ P ^ e U l p J d v A . 5 ^ * 0 . t c l * * c * k The University of British Columbia Vancouver, Canada Date ^SflLiyVO^ l ^ i t a DE-6 (2788), Abstract Some researchers believe that all French immersion students speak the same way and that they make many transfer errors (borrowing English structures). Although these beliefs are not founded on rigorous observations, current teaching strategies emphasize repeating correct patterns rather than teaching according to student's individual differences. My study posed two main questions. First, are there individual differences in the quality of French spoken by French immersion students? Second, do individual differences in cognitive and linguistic processes influence the type and frequency of errors observed in French immersion students' language production? To analyse individual differences in linguistic processes, I used variables from verbal learning theory that measure Paradigmatic and Syntagmatic processes. I analysed differences in cognitive processes with the PASS (Planning, Arousal/Attention, Simultaneous and Successive processes) cognitive theory. PASS is operationalized in the Cognitive Assessment System test (CAS). These theories generated a secondary purpose for the study: to verify and explain the findings of Jarman's (1980) study concerning the parallelism between Paradigmatic and Simultaneous processes, and between Syntagmatic and Successive processes. Data came from 89 girls and 63 boys from anglophone families in grade three French ii immersion schools in British Columbia. I analysed four areas of individual differences (Paradigmatic and Syntagmatic processes; Simultaneous and Successive processes; Planning; and Attention), and the relationship between these processes with Vocabulary, Grammar, and Transfer Errors in oral and written French language production. LISREL measurement models did not converge due to variables' low correlation. Relationships between the variables were not found probably because of low reliabilities of the CAS measures. A subsequent analysis used observed variables' models (multiple regressions), rather than latent variables' models. The strongest relation was between Planning and Attention, and Transfer Errors. Grammar Errors, rather than Transfer Errors prevailed. Consequently, I suggested teaching methods reinforcing communicative sentence structures. As in Jarman (1980), cognitive and linguistic processes were not demonstrably parallel. Contrary to what verbal learning researchers thought, various linguistic processes' measures are not equivalent. Linguistic measures' constraint level as well as Planning level affected my results. For further research, I proposed a model of linguistic tasks of various constraint levels. iii Table of Contents Abstract ii Table of Contents iv List of Tables vii List of Figures viii List of Appendices ix Acknowledgement xi Chapter I. Introduction 1 Chapter IT. Review of the Literature 8 A. Research in Second Language Learning 8 a. Errors in Second Language Production 9 b. Language Errors in French Immersion Schools 12 Negative Transfer Errors: Definition 13 The Debate on Negative Transfer 14 Negative Transfer in French Immersion Classrooms 19 Grammar Errors other than Transfer Errors 22 Lexical Diversity and Vocabulary Errors 23 c. Control in Language 25 d. Recent Models of Second Language Acquisition and Production 31 e. Summary 37 B. Cognitive and Linguistic Processes 39 a. From Luria's Theory to the PASS Theory 39 b. Control of Linguistic Processing as Planning and Attention 50 c. Paradigmatic/Syntagmatic Linguistic Processes 51 Linguistic Processes: a Century of Research 51 Parallel between Linguistic Processes and Cognitive Processes 57 C. Summary 59 Chapter 111. Statement of the Problem 62 A. Overall Statement of the Problem 62 B. Rationale 64 C. Questions 71 Chapter IV. Method 72 A. Subjects 72 B. Tasks 72 a. Planning, Attention, Simultaneous/Successive Processes (PASS) Tasks 72 iv Visual Search 73 Planned Codes 75 Planned Connections 75 Simultaneous Verbal 76 Figure Memory 76 Matrices 76 Word Series 77 Sentence Repetition 77 Sentence Questions 77 Expressive Attention 78 Receptive Attention 78 Number Finding 79 b. Paradigmatic and Syntagmatic Linguistic Processes Tasks 80 Closed Word Association Task 83 Free Word Association Task 84 c. Language Tasks 85 Language Proficiency 86 Cloze Test 87 Composition 87 Grammar Diversity 88 Vocabulary Diversity 88 Grammar Errors 88 Transfer Grammar Errors 88 Grammar Errors other than Transfer 89 Vocabulary Errors and Transfer Vocabulary Errors 89 Oral Language Sample 89 C. Procedure 91 Chapter V. Results 94 A . Descriptive Statistics 94 a. Demographic Statistics 94 b. Descriptive Statistics of the Cognitive and Linguistic Variables 95 c. Descriptive Statistics of Language Measures 96 d. Sex Differences 98 e. Correlation Matrix 103 B. Preliminary Analysis 108 a. Principal Components Factor Analysis of the PASS Model 108 b. Distinction between PC A and Confirmatory Factor Analysis 113 c. Confirmatory Factor Analysis of the Variables of the PASS Model 114 C. Main Analyses 118 a. Structural Equation Modelling with LISREL 118 Multiple Measures of the same Concept 119 Observed Variables Reliability 119 Structural Comparisons of Latent Variables 120 v Comparison of Models 120 b. Cognitive Processes and Language Variables 121 c. Parallelism between Cognitive and Linguistic Processes 122 D. Reanalyses 127 a. Level of the correlations 128 b. Other factors 129 Normality 129 Sample size 129 Number of indicators per latent variable 130 Outliers 130 Transformations 130 Technical causes 131 Models 131 Scoring 132 c. Summary 133 d. Regression Analyses 133 E. Post Hoc Analyses 138 a. Homogeneity and Heterogeneity of the Task Processes 138 b. High General Ability and Precocious Development 142 c. Planning and Strategies 143 Chapter VI. Discussion 151 A. Main Analyses 151 a. General considerations 151 b. Parallelism between Cognitive and Linguistic Processes 152 c. Results about Language Variables 162 Descriptive Statistics 162 Statistical Analyses 164 B. A d Hoc Analyses 166 C. Summary of Main Findings 179 D. Limitations of the Study 180 E. Suggestions for Future Research 181 a. Second Language Learning 181 b. Planning 183 c. Task Specificity 185 References 186 vi List of Tables 1 Norms for the Written Tasks of the French Language Speaking and Writing Evaluation Units (1983) 91 2 Successive Eliminations from the Original Potential Sample to form the Final Sample . . . 94 3 Means and Standard Deviations (SD) of the PASS Model and the Linguistic Processes Variables by Sex and for the Total Sample 95 4 Means and Standard Deviations (SD) of the Language Variables by Sex and for the Total Sample 98 5 Results of t Tests for Boys and Girls for the Cognitive and Linguistic Variables . 100 6 Results of t Tests for Boys and Girls for the Language Variables 101 7 Correlation Matrix of all the Variables calculated from Transformed Data (Sex Differences removed) 107 8 PC A with Orthogonal Rotation of the Tasks from the Das-Naglieri Test 110 9 Design Matrix for the Confirmatory Analysis of the PASS Model Variables . . . . 114 10 Indices of Fit of the Confirmatory Analysis of the PASS Model Variables 115 11 Reliabilities of the CAS measures (SMC) 118 12 Indices of Fit of the Models testing the Relationships between Cognitive and Linguistic Processes 123 13 Principal Components Factor Analysis of the Simultaneous and Successive Processes Variables and of the Paradigmatic and Syntagmatic Processes Variables 126 14 Common Factor Analysis of the Simultaneous and Successive Processes Variables and of the Paradigmatic and Syntagmatic Processes Variables 127 15 Regression Analyses of Successive Processes Measures on Grammar Variables. . 135 16 Regression Analyses of Simultaneous Processes Measures on Vocabulary Variables. 136 17 Regression Analyses of Planning and Attention Measures on Transfer Errors. . . 137 18 Correlations between the Linguistic Processes Tasks of the Low Planning Group and of the High Planning Group 145 vii List of Figures 1. Path diagram for Syntagmatic Processes, Successive Processes, and Grammar Diversity and Errors other than Transfer 68 2. Path diagram for Paradigmatic Processes, Simultaneous Processes, and Vocabulary Diversity and Errors other than Transfer 69 3 . Path diagram for Planning, Attention, and Vocabulary and Grammar Transfer Errors. 70 4. Distribution of all errors between vocabulary and grammar 97 5. Distribution of errors between vocabulary, grammar, transfer vocabulary, and transfer grammar 97 6. Principal Component Analysis of the variables of the PASS model 112 7. Maximum Likelihood Confirmatory Factor Analysis of the PASS model 117 8. Models testing the correlation between, Simultaneous and Paradigmatic Processes and between Successive and Syntagmatic Processes 124 9. Graphs showing the effects of high and low Planning, paradigm used (Free Word Association and Closed Word Association), and the linguistic type of task used (Syntagmatic or Paradigmatic) 149 10. Distribution of the results on the four linguistic tasks, CAP, FAP, CAS, FAS for the total sample (n= 152) 150 11. Theoretical model showing the level of performance on various paradigmatic and syntagmatic linguistic tasks depending on the level of Planning 176 viii List of Appendices A 1. Visual Search 196 A 2. Planned Codes 197 A 3. Planned Connections 198 A 4. Simultaneous Verbal 199 A 5. Figure Memory . . : 200 A 6. Matrices 201 A 7. Word Series 202 A 8. Sentence Repetition 203 A 9. Sentence Questions 204 A 10. Expressive Attention 205 A l l . Receptive Attention 206 A 12. Number Finding 207 B 1. Closed Word Association Task 208 B 2. Free Word Association Task 209 C. Written Language Tasks 210 D. Oral Language Task 211 E 1. Letter of Contact to the Parents 212 E 2. Parent Consent Form 213 E 3. Background Information Questionnaire 214 F. Repartition of Errors in Oral and Written French Language Production 215 G. Correlation Matrix of All the Data (Calculated From Raw Data) 216 H. Examples of classification of language Errors 217 ix To my husband, Andrew, who met me in a bar and made me believe that my dream could come true. x Acknowledgement Starting a Ph.D late in life with concomitant responsibilities, such as a family to love and support, a job, research projects, a Scout group to manage, and all the additional chores that my gender accumulates by default, would be an impossible task to undertake alone. I would like to acknowledge and thank all who encouraged me, gave me support, helped me along this interminable and difficult path. A few lines can never do them justice. At the outset, I wish to thank the members of my committee. I would like to acknowledge the late Dr. Ron Jarman, my supervisor for five years. He was to me as Marpa was to Milarepa, a teacher of patience and modesty. Dr. Marshall Arlin and Dr. Marion Porath agreed without hesitation to join my committee late in the day. This did me honour and I feel indebted to them for their kindness and generosity. Their insight and input has been extremely valuable. I owe most to Dr. Nand Kishor who worked with me from beginning to end, when he became my supervisor. Dr. Kishor taught me structural equation modelling without counting his time. Although he had a full timetable, he always made himself available to me and was enthusiastic in helping me solve the many problems I encountered. His working power and his perspicacity always amazed me. Furthermore, Dr. Kishor enabled me and at the same time he knew when I needed a boost and gave it to me with the right intensity. I want also thank the people who administered tests with me in the schools, Catherine Blancard, Sylvie Kirouac, Dr. Buff Oldridge, and especially Sally Ann Lee who further helped me with coding. My thanks to the North Vancouver, Vancouver, and Richmond xi School Boards and all the participating schools. Special thanks to the students who worked so well with me and to all the teachers for their patience with my disruptions. Some people have been like pillars along my university road and I will never forget them. Some have been examples for me, others have said the right word at the right time, and others have given me lots of their time and energy. I would particularly like to acknowledge, Jane Flick, Dr. Carol Herbert, Nancy Horseman, Dr. Daniel Kahneman, Mike Kayo, Dr. Merv McArthur, Dr. Buff Oldridge, and Dr. Todd Rogers, I owe a special "thank you" to Dr. Patricia Arlin for being in turn a brilliant professor, a model, a mentor, an advisor, and a friend. I want to thank all my friends who believed in me, supported, me and helped me as much as they could. I want to thank them for being patient and accepting my lack of free time. Thank you to Kyme Wegrich who checked my numbers and did so many things to help me. A special acknowledgement goes to my long-standing friend and colleague Dr. William Bruneau who edited many drafts and spent hours discussing my research and related matters. A special thank you to my friend Debbie Johannesen, a nurse doing research on premature babies, who used both lines of her qualifications to help me deliver my big post-mature thesis. My companions on the cross deserve special acknowledgement. Thank you to the group of seven for being such brilliant scholars from whom I learned so much. A huge hug goes to my dear friend and colleague Patricia Lamarre for her invaluable sense of humour, kindness, and wisdom. Finally, a very special thank you goes to my friend, running partner, and colleague Dr. Ross Barbour, who became my mainstay during these last two years. Thank you also to Atsuko, his wife, for all of her homemade jam. xii My parents merit a distinctive recognition for their role in my achievements. My mother has an immense reverence for education and never had the chance to go to school to study. She became an autodidact and kept repeating to me that education was the surest way to independence and women liberation. My father agreed knowing that my mother could not take her own advice. I know that the best acknowledgment I can give her is to graduate with a Ph.D from an English-speaking University. I thank my parents-in-law from the bottom of my heart for always being very encouraging and positive towards my endeavour. I deeply regret that both my father and father-in-law have not lived long enough to see me finish. At last and more than anyone else, I thank my immediate family: my sons, Vincent and Daniel, for being extremely understanding and for accepting to make many sacrifices to allow me to study; and especially Daniel, who has been a great companion and has used his innate wisdom so often to show me the perspective I had lost or to give me the courage I needed. Finally, my husband Andrew deserves a whole paragraph to himself because I would have never done it without him. I want to thank Andrew for seeing that I could do it, for showing me the way, and for making it happen. Although Andrew was as financially poor as a student can be, he put me through school, became an instant father to Vincent, and paid off all my debts. Thank you, Andrew, for being strong, stable, understanding, and so loving. xiii Chapter I. Introduction Over the past two decades, linguists and pedagogues have compared the results of French immersion students on French achievement tests to those of French as a second language students and native French speakers (Day & Shapson, 1987; Lapkin, 1984; Lyster, 1987; Pellerin & Hammerly, 1986). These studies show that French immersion students speak fluently and score highly, although not at native-speaker levels, on most measures. However, while French immersion students are comparable to Francophones on the communicative aspect of speech such as the amount of information transmitted, they still differ in the frequency and type of grammatical errors they make (Day & Shapson, 1987). Furthermore, Hammerly (1982) claims that French immersion students speak a "faulty interlanguage" made up of structures borrowed from the English language. A detailed description of the kinds of errors found in French immersion classrooms is contained in Lyster (1987). New pedagogical approaches and reforms of educational programs have been proposed to improve the quality of French spoken and written by French immersion students (Nemni, 1985; Pellerin & Hammerly, 1986). Most of these proposals adopt a behaviourist input-output model based on achievement and on premises such as: "if we present the correct elements of language to the students, they will produce sentences without errors" or "if we teach grammar more systematically, students will produce more grammatically correct 1 sentences." Such interventions assume an experimental perspective, manipulating variables in the learners' environment and ignoring potential systematic variation among learners that could explain achievement. As Cronbach (1957) states, "Individual differences have been an annoyance rather than a challenge to the experimenter. His goal is to control behaviour, and variation within treatments is proof that he has not succeeded" (p. 674). Reforms of educational programs and new pedagogical approaches generally should stem from the results of research. However, studies on French achievement such as the ones mentioned above give global information only on French achievement and relate the quality of French to the methods of teaching. The current studies on which proposals are based bypass important considerations in the question of second language production through several areas of deficiencies or limitations. First, they don't describe individual differences among students' results. They consider French immersion students as a homogeneous group. Nonetheless, it is doubtful that immersion students make the same errors with the same frequency. There is no discussion of range in the frequency of errors, nor is there discussion of variability of any sort among students. For example, qualitative studies such as Lyster's (1987), list and classify errors made by French immersion students, while quantitative studies globally compare the results of French immersion students to native speakers, or to other French-program students. Thus, it does not come as a surprise that the proposed programmes are a direct consequence of these global observations and of inferences made from them rather than on correlational studies analyzing individual differences. They propose to change the general organization of programmes or the methods of teaching in order to improve the quality of written or spoken French. Curriculum advisors have to take into account both sets of variables included in the equation, environmental variables as well as learners' variables. The role of the latter set of variables can be uncovered with correlational studies of individual differences. Modern pedagogy, such as the policy document Year 2000 (1990) and its most recent alterations from the British Columbia Ministry of Education, emphasizes individual instruction. It is the role of studies of individual differences to inform pedagogues on the most appropriate decisions on matters of education. The importance of environmental variables should not be discounted, but rather both disciplines, "experimental psychology" and "correlational psychology", should complement each other (Cronbach, 1957). Secondly, models of second language acquisition which drive the studies are not grounded in a general theory of cognition. The aforementioned studies do not ask questions about the various cognitive processes associated with the results of French achievement tests. Only the programmes or the pedagogy are considered as possible causes for the results. Individual differences in cognitive and linguistic processes among French immersion students and their relationship to language production do not seem to have been explored. It is therefore not surprising that various pedagogies are tried one after the other and do not result in considerable improvement. As Crookes (1988) notes, with respect to second language (SL) pedagogies: The conception of the swinging pendulum, of one fashion succeeding another has been put forward by critics of SL pedagogy, and it is probably accurate because SL pedagogy has 3 not been informed by, nor developed within a scientific model of knowledge development. For that reason, little or no progress could be made. (p. 43) Bialystok (1988) has done a study related to individual differences and second language production. She showed that levels of linguistic awareness—the ability to step back and reflect on one's own language—correlate positively with levels of bilingualism. This study is important in that it shows a first step in analysing learners' cognitive processes. However, loose definitions of levels of bilingualism and methodology problems prevent definitive conclusions about the relationship of linguistic awareness and quality of second language production. Research linking linguistic awareness as defined within a general cognition model and types of errors (transfer and others) would (1) demonstrate the role of linguistic awareness in the quality of second-language production and (2) provide a starting point for more effective pedagogy. Furthermore, a review of the literature in second language learning and processing shows that a study of the relation of linguistic awareness and linguistic processes as well as cognitive processes to the types and frequency of errors in Canadian French immersion classrooms is needed in order to guide pedagogical decisions. Most recent models of second language learning and processing (Bialystok, 1978; Ellis, 1985; Lamendella, 1977) are reminiscent of information processing and artificial intelligence concepts. They represent, using a series of boxes, a temporal succession of 4 identified processes taking place between the intention of the speaker and the production of utterances. Each model takes into account one aspect of human cognitive behaviour such as the planning aspect in Ellis's model, or memory and consciousness in Bialystok's model, or metasystems in Lamendella's model. As Ellis (1985) remarks, "Ultimately, a theory stands or falls according to its ability to explain the available facts" (p.55). Unfortunately, many such models could explain the same key issues in second language learning or production. A model which could be validated by its correspondence to the human neuropsychological system as a whole as well as by its explanatory power of linguistic behaviour would be more accurate. Only a reduced number of models could fit neuropsychological functioning as well as linguistic production. Furthermore, such a model would have to take into account complex human behaviour rather than one specific aspect of it. The PASS theory (Planning-Attention-Simultaneous-Successive) accommodates the preceding requirements. It is a general model of human cognitive processes rooted in Luria's theory of neuropsychological functioning. Kirby (1980, cited in Kirby & Das, 1990) indicates that, "PASS theory offers a plausible description of cognition, and of the cognitive processes involved in achievement tasks. Combined with theory-based assessment, this provides the basis for prescriptive theory-based instruction" (p. 329). Additionnally, Luria's neuropsychological theory of cognitive functioning strongly associates language with cognition. More specifically, two distinctive linguistic processes recognized by developmentalists and linguists (for example, Galton, 1883; Jakobson, 1971; Jenkins, 1954) 5 have been identified independently by Luria (1973a) in his neurolinguistic studies. These are Paradigmatic and Syntagmatic Linguistic Processes. In a word classification task, Paradigmatic Processes are used when words are classified by categories whereas Syntagmatic Processes are used when words are classified on the basis of concrete or situational associations. For example, in the list, "wine, glass, milk" a classification of the words milk and wine together are the result of Paradigmatic Process for classification, while the classification of milk or wine with glass is the result of Syntagmatic Processes (Bournot-Trites, Jarman, & Das, 1995). Such processes have been shown to be related to two general cognitive processes of the PASS theory, namely Successive and Simultaneous Processes (Jarman, 1980a). A framework of general human cognition and cognitive processes, based in neuropsychology, appears very promising to explain second language production. Such a unique framework is found in the PASS theory and was used in the present study to explain types of errors in second language production. The purpose of this study was to analyze individual differences in second language production within a general theory of linguistic processes, namely Syntagmatic and Paradigmatic Processes, as well as to analyze these individual differences within a broader theory of cognitive processes based in neuropsychology, namely the PASS theory developed by Das, Kirby, and Jarman (1975, 1979). The Cognitive Assessment System (CAS), now being developed by Das and Naglieri, measures the constructs proposed in the theory (Planning, Attention-arousal, Successive and Simultaneous Processes). By way of contrast to studies such as those of Pellerin and Hammerly (1986) or Lyster (1987), the cognitivist approach of the present study based on a psychometric measurement method emphasizes individual differences in cognitive and linguistic processes and conceptually linked variables in second language production. It provides a theoretical basis for devising programs to help students achieve higher standards in their second language. I analyzed four areas of individual differences: (1) Syntagmatic and Paradigmatic Linguistic Processes, (2) Simultaneous and Successive Cognitive Processes, (3) Planning, and (4) Attention, and related each of the differences to individual differences in French language production, especially errors. These areas were analyzed through a study of 89 girls and 63 boys from grade 3 (age 7 to 9) French immersion classrooms and anglophone families of the British Columbia Lower Mainland. 7 Chapter n. Review of the Literature A . Research in Second Language Learning French immersion in Canada started thirty years ago when a group of anglophone parents from Montreal came to believe that learning French would be easier, more interesting, and more efficient were it taught through the presentation of content. In 1965 a primary class of anglophone children from St Lambert, Quebec, was taught all subjects in French. The result was deemed a success. Anglophone pupils were able to speak French fluently by the end of the first year with no apparent disadvantage in other academic subjects including English. French immersion programmes became very popular across Canada and in 1975 the first French immersion classroom in British Columbia opened in Coquitlam. Since then, enrolment has grown at the approximate rate of 25% per year until 1990 when it reached a plateau. However, some opponents of French immersion discount the fluency in French acquired by French immersion pupils and criticize the incorrectness of their French (for example, Hammerly, 1989). Significantly, the goal of French immersion programs was functional bilingualism and not perfectly balanced bilingualism. In that sense, the programmes succeeded. The main concern of critics of French immersion is that an incorrect way of speaking and writing French is so ingrained in French immersion pupils that they will never be able to speak correctly. This may be so. Yet, before dismissing French 8 immersion, which has the advantage of bringing pupils to a fluent communicative stage, it is essential to analyze the types of errors made by French immersion pupils and to see whether theories or models of second language acquisition and production could explain both such errors and individual differences among pupils according to the type of error. Such an analysis would cast additional light on the problem. I here describe the various types of errors made by French immersion pupils and discuss the explanatory capacity of recent models of second language acquisition and production, regarding errors. a. Errors in Second Language Production Errors can be classified along various taxonomies. They can be classified according to some observable surface characteristics such as errors based on linguistic categories (e.g., syntax and morphology, semantics and lexicon, phonology, and discourse) or they can be classified with reference to their underlying cause or source. Unlike the objective classification of errors based on linguistic category, classification of error based on sources is a difficult enterprise because it is subjective. First, the specification of the source of the error is not always straightforward. For example, a single error could be attributed to various single sources or even to multiple sources. Second, since the attribution of errors to sources is based on inferences, the theoretical perspective of language learning and/or acquisition held by the researcher doing the classification plays an important role in the decision. Third, the learning context is also a key factor in the classification of errors according to the inferred source. In particular, the learning context 9 interacts with the learner and one error could be attributed to one source in one learning context and to another source in another learning context. Taxonomies of sources of errors in second language production are numerous. For example, at one end of the spectrum Brooks (1964), who holds a behaviourist view of second language learning, gave four possible causes of learners' errors: (a) the student may make a random response, that is, he may simply not know which of many responses is the right one; (b) the student may have encountered the model but not have practised it a sufficient number of times; (c) distortion may have been induced by dissimilar patterns in English; or (d) the student may have made a response that follows a sound general rule but, because of an anomaly in the new language, is incorrect in this instance, (p. 58) At the other end of the spectrum, Dulay, Burt, and Krashen (1982), who hold a nativist view of second language learning, gave four causes of learners' errors as follows: These comparisons have yielded two major categories in this taxonomy: developmental errors and interlingual errors. Two other categories that have been used in comparative analysis taxonomies are derived from the first two: ambiguous errors, which are classifiable as either developmental or interlingual; and, of course, the grab bag category, other, which are neither, (p. 164) Dulay, Burt, and Krashen define developmental errors as "errors similar to those made by children learning the target language as their first language" (p. 165), interlingual errors as "similar in structure to a semantically equivalent phrase or sentence in the learner's native language" (p. 171), and ambiguous errors as those which "reflect the learner's native 10 language structure, and at the same time, they are of the type found in the speech of children acquiring a first language" (p. 172). Brooks' (1964) taxonomy reflects the idea that language learning is the result of "the formation and performance of habit" (p. 49), whereas Dulay, Burt, and Krashen (1982) put the emphasis on "the developmental processes in L2 acquisition" (p. 164). Nowadays it would be unthinkable to entertain Brooks' pure behaviourist view of language learning. Modern pedagogy does not reflect that view. Languages are not learned by the formation of habits through repetition, especially in French immersion classrooms where learning is more implicit than explicit following the model of first language acquisition. Errors are not considered as being the result of low habit strength or sloppy thinking but as being learners' rule-based productions which do not correspond to the target language. The problem of error classification according to sources has been extensively discussed by Dulay, Burt and Krashen (1982): "Any researcher who attempts to use an error taxonomy to posit sources of errors must make a number of difficult and ultimately arbitrary decisions in order to attribute a singular source to an error." (p. 144). For the purpose of this study, a classification of errors based on linguistic categories was used. As for the classification according to the inferred source of error, only the distinction between interlingual errors and other errors was made. In particular, only the errors that could be inferred as originating in the first language of the learners were distinguished from the others. 11 b. Language Errors in French Immersion Schools The context of this study is French immersion which is a special case of second language learning. Second language learning can occur in various pedagogical contexts ranging from explicit learning of rules to natural acquisition of the language. For example, Core French programmes generally consist of one hour of French language teaching three times a week. In this case, grammatical rules and vocabulary are taught formally and the use of French language is closely monitored. In contrast, in French immersion programmes content is taught in French and only a few hours a week, if any, are dedicated to the formal teaching of French language. Grammar rules and vocabulary are implicitly learned by the students from input data and first-hand communication experience. The pedagogical model follows closely the acquisition of a first language; however, it differs from various points of view. In particular, the learners have the advantage of already speaking a language and of being cognitively more developed than when they acquired their first language. They are at a disadvantage from the point of view of input data which is not as rich as it is for their first language where the entire society speaks the language more or less correctly. In French immersion classrooms, the only sources of correct input data are the teacher and the pedagogical material. French immersion pupils make errors of various linguistic types when speaking French. As for the source of these errors, the most documented and obvious type has been called negative transfer errors, or interlingual errors. These errors occur when second language speakers inappropriately fall back on their first language in speaking French. 12 However, French immersion pupils also make intralingual errors, including developmental errors as do all children when learning their own language. The latter are rule-based and cannot be traced to the first language of the learners. In the following, I first define "negative transfer," and I present the debate that arose over that concept in second language research. Second, I discuss other types of errors, those that cannot be inferred as originating in the first language of the learners. Negative Transfer Errors: Definition Odlin (1989) defines the phenomenon of transfer as follows: "Transfer is the influence resulting from similarities and differences between the target language and any other language that has been previously (and perhaps imperfectly) acquired" (p. 27). In this study, I considered negative transfer from English to French in grade 3 French immersion classes in Canada. In other words, I was concerned with those errors which come from the transfer of English grammatical structures into the French language (for example, "j'aime toi" instead of "je t'aime," the equivalent of saying "I you like" instead of "I like you"), idiomatic expressions (for example, "je suis huit ans" instead of "j'ai huit ans" the equivalent of saying "I have eight years" instead of saying "I am eight years old"), or vocabulary (for example, "je vais checker" instead of "je vais v,rifier," the equivalent of saying "He razed this morning" instead of "He shaved this morning"). Although positive transfer also occurs in French immersion classrooms, when pupils appropriately use their 13 first language knowledge and apply it to second language, I did not consider that for two reasons: first, it is not a source of error, and second, it would be difficult to substantiate. The Debate on Negative Transfer In the history of second language learning research, two hypotheses may explain errors of second language learners-the contrastive analysis hypothesis (CA) and the creative construction hypothesis (CC). These hypotheses stem from a wish to be able to predict errors and therefore to prevent them if possible. The CA hypothesis, whose principal proponent is Lado (1957), states that children have a tendency to use structures from their first language when learning a second language and that children make errors when second language structures differ from those in the first language. For example, anglophone children have learned to say "I am eight" in their mother tongue and in a literal translation they say "je suis huit." However, the French structure expressing the same idea does not use the auxiliary "to be" but rather "to have." Furthermore, because French uses the auxiliary "to have" it is essential that the number be followed by the word "years," otherwise the interlocutor would not know what it is that the person has. This gives the correct expression "j'aj. huit ans," which is quite different from "je suis huit." The latter expression not only is incorrect but does not carry the meaning intended by the speaker. The contrastive analysis hypothesis has its roots in behaviourist theory. In this 14 perspective language learning is seen as habit formation through modelling and reinforcement. A particular stimulus becomes linked to a particular response to the point of automaticity. When habits are formed, these automatic responses are transferred to second language and create errors when the response that should be linked to the stimulus is different in French and in English, as shown in the previous example. In contrast, the CC hypothesis states that children actively organize second language samples they hear and make generalizations about the second language as they did when learning their first language. Dulay and Burt (1974) say that "the child's organization of L2 syntax does not include transfer from (either positive or negative) or comparison with his native language, but relies on his dealing with L2 syntax as a system" (p. 115). Dulay and Burt (1973) claimed that 85% of errors in second language were developmental. As an illustration, the way French immersion pupils in some cases use past participle forms of verbs in French offers a good example of a developmental error known as overregularization, or overgeneralization (Berko Gleason, 1985). In French, verbs are classified in three groups of which the first group contains the largest number of verbs. The infinitive form ends in "er" and the past participle ends in ",." Often, French immersion speakers use their knowledge about first-group verbs to apply to other groups. Therefore, when they use the verb "prendre" ("to take" which is from the third group) in the past participle form, they sometimes say "prend," instead of "pris." This is the equivalent of forming the past of all English verbs by adding "ed" at the end of the verbs, which produces such forms as "I buyed" instead of "I bought." In contrast to the CA hypothesis, the CC 15 hypothesis is rooted in nativist theory and developed when Chomsky (1959) wrote an attack of Skinner's Verbal Behaviour. Chomsky used the argument of "poverty of stimulus" to show that language learning could not be explained by habit-formation. Imitation and reinforcement could not account for the creativity found in language. Chomsky's ideas were directed at first language acquisition, but they were subsequently applied to second language acquisition. As Ellis (1989) indicates, "These criticisms of behaviourist learning theory were directed to begin with at LI acquisition. However, they soon spread to SLA [Second Language Acquisition]" (p. 30). According to the CC hypothesis for example, learning a second language is innate and follows a natural development independent of knowledge of other languages (Dulay, Burt, & Krashen, 1982). In contrast with the behaviourist view of the CA hypothesis and the nativist view of the CC hypothesis, Faerch and Kasper (1987) offer a cognitivist view. They think learners use accumulated linguistic knowledge as input in creative construction, of which their first language is an important source of knowledge. The concept of "strategy" becomes important in this perspective. The learner is no longer passive, subjected to the power of habit-formation as in the CA hypothesis, or to natural developmental processes as in the CC hypothesis, but rather takes an active role in language production. Thus in the examples above, when speakers say "je suis huit" instead of "j'ai huit ans," they are not seen as subjected to habit-formation but as active users of their first language knowledge applying it to a second language in order to express themselves. When they say "prend," instead of "pris," they actively use their knowledge about some verbs in a second language and 16 generalize it to other verbs. These are strategies speakers use in order to communicate. As a digression, it is interesting to note that there has been some confusion in the literature about the distinction between processes and strategies and that the terms have been often used interchangeably. For example, for Kirby (1984), "A strategy is essentially a method for approaching a task, or more generally attaining a goal. Each strategy would call upon a variety of processes in the course of its operation" (p. 5). In contrast, Bialystok (1990) does not make such a distinction between processes and strategies. She concludes her analysis of communication strategies by saying, Strategies are a normal and fundamental aspect of ordinary language processing. They are rooted in the same processing mechanisms as is nonstrategic language use. They are the adjustments to the ongoing processes responsible for language acquisition and use that allow processing to be maintained, (p. 146) In my study the term "process" was used to designate the general mental operations associated with language. It included attention, encoding, transforming, and storing of information as well as planning and controlling functions. Selinker's (1969, 1972) interlanguage hypothesis reconciles the CA and CC hypotheses. "Interlanguage" is a term coined by Selinker (1972) and refers to second language in development that is neither equivalent to the first language nor to the target language but follows systematically its own rules. Selinker considers the various sources of interlanguage development: linguistic transfer, reorganization of linguistic material, learning 17 strategies, transfer-of-training, and strategies of communication are the five elements responsible for the interaction between first and second language and the learner's cognition. This interaction results in interlanguage, an erroneous language compared to target language. However, according to Selinker this erroneous language is permanent because of "fossilisation," a characteristic of interlanguage. Fossilisation represents the "permanent [italics added] characteristics of the speech of bilinguals irrespective of the age at which the second language is acquired or the amount of instruction or practice in it" (Richards, 1973, p. 117). Therefore, according to the principle of fossilization, second language development stops before it is identical to the target language. Interlanguage is a new language with a life of its own. It has its own rules which are not the rules of the first language, nor those of the target language, but rather the result of an interaction between the two languages and the creativity of the speaker. This presents difficulty for second language teachers. For them, what Selinker calls "interlanguage" is simply second language with errors that have to be corrected. Yet, if it is the case that these characteristics or errors are permanent, there is little hope for an efficient pedagogy capable of changing the interlanguage into target language. The fossilization hypothesis could be supported only if no counter-examples could be found. However, Taylor (1975) shows second language learners make fewer negative transfer errors as they become more bilingual. Tomasello and Herron (1989) have further shown that "the Garden Path Technique" corrects negative transfer errors; that is, when students learn a new sentence for which transfer would not produce an adequate translation, 18 they are first allowed to make the negative transfer error, which is immediately corrected by the teacher, instead of being given the correct sentence in second language and shown the difference between the two languages before trial. Research shows, then, that errors caused by negative transfer are difficult but not impossible to correct. The CC and CA models both explain errors in second language production with overgeneralization. CC states that most errors are developmental and are the sign of overgeneralization or lack of discrimination in learning within second language. CA states that errors are the sign of overgeneralization of vocabulary or grammar rules across languages. For our purpose, these models are useful because they show that intralingual errors as well as transfer errors are present in second language production. As they are defined presently it is not possible to submit these models to a direct test. However, the main problem with both models is that they do not include subject variables. As a consequence they predict the same outcome for all speakers. Subject variables could explain individual differences regarding errors in second language production. Negative Transfer in French Immersion Classrooms Negative transfer in French immersion classrooms has been studied qualitatively (Lyster, 1987) and quantitatively (Pellerin and Hammerly, 1986). Lyster (1987), working in a grade 8 French immersion classroom, describes errors due to negative transfer in the use of verbs, pronouns, prepositions, gender, and the distinction 19 between "tu" and "vous." These transfer errors support the CA hypothesis. However, Lyster also lists intralingual errors resulting from overgeneralization of second language rules (cf. Dulay & Burt, 1974), a finding that makes Selinker's interlanguage hypothesis more tenable than the extreme CA or CC hypotheses. Lyster made his observations in his classroom over a period of five years, drawing evidence from compositions and classroom discussions. His list of errors is representative of French immersion classrooms. Pellerin and Hammerly's study (1986) is more quantitative than Lyster's study. Unfortunately, their sample included only 6 students from grade 12. Moreover, they conducted group interviews (2 or 4 students together) for a duration of 40 minutes. Their results show that 52% of students' utterances contained grammatical errors. Pellerin and Hammerly note that their results for grade 12 students are comparable to Spilka's (1976) grade 6 students. They make the argument that if grade 12 students make as many errors as grade 6 students, this confirms Selinker's fossilization hypothesis. Furthermore, they make inferences about the cause of their results: The assimilation of erroneous grammatical rules: here is the disastrous result of the incessant use, year after year, of faulty grammatical forms or structures without their being sufficiently or efficiently corrected, (translated from Pellerin et Hammerly, 1986, p. 593) Hammerly's studies on French immersion present some methodological problems which have been outlined in a review (Bournot-Trites, 1991) of his book: French immersion: Myths and reality. Similarly, the Pellerin and Hammerly study presents some flaws. First, their sample is so small that it is not possible to generalize the results to the entire French 20 immersion population. Moreover, the comparison with Spilka's study is not a good argument for the fossilization hypothesis since secondary French immersion students are in a different linguistic context from that of primary students. Secondary French immersion students are usually in anglophone schools and take most of their courses in English. Consequently, they have fewer occasions to speak French than do primary students. This is also well documented by Lyster (1987, p. 705). In fact, we could reasonably expect an increase in the number of negative transfer errors children make as they move from grade 6 to grade 12. Pellerin and Hammerly's results point to linguistic context rather than to fossilization. Yet, neither Lyster nor Pellerin and Hammerly tell us whether all students make the same errors with the same frequency. On the one hand, Lyster says that French immersion students speak "Immersion" as though these students were using a new linguistic code, as they would a dialect or an interlanguage, similar for every member of French immersion classrooms. On the other hand, Hammerly blames the French immersion system as a whole, based on the results of his study with six students. To reach such a conclusion, he has to assume that the frequency of errors reported in his studies is the same for all French immersion students and that there are no individual differences. Such a viewpoint is not defensible. It is probable that individual differences among speakers exist when their productions are compared for errors. For example, some speakers would be found to make more errors in general than others. Furthermore, when errors are classified by type, profiles of speakers' errors could be compared and found to differ. Within-subjects comparisons 21 would probably show that some speakers make more errors of one type (negative transfer, grammar, vocabulary) than of other types. Additionally, the analysis of individual differences in the frequency and type of errors would permit us to concentrate on the cognitive and linguistic processes leading to the results described in studies such as Lyster's or Hammerly's. Grammar Errors other than Transfer Errors As mentioned earlier, grammar errors can be classified in different ways. Lyster classified errors according to whether they could be explained by negative transfer. He classified the errors caused by negative transfer as crosslingual errors. He called errors not explicable by negative transfer, "intralingual." These are "caused by the students' tendency to simplify and overgeneralize the morphological rules of the French language based on what they already know" (Lyster, 1987, p. 711). Because they lack knowledge of grammatical rules or have an incomplete knowledge of the rules, French immersion pupils simplify or overgeneralize. For example, if they do not know how to conjugate a verb, they use it in its infinitive form (simplification) or conjugate it as verbs of the first group, which are the first learned, the simplest, and the most numerous in French language (overgeneralization). Simplification brings the utterance closer to inner speech than to the fully developed and grammatically correct external speech. It is more or less comprehensible to the interlocutor since it contains what the speaker wants to communicate about an object, an event, an attitude or an action. According to Luria (1976) inner speech is the translation of an initial thought into a speech pattern. In other words, all the ideas are there, grammatically or not. 22 Corder (1967) offered another classification according to the linguistic knowledge of the speakers. Following this principle he classified errors into two groups, competence errors and performance errors. He called the former "errors" and the later "mistakes." Mistakes take place when learners know the rules but fail to apply them, whereas lack of knowledge of the grammatical rules results in errors. Although the end result is the same, performance errors, or mistakes are different from competence errors in that they result from learners' lack of control over linguistic production and lack of integration of knowledge with production. Johnson (1988) argued that if performance errors are due to learners' inability to produce correct utterances under difficult circumstances, a large percentage of our students' malformed utterances are mistakes and not errors. He takes the view that French immersion speakers know the rules but are subjected to the power of habit in using the structures of their first language. Although this distinction is relevant to French immersion pupils' language production, it is difficult to assess with certainty whether a speech production is an error or a mistake. I have thus used the term "error" in this paper while being conscious that it may represent mistakes as well as errors. Individual differences in type and frequency of grammar errors could be explained by individual differences in various cognitive processes. Intralinguistic grammar errors especially, could be linked to Syntagmatic Linguistic Processes and Successive Cognitive Processes. I explain this hypothesized relationship later. Lexical Diversity and Vocabulary Errors Second language vocabulary of most second language speakers is not as well developed 23 as is their native language vocabulary. This has been shown by Vasos (1983, cited in Segalowitz, 1986) who "found that semantic activation was less extensive and shorter in duration in the second language than in the first for moderately skilled bilinguals" (p. 11). Luria (1976), describing the process underlying speech communication, explains that the activation of a multidimensional matrix generating a complete semantic net of all the words associated with the idea being discussed by the speaker, is at the basis of lexical selection. Thus, according to Luria (1976), the speaker must select some words and inhibit others. However, in some cases a word has been insufficiently reinforced by rehearsal and different words appear to the speaker with an equal probability of being selected. Furthermore, Luria indicates that this problem is known as the "tip of the tongue" phenomenon (Brown & McNeill, 1966). Transposing this theory to second language speakers in French immersion programmes means that lexical access to French words could present such a problem for words insufficiently heard, used, or rehearsed. In such cases, the English equivalent word would have the greatest probability of being selected. In the French immersion classroom, when a word is unknown in French or has been rarely used or heard, it is common for pupils to use the English equivalent with a French pronunciation. This is a case of negative transfer when the word is different in French, as in the example of "checker" instead of "v,rifier" given earlier. Besides, a lack of vocabulary knowledge can also translate into a lack of vocabulary diversity. For example, second language speakers use a restricted vocabulary and compensatory strategies based in their second language (L2) knowledge (Ellis, 1989, p. 24 182). They sometimes replace difficult items by easier and less precise alternative forms (e.g., "animal" instead of "rabbit"). They also repeat words and use analytic paraphrases (e.g., "tool for hitting with," instead of "hammer") or even make up words (e.g., "picture place" instead of "gallery"). Individual differences in vocabulary errors as well as vocabulary diversity could also be explained by individual differences in cognitive processes, especially Paradigmatic Linguistic Processes or Simultaneous Cognitive Processes. These hypothesized relationships are discussed in greater detail later since such processes have been shown to be related to the nominative part of speech. c. Control in Language The control aspect is an important variable in the negative transfer problem. In this research, control was viewed as the capacity to intervene at the moment of production in order to either correct negative transfer errors as soon as they are produced, or not make these errors at all. In French immersion classrooms, students generally know the correct forms but automatically make errors when they speak or write because they tend to generalize the structures of the English language in which they are familiar. Note that these would be called "mistakes" in Corder's framework (1967) as suggested previously. The communicative approach of French immersion classrooms puts the emphasis on the message rather than on the form; therefore, teachers do not insist on correcting errors when the meaning is clear. It may be the case that emphasizing differences in forms could help them become aware of differences in grammatical structures between the two languages, and control the tendency to use English structures they have practised at length. 25 If it could be shown that the number of negative transfer errors correlates negatively with the learner's degree of control, teachers could use pedagogical means to improve their pupils' degree of awareness and control over problematic grammatical structures. The metacognition literature identifies two cornerstones: control and awareness, the latter of which is sometimes called monitoring (for a detailed review, see Jarman, Vavrik, & Walton, in press). Nelson and Narens (1990) give a clear description of what is meant in the literature by the two main metacognitive processes, control and monitoring. They indicate that cognitive processes are split into two levels, the meta-level which is the executive level and the object-level which is the content upon which the executive acts. Information flows from one level to the other and the two main metacognitive processes, control and monitoring, are defined by the direction of the information between the two levels. In the case of monitoring, the information flows from the object-level to the meta-level, whereas in the case of control, the information flows from the meta-level to the object-level. As a result, in monitoring "the meta-level is informed by the object-level" and in control "the meta-level modifies the object-level" (Nelson & Narens, 1990, p. 127). Relating these metacognitive processes to bilingualism, Bialystok (1988) claimed that levels of linguistic awareness ability-the ability to step back and reflect on one's own language—correlates positively with levels of bilingualism. However, her classification and labelling of the processes is not in direct correspondence with the classification in the metacognition literature. Although she distinguishes between control and awareness, which 26 she calls analysis of linguistic knowledge, she classifies both of these processes under the label of linguistic awareness. Bialystok's model follows an information processing framework comprising three types of linguistic knowledge sources necessary for language acquisition, comprehension or production. These are implicit linguistic knowledge, explicit linguistic knowledge, and other knowledge. Her model, in which the degree of linguistic awareness and control are important, is well suited to the study of the phenomenon of negative transfer errors in French immersion classrooms. In two studies, Bialystok (1988) associated different levels of bilingualism with different levels of linguistic awareness defined as control of language processing and analysis of linguistic knowledge. She identified two skill components necessary for metalinguistic tasks and differentially necessary in specific tasks. These are analysis of linguistic knowledge and control of linguistic processing which she defined as follows (1988, p. 561), This analysis component corresponds to the memory base, which becomes organized into networks, schemata, or systems (crystallized ability). Development involves making explicit the structure of the mental representation that had initially been stored in some holistic or implicit form...Control of linguistic processing is the executive component responsible for directing attention to the selection and integration of information. Her subjects were at three levels of bilingualism: unilingual anglophones, partial bilinguals, and full bilinguals. She found that bilinguals had better control over processing of language than monolinguals. However, she found no difference in control among bilinguals. She also found that full bilinguals scored higher than partial bilinguals or monolinguals on analysis of linguistic knowledge. In French immersion, explicitly knowing the rule is not sufficient: students must resist using the English structure when the French is different. Therefore, 27 control is a variable worth studying in the French immersion context. However, methodological problems underlying Bialystok's results should cause us to interpret them carefully. Two methodological problems arise in Bialystok's study (1988). First, her operational definition of bilingualism in both studies may confound the results. Second, her measure of control of language processing does not include enough items to be reliable and valid. In her first study, Bialystok investigated differences between grade 1 monolinguals, partial bilinguals (French immersion pupils), and full bilinguals (Programme Cadre pupils, that is, Francophones outside of Quebec in French schools) on control of processing and analysis of linguistic knowledge. She defined levels of bilingualism not with a language test but by reference to students' learning contexts. The distinction between French immersion and Programme Cadre (French schools for Francophones outside of Quebec) is more political than linguistic. Often, French school students outside of Quebec come from families where English is spoken most of the time, since the criterion for admission in those schools is that one parent be Francophone or have attended a French school for at least one year during childhood. This criterion opens the door to many anglophones and as a result, there are few differences between French schools and French immersion schools. Bialystok verified that the level of bilingualism of her three natural groups was different by administering them the Peabody Picture Vocabulary Test (PPVT). She gave two versions (English and French) to the bilingual groups. As she found a significant difference between 28 the French immersion group and the Programme Cadre group, she concluded that their level of French was significantly different. However, the PPVT focuses on receptive vocabulary and does not give information on level of syntax. The French group, since one of their parents is French, might have heard more French vocabulary and scored higher on the PPVT in French than the other groups. At the same time their linguistic production might be equivalent to that of a person in the French immersion group, and they might make as many negative transfer errors as their peers in French immersion. In Bialystok's study full bilinguals and partial bilinguals could be equivalent groups as far as level of bilingualism is concerned. This would explain why she found no difference on control between full and partial bilinguals. In her second study, Bialystok (1988) investigated differences within grade 1 bilinguals on control of linguistic processing and analysis of linguistic knowledge. Her subjects were children who had varying exposure to Italian in their homes and who went to school in English. Their degree of bilingualism was measured using the PPVT translated into Italian. In that second study, significantly, none of the subjects could read Italian. Although this would be expected for a minority language at grade 1 level, it shows that their level of bilingualism was probably at the low end of the bilingual range. Furthermore, their results on the English and Italian PPVT showed that they were more proficient in English than in Italian. Variables other than level of bilingualism may confound the results. As Bialystok discusses, it is possible that degree of literacy in the home was responsible for the differential results on the PPVT in Italian and for the results on the tasks measuring analysis of linguistic knowledge. 29 In fact, in her first study where the three groups were similar in socioeconomic factors, Bialystok found no conclusive difference in analysis of linguistic knowledge between the three groups. She measured analysis of linguistic knowledge with three tasks. For one of the tasks (word judgement where the subjects were presented with a list of 10 words and asked to say if they were words or not), she found no difference between the three groups. For the two other tasks (definition of what a word is, and syntax correction), the full bilinguals outperformed the partial bilinguals and the monolinguals but there was no difference between the latter two groups. Since the level of socio-economic status was equivalent for the three groups, this result could be explained by a difference in pedagogy in the Programme Cadre where grammar is systematically and formally taught whereas it is not in the other programmes. From her results I conclude that a high level of analysis of linguistic knowledge may be explained by a high level of literacy in the home, or a different pedagogy at school. Thus, some of Bialystok1 s results could be an artifact of the sampling procedure rather than an effect associated with levels of bilingualism. Control, the second metacognitive process, is more definitely associated with bilingualism. Although Bialystok found no difference on control among bilinguals, she found a difference between bilinguals and monolinguals. This result is questionable because of the sampling of the three groups and also because of the way control was measured in the study. To test the level of control of linguistic processing, Bialystok used a task comprising only two items, Piaget's (1929) sun/moon problem adapted by Ianco-Worrall (1972). The first item was Piaget's original sun/moon problem, and the second item was apparently 30 equivalent (dog/cat) but resulted in different findings. Before concluding with assurance that levels of bilingualism are related to levels of control of linguistic processing and analysis of linguistic knowledge, such methodology problems must be resolved. Bialystok (1988) mentions these problems in her conclusion, "Further research must examine these factors more carefully, possibly using longitudinal study, and certainly developing more rigorous methods for definition and measurement of the relevant factors" (p. 567). There is a need for a design measuring level of bilingualism with a language test. Furthermore, relating frequency of negative transfer errors to levels of control of linguistic processing with a task comprising more than two items would be more suited to the understanding of the role of cognitive processing in second language production. d. Recent Models of Second Language Acquisition and Production Because such metalinguistic processes as control of linguistic processing and linguistic awareness are possible variables related to levels of bilingualism, I now examine how recent models of language acquisition and production include these variables and discuss their ability to explain second language errors. I show that second language models have been influenced by the advancement in metacognition. With the development of cognitive science and metacognition research, metalinguistic aspects of language started to appear in the recent 31 models of language acquisition and production under various labels such as monitoring, linguistic awareness, control of language processing and planning. Among second language acquisition researchers, Krashen (1978) is well known for his "Monitor Model". Krashen has made the distinction between learning and acquisition of a language. Acquisition is "a subconscious process identical in all important ways to the process children utilize in acquiring their first language" (Krashen, 1985, p. 1) and learning is "a conscious process that results in 'knowing about' language" (1985, p. 1). According to Krashen, the only function of learning is that of a Monitor, which can alter speech productions before or after they are written or spoken. Krashen's notion of the monitor contains both metacognitive processes identified in the metacognition literature. The Monitor holds both functions of awareness and control. In contrast, monitoring is only synonymous with awareness in the metacognitive literature. However, according to Krashen, three conditions must be met before the Monitor can function: speakers must have time to think and consciously use linguistic rules, they must focus on form, and finally they must know the rules. Krashen dismisses learning in favour of acquisition by showing that the three requirements for the Monitor to be effective are rarely met in language communication. Further, learning cannot be transformed into acquisition. Krashen's hypotheses have been used to support immersion pedagogy. Since conscious learning was declared useless, no systematic teaching of grammar was advocated in French immersion schools. Following Krashen's ideas, pupils acquire rules naturally through communication. In the same way, Krashen used the results of French immersion to show his hypotheses were 32 confirmed. Krashen's hypotheses have been criticized, especially by McLaughlin (1987), mainly because there is almost no evidence to support them apart from Krashen's arguments. The main problem comes from the fact that it is impossible to isolate empirically which part of speech is learned and which part is acquired. Yet Krashen's concept of the Monitor was so close to ordinary language usage that it "became the mainstream of second language research" (Crookes, 1988, p. 26). Almost every article in second language literature refers to Krashen's model although it has never been empirically supported. As Crookes (1988) notes, "Interest in the topic resulted subsequently in a large number of studies in this area, though comparatively little insight into the underlying issues emerged" (p. 26). Information processing approaches including metacognitive processes followed Krashen's theory and have been more successful. The influence of the information processing perspective transpires in Ellis' definition (1989) of second language acquisition, "the way in which the learner works on samples of the input data, converting them into intake and then using his knowledge to produce output." (p. 249) Bialystok's (1978) model is in this new generation. She distinguishes three knowledge sources for language production: implicit knowledge, explicit knowledge, and other knowledge. In her model, learning can be transformed into acquisition through conscious practice. Only the explicit knowledge is the source of monitoring and correction of errors 33 of utterances. However, she notes that, "Linguistic knowledge source is associated with extensive knowledge of formal aspects of the language but does not necessarily imply an ability to use this information," (1978, p. 73). Although knowledge and awareness are necessary, as Bialystok has shown in her model, they are not sufficient for correct production in second language. The executive system of control requires planning, the intention of using linguistic knowledge and a plan including attention to execute this intention. Oddly enough, planning has not been given much attention in second language research except in the "Variability Model" of Ellis (1985). Ellis's model is based on the concept of discourse continuum developed by Ochs (1979). The variability in speech stems from whether the discourse is planned or unplanned. Two types of procedure correspond to the two types of discourse: primary processes and secondary processes. When the discourse is unplanned, primary processes such as semantic simplification (omission of elements in a sentence) are in action, whereas secondary processes such as monitoring (which Ellis defines as editing) or borrowing (deliberately using first language knowledge and applying it to second language) are in action when the discourse is planned. Although Ellis does not define the terms of his language processing model clearly, he shows the importance of planning in language production and its variability. However, he equates planning with attention. As Crookes (1988) remarks, "the admittedly close relationship between planning and attention is oversimplified to one of equivalence" (pp. 16-17). 34 Although these recent models are coloured with the new trends in research, each model contains only some aspects of metacognitive processes. Further, these concepts are defined differently in each model and the definitions are abstract. The processes or parts of the models are not operationalized and this makes their measurement and empirical verification difficult, if not impossible. Another problem with these models of second language acquisition and production is that they are not embedded in larger models of cognition developed in the disciplines of psychology and neuropsychology, or even in generic models of language learning. As Crookes (1988) demonstrates in his report, The study of second language (SL) learning is obviously an interdisciplinary endeavour. However, its development has been hampered by a tendency in the SL field until recently to be inward-looking-to base theories or (usually more accurately) conceptual models on an unduly narrow reading of applied linguistics research, and of the major contributory disciplines...The general psychological processes involved in learning (whether unconscious or not) and the existing psycholinguistic mechanisms possessed by the learner (by virtue of speaking a first language) must be taken into account by SL researchers...Previous failure to do this has led to an inadequate treatment of two important cognitive processes in language: monitoring and planning, (p. 1) In the conceptual models I described earlier, the notions of monitoring or linguistic awareness and planning are not related to research on metacognition in psychology and more specifically in cognitive science. They are only defined by their meaning in ordinary language. In addition, establishing a relation between the models of second language 35 acquisition and production and research in neuropsychology could offer a further validation of these models. Lamendella's (1977) model seems promising in this regard. In his paper, Lamendella presents his hypothesis "about the organizational character of the systems underlying the learning of a language other than one's native language" (p. 156). He tries to link neurofunctional organization to language acquisition. Lamendella's distinction between primary language acquisition and nonprimary language acquisition is central to his model. The former refers to unconscious acquisition of a language, were it first or second, whereas the latter refers to the learning of a second language which he labels "foreign language learning." According to Lamendella, a metasystem for communication underlies primary language acquisition, whereas a metasystem for cognition underlies nonprimary language acquisition. During language development, differential infrasystems (information frames and skill schemata) are constructed within each metasystem. Infrasystems are defined as "functional constructs of brain systems derived in relation to particular environmental experience" (p. 159). Lamendella's model emphasizes the distinction between conscious learning and unconscious acquisition and shows the influence of Krashen's ideas. However, what is new in his model is the attempt to link these concepts to neurological structures in the brain. In his 1977 paper Lamendella presents a list of general principles of neurofunctional organization from which he derives manifestations of these general principles in primary and 36 secondary language acquisition. For example, one of the principles of neurofunctional organization is, "A given system, subsystem, or metasystem may form part of more than one functional hierarchy and/or carry out more than one function." (Lamendella, 1977, p. 160) The manifestation in primary and secondary language acquisition he derives is, There is every likelihood that the communication hierarchy and the cognitive hierarchy share subsystems in common and are highly integrated. These two functional hierarchies interact in different ways at different levels. In addition to their role in interpersonal communication, language systems possess internal coding functions in the symbolic representation of external reality. (Lamendella, 1977, p. 161) Notwithstanding Lamendella's effort to link second language to neuropsychology, he fails to give references for the principles he presents and his list of manifestations are only hypotheses based on informal logic and deduction as seen in expressions such as "there is every likelihood." His model is based on speculations for which he does not offer supportive studies, and although he uses the vocabulary of neurology and tries to attach his representation of language to physical structures of the brain, he offers no validation for this possible correspondence. This leaves the model at the stage of an hypothesis still to be verified. In contrast, the cognitive model of the present study is rooted in neuropsychology and validates the correspondence between the processes composing the model and neurofunctional systems (see p. 36 and following pages). e. Summary In summary, previous studies examining the quality of French language spoken in French immersion classrooms have resulted in various taxonomies of errors. However, the 37 main limitations of these studies reside in the fact that they ignore individual differences in the frequency and type of errors and do not facilitate the study of cognitive or linguistic processes leading to these errors. Differential cognitive and linguistic processes could be associated with Vocabulary Diversity and Errors, Grammar Diversity and Errors, and Transfer Errors. Intensive research in cognitive science and particularly in metacognition has had some ripple effect on recent models of language acquisition and production. Such processes as monitoring, planning, and control of linguistic processing as well as analysis of linguistic knowledge have begun to appear in various models or theories of second language acquisition and production. However, none of the models includes all of the processes and only a few offer operational definitions of the processes of which they are made. Furthermore, the definitions given in these models are not linked to human cognition theories or to findings in neuropsychology, nor do they correspond to the definitions generally accepted in the metacognition literature. Only Lamendella sought to connect his second language acquisition model to neurological structures; however, he offered no validation for his hypothesized links. In order to explain errors in French immersion classrooms, a study of individual differences using a model based on cognitive and linguistic processes underlying language production is necessary. Such a model, based in neuropsychological theory and research is used in the present study. One of the characteristics of the model used here is that it is rooted in the socio-38 historical tradition. In this model, higher mental faculties are based on functional systems which are social in origin (Luria, 1980). Following Luria, other writers have said that "Intellectual behaviour is a covariate of learned behaviour" (Hynd & Willis, 1985, p. 132). If we accept this notion, the study of cognitive and linguistic processes resulting in errors in French immersion classrooms should help curriculum developers and pedagogues in their task. It should help them to implement teaching and learning strategies directed at the development of important cognitive processes in second language learning. B . Cognitive and Linguistic Processes a. From Luria's Theory to the PASS Theory In the present study I analyzed individual differences in second language production as they relate to individual differences in levels of cognitive processes through the PASS model of Das, Kirby, and Jarman (1979). It is an alternative model of intelligence based on the work of the late Soviet neuropsychologist, Luria. In this section, I explain the origins of the model, describe the model, and present evidence of its validity. Earlier models of individual differences were based on factor-analytic methodology. The construct of intelligence was equated to that of ability. However, different researchers had different conceptions of these abilities. Some used a hierarchical model (e.g., Cattell) whereas others, such as Guilford, used a non-hierarchical model. There was also a lack of 39 consensus on the number of abilities included in the construct of intelligence. For example, Spearman isolated one general factor g', whereas Guilford proposed 120 intelligences. Because of the lack of consensus, a paradigm shift took place. Keeping the same methodology, or in some cases using new methodology (for example, Sternberg's componential analysis), researchers presented new conceptual structures. With the advent of cognitive psychology and theories of information processing, they became more interested in mental processes concerned with learning, thinking, and achievement, and individual differences in level of activity of these processes. The PASS model is part of this new research movement on individual differences in intelligence. It views intelligence as a series of cognitive processes and is theoretically rooted in cognitive psychology and in the work of Luria in neuropsychology. It is based on factor-analytic methodology and is concerned with cognitive functions or processes rather than abilities. The first articles on the model were published in 1975 and the model is still under development. The PASS model has its roots in the work of Luria, who used syndrome analysis as a research method. Luria (1966, 1970, 1973a, 1973b) examined the behaviour of people who had brain lesions and made inferences about the cortical zones responsible for some specific factors common to the various disturbed behaviours which resulted from a lesion. He identified three principal functional units of the brain, which are dynamic and interact with each other. According to Luria, the three units are necessary for each and all tasks. 40 However, the proportional involvement of each functional unit may differ from task to task. The first unit, which regulates the state of cortical activity and the level of alertness, is situated in the brain stem and the reticular formation. The second unit of the brain is responsible for the reception, analysis, and storage of information and is located in the lateral regions of the neocortex. The structures of the third functional unit which is responsible for programming, regulation, and verification are situated in the anterior regions of the hemispheres. Luria indicates that disturbance in information integration is associated with damage to the second functional unit of the brain. He distinguishes between two kinds of integration, namely simultaneous integration and successive integration. Simultaneous integration refers to the synthesis of separate elements into groups, whereas successive integration refers to the processing of information in a serial order. Simultaneous processing is required to reproduce a design for example, whereas successive processing is required to repeat a list of words in order. Simultaneous as well as successive processes are not modality specific (Jarman, 1980b). Visual information can be processed successively as well as simultaneously. Verbal information requires simultaneous and successive coding. In terms of localization, the simultaneous processing corresponds to the temporo-parieto-occipital zone of the brain, and the successive processing is associated with the fronto-temporal areas of the brain. Based on Luria's work, Das, Kirby, and Jarman (1975, 1979) researched and developed an alternative model for intelligence which was later named the PASS model 41 (T4anning, Attention, Simultaneous, Successive cognitive processing model) by Naglieri and Das (1988). This model is constituted of four blocks: Attention and Arousal, Simultaneous coding, Successive coding, and Planning, corresponding to the three functional units of Luria. Each block is dependent upon the knowledge base and functions at the memory, conceptual, and perceptual levels. There is interaction between each of the parts. The model can be classified as an information processing model, although the processing of information is not linear as in the information processing models of memory, for example. By confirmatory analysis, various tasks have been shown to load on the four different factors. The tests for Arousal-Attention were Selective Attention-Receptive and Stroop Colour-Word Interference Card. The planning tasks were Visual Search, Matching Numbers, and Trail Making. The simultaneous tasks were Figure Memory and Matrices. The successive tasks were Sentence Repetition and Word Recall. Currently the model is operationalized into a standardized test, the Cognitive Assessment System (CAS) being developed by Das and Naglieri, which gives scores corresponding to the four parts of the theoretical model (see detailed description in the methodology section). The validity of the PASS model is based on the parallel between syndrome analysis in neuropsychology and factor analysis in cognitive psychology. The conceptual model is derived from findings and theory in neuropsychology through this validation. Syndrome analysis could easily be mistaken for phrenology, developed by Franz Gall at the beginning of the nineteenth century. However, it is quite different. Franz Gall thought he could localize complex psychological processes in the brain. We have learned that only elementary 42 functions can be localized in precise areas of the brain: for example, motor speech is localized in Broca's area. Complex psychological functions, such as language in general, cannot be reduced to a specific part of the brain. As Hughlings-Jackson wrote in 1915, "To locate the damage which destroys speech and to locate speech are two different things" (p. 81). A disturbance in behaviour is a sign of a disturbance in the brain, but says nothing about specific localization of the centre causing the appearance of the disfunction. Each behaviour is the result of the activity of a complex functional system made of different components or factors, working together. These components can be located in completely different and often far distant areas of the brain. To make inferences about the way these complex psychological functions work, neuropsychologists use syndrome analysis, whereas cognitive psychologists use statistical analyses such as factor analysis. Syndrome analysis consists of examining various disturbed behaviours as a result of specific brain lesion. By analysing the symptoms in a large number of patients with the same brain lesions, neuropsychologists can find the common factor lying behind the observed symptoms. The syndrome (a constellation of signs and symptoms) must be subjected to a complex structural analysis before neuropsychologists come to a conclusion. Disturbance in disparate behaviours can be caused by a faulty common factor among them. Luria (1973b) shows that differing psychological processes such as arithmetical calculations, spatial orientation, and the understanding of complex logico-grammatical structures can be united into a single group of psychological processes. He indicates that researchers must resist the attempt to seek the direct localization of mental processes in the cortex. 43 Luria remarked (cited in Das, Kirby, Jarman, 1975) that there is a close similarity between syndrome analysis and factor analysis. In syndrome analysis the number of subjects is one and the experiment has to be replicated a large number of times before a reliable conclusion can be reached. Factor analysis is based on variations due to individual differences; therefore, many tasks are administered to a large group of people. In parallel with syndrome analysis Das, Kirby, and Jarman (e.g. 1975) used factor analysis of the results of large numbers of individuals on many different tasks to validate their cognitive model. First, they had to identify groups of tasks involving mainly the same cognitive processes. Luria developed his theory from his examination of the common aspects of various tasks that were associated with lesions in specific areas of the brain. Das, Kirby, and Jarman analyzed tasks according to Luria's model in terms of the main processes (among the four mentioned earlier) necessary for the solution of these tasks. Some of the tasks they chose to statistically validate their cognitive model are the same as those used by Luria. For example, Luria used Figure Copying extensively to show disturbance in simultaneous processes. Their choice of tasks was not random, but rather, was based on Luria's model, and the tasks chosen were closely related to those used by Luria. Then they hypothesized that the tasks would load on the factors according to the classification done from a logical and analytical point of view. The tasks classified as depending mainly on the same factor should load on that factor and be shown to be mainly under the control of the same cognitive processes. When it is established that specific tasks show the functioning of specific processes, different performances on these tasks indicate different levels of the 44 underlying cognitive process. While Luria used reasoning and logic to make inferences about a great number of observations in his experimental work, Das, Kirby, and Jarman used analysis of tasks followed by statistical analysis, and more specifically factor analysis, to discover the same cognitive processes in normal subjects with individual differences being the source of variation in their analyses. As noted here, syndrome analysis and factor analysis are parallel methods of investigation of cognitive processes; however, there are some differences between the two. First, the source of variation is the type of brain lesion in syndrome analysis, whereas it is individual differences in factor analysis. Syndrome analysis uses multiple tasks on single individuals with an infinitely large number of replications, whereas factor analysis uses some tasks on many individuals. Syndrome analysis presents some difficulties. It necessitates a very large number of observations of individuals with the same type of lesion in order that researchers can be confident of the results. Furthermore, the brain shows resilience and plasticity (some parts of the brain can compensate for parts where the lesions occur). This phenomenon causes some paradoxical results and makes it difficult to identify with certainty localization of cognitive process. Factor analysis also shows some limitations and the use of this method for studying processes that constitute intelligence is highly contentious. The use of factor analysis for the study of intelligence has been criticized by Sternberg and Baron, for example. Sternberg (1977, Chapter 2) criticized the globality of the method and Baron (1987) argues that the 45 method cannot distinguish between the determinants of the correlations (for example, correlations in educational opportunities compared to common cognitive processes). Carroll (1988) tempered these criticisms by proposing proper designs to eliminate confusion on the findings of factor analyses. Even Sternberg (1977), who wrote an article against factor analysis, now is more willing to recognize the value of this statistical method in the field of intelligence (Carroll, 1988). The PASS theory is supported by a large number of substantial studies. Task selection was done with confirmatory analyses whereas data analyses were done with quasi-confirmatory factor analyses or confirmatory factor analyses. None of the factor analyses done for the PASS model is of the exploratory type, because the model was driven by Luria's neuropsychological model of cognitive functions. Quasi-confirmatory analyses have been done to confirm that the same patterns were observed from syndromes in neuropsychology to factors in statistical analysis. True confirmatory analysis has been used to select tasks and to test the PASS model against other models of intelligence (Naglieri, Das, Stevens, Ledbetter, 1991). As to task selection, confirmatory analysis should be distinguished from non-confirmatory analysis. In non-confirmatory analysis tasks are chosen randomly and the results of these tasks from a large number of subjects are submitted to factor analysis. The obtained factors are then interpreted according to the model hypothesized. In contrast, in the PASS model confirmatory analyses were used for task selection. The tasks chosen were 46 extensively used by Luria, or tasks analyzed as requiring the same cognitive processes. Thus, factor analysis was confirmatory, in the sense that the new tasks were confirmed to measure the same cognitive functions as Luria's. As I indicated previously, data were analyzed with quasi- or true confirmatory analyses. When the authors of the PASS model used quasi-confirmatory analysis, they analyzed the cognitive demands of various tasks according to the findings of Luria and developed hypotheses about the loading of each task on the factors. For example, Figure Copying was hypothesized to load more strongly on the simultaneous processes factor, whereas Forward Digit span was hypothesized to load more on the successive processes factor. A number of tasks were analyzed and classified according to their main cognitive demands. If the theory is correct, all the tasks which hypothetically make more demands on the same cognitive process should load on the same factor when the data from tasks results are submitted to factor analysis. Furthermore, the number of factors obtained should number four to correspond to Luria's four cognitive processes. Therefore, after careful analysis, the tasks were administered to various populations and the data were factor analyzed. The reason they should be called quasi-confirmatory analyses and not confirmatory analyses is that the results from these kinds of factor analyses were not submitted to a statistical test but to researchers' judgements. In contrast, non-confirmatory analyses of data would be the same kind of analyses but without theoretical basis. For example, in the PASS 47 quasi-confirmatory analyses, Forward Digit Span consistently loaded on the successive processes factor along with other tasks such as Sentence Repetition as predicted from Luria's findings and theory. In the same way, Figure Copying consistently loaded on the same factor as Figure Recognition which was hypothesized to load on the simultaneous processes factor. This kind of confirmatory analysis enabled researchers to establish the bridge between neuropsychology findings and statistical findings. Moreover, the number of factors expected from Luria's theory was found. The model was thus shown to be viable. However, the authors had still to validate their new model against other models in the field. This second stage in research necessitates a test of statistical significance. True confirmatory factor analysis is a variety of factor analysis that establishes statistical significance. In the confirmatory maximum-likelihood model the data matrix is rotated on each of the target matrices representing competitive models. The fit of the data is verified with a chi square test. Concurrent validity of the PASS model was established statistically using the confirmatory maximum-likelihood model by Naglieri, Das, Stevens, and Ledbetter (1991). The model was tested against three competing models: a verbal/spatial/speed division of the tests, a memory/reasoning division, and a g' model. The data were shown to fit the PASS model best. This kind of analysis represents a second step in research which complements the first step using quasi-confirmatory analysis. From a statistical point of view, true confirmatory analysis is more stringent and statistically correct than quasi confirmatory research. However, one kind of analysis is not 48 better than the other and quasi-confirmatory analysis has its place in model development. It establishes a strong basis from which more sophisticated statistical tests can be done. It is as important for a model to be validated across fields of research as well as within its own field. Quasi-confirmatory analysis and true confirmatory analysis each have their place along a continuum of the development of a model of intelligence and cognitive processes. At the moment, the PASS model is being operationalized through the Cognitive Assessment System and will encompass more cognitive processes than previous intelligence tests. Because the test gives results on a greater range of processes than previous intelligence tests it will have diagnostic power. It will enable school psychologists not only to establish the competence of the individuals compared to the general population of the same age in terms of processing of information, but will also offer guidance for remediation as well. In summary, the PASS model has good internal, external, and predictive validity. However, the internal validity would be improved with more experimental studies of the type described by Das (1984a). Furthermore, studies at the micro level would permit more refinement of the model and would reconcile its findings with other theories based on experimental studies at the micro level, such as Sternberg's triarchic theory of intelligence (1979). The PASS model seems very appropriate for the study of second language production. It is a general cognition model validated by findings in neuropsychology and by statistical 49 analysis. The fact that these processes have an operational definition and can be measured allows study of the role of general cognitive processes in second language production. It will be shown hereafter that control in second language production could be accounted for by two functional units, namely planning and attention/arousal. Furthermore, the two parts of the coding unit (simultaneous and successive) could be shown to have an important role in the grammar and vocabulary aspects of second language production. b. Control of Linguistic Processing as Planning and Attention As I have mentioned in the preceding chapter, control of linguistic processing plays an important role in the quality of language production. This variable should be defined and operationalized in the framework of the PASS model. Control is comprised of two elements: first, based on awareness or intention, a plan has to be established on what to control; secondly, attention is necessary if the plan is to be executed at the right time, and correctly. In Luria's functional organization of the brain, the third functional unit located in the frontal lobes is responsible for planning and the first functional unit located in the brain stem is responsible for attention. Luria (1973b, p. 67) noted the close connection between the two units, The systems of the first functional unit not only maintain cortical tone but also, themselves, experience the differentiating influence of the cortex, and the first functional unit of the brain works in close cooperation with the higher levels of the cortex, (p. 67) Planning and Attention are blocks three and one respectively in the PASS model. Their level can be assessed using the corresponding tasks of the CAS. It is, therefore, 50 possible to verify whether negative transfer errors are correlated with level of control, and more specifically with planning and attention. This was not possible with Ellis's model where planning was not operationalized. Furthermore, Luria (1966) indicated that the third functional unit is responsible for activities such as impulse control, and linguistic functions such as spontaneous speech. He explained how his patients with lesions in the frontal lobe had difficulties finding right words and were unable to inhibit the production of irrelevant words that came to their mind: It may be suggested that fluent spontaneous speech that stems from an appropriate plan and is not based on a ready-made assortment of images requires rather active attempts to find the necessary words as well as an inhibition of irrelevant words that involuntarily spring up. That is why patients with particularly severe lesions of the frontal lobes who attempt to find a particular word required in active speech have so weakened a process of inhibition that they are unable to suppress irrelevant associations and resort to paraphasias on a wide scale, reproducing words of similar sounds but inappropriate meaning, (p. 517) This description of behaviour shares a striking similarity with that observed in French immersion classroom. It warrants an investigation of the relationship between negative transfer errors and planning and attention as operationalized in the PASS model by the CAS. c. Paradigmatic/Syntagmatic Linguistic Processes Linguistic Processes: a Century of Research Besides planning and attention, a consideration of the linguistic processes known as 51 Paradigmatic and Syntagmatic could help us to understand the mechanisms underlying grammatical or vocabulary errors others than transfer as well as grammar and vocabulary diversity. Furthermore, findings about the relationship between linguistic and cognitive processes could bring even more insight into the understanding of the mechanisms underlying second language production. In the following section, I define the Paradigmatic and Syntagmatic Processes and describe research and findings concerning them in three traditions, namely linguistic, developmental and neuropsychological. I then explain how these linguistic processes can be related to second language errors. Linguistic, developmental and neuropsychological research independently suggest the existence of Paradigmatic and Syntagmatic Linguistic Processes. First, in the linguistic tradition, Jakobson indicated that the Paradigmatic and Syntagmatic Linguistic Processes underlie two fundamental operations of our verbal behaviour: selection and combination (Jakobson, 1971), At the lexical level this is readily apparent: the speaker selects words and combines them into sentences according to the syntactic system of the language he is using; sentences in their turn are combined into utterances, (pp. 51-52) In combination "any linguistic unit serves as context for simpler units and/or finds its own context in a more complex linguistic unit" (p. 51). Combination is based on contiguity and context. It is reflected in the ability to use syntactic rules to organize words into higher units, and to "propositionize" (Hughlings-Jackson, 1879). In combination or contiguity 52 disorders, the ties of grammatical coordination and subordination are dissolved. Such a loss is called agrammatism. On the other hand, selection implies the possibility of substituting one word for an equivalent word; therefore, substitution is linked with the concept of similarity, which ranges from synonyms to antonyms and with the concept of metaphor. Jakobson (1971) studied aphasia from a linguistic point of view and came to the conclusion that: A linguistic reinterpretation of the varied cases described in the multilingual literature on aphasia, as well as observations made by myself on aphasics of different languages, made it clear to me that we have to deal with two basic types of aphasia. Either the internal relation of similarity and correspondingly the selective ability is impaired or conversely, the external relation of contiguity and, hence, the capacity of combination appears to be affected, (p. 97) Jakobson indicated that in normal behaviour both processes are continually operative, but careful observation will reveal that under the influence of a cultural pattern, personality, and verbal style, preference is given to one of the two processes over the other. In the developmental tradition, Paradigmatic and Syntagmatic Processes were first noticed in the analysis of free word-association tests (Galton, 1883). Various classification schemes have been used in the literature. A criterion of paradigmatism and syntagmatism was proposed by Jenkins (1954), and recently used by Beauvois and Ghiglione (1981) and cited in Pons and L'homme (1983, p. 412): "Two words are considered paradigmatically similar to the extent that they are substitutable in the identical frame.. .and syntagmatic to the extent that they follow one another in utterances" (p. 412). Some authors such as Mefferd, 53 Dufilho and Dawson (1979) consider that syntagmatic responses involve not only syntactic contiguity but also functional, spatial, temporal contiguity as well as descriptions of objects denoting the words. In other cases, Brown and Berko (1960), and Ervin (1961), used a simplified classification of answers. Paradigmatic and syntagmatic responses were classified on the basis of whether they were of the same grammatical class as the stimuli words. Responses made of words of the same class (for example, table - chair) were classified as paradigmatic whereas responses made of words of a different class (for example, table - eat) were classified as syntagmatic. Such free word association tests allowed observation of language development: younger children (7 or 8 years old or younger) generally gave answers of a different grammatical class from the stimuli, whereas older children mostly gave answers in the same grammatical class. This shift, designated as Syntagmatic-Paradigmatic Shift, has been observed in various studies (for example, Entwistle, 1966; Entwistle, Forsyth, & Muus, 1964; Ervin, 1961; Palermo & Jenkins, 1963; Palermo, 1971). White (1965) analyzed the changes in learning taking place between 5 and 8 in various domains, including the shift to paradigmatic word associations and saw them as "a broad spectrum of change" (p. 208). He concluded that "the material on the various shifts in the 5-7 age period may define something about the structure of adult mental processes" (p. 215). White noticed that the appearance of the regulatory role of speech on behaviour as described by Vygotsky (1934, 1962, 1965, 1978) may be the linking element between the findings in the literature about the various shifts. He also adopted Luria1 s (1980) idea of higher and lower functional levels of behaviour 54 mediated by higher and lower brain structures: The reorganization of mental activity by means of speech and the incorporation of the system of speech connections into a large number of processes, hitherto direct in character, are among the more important factors in the formation of the higher mental functions, whereby man, as distinct from animals, acquires consciousness and volition. (p. 32) White (1965) proposed a two layer model of cognitive processes composed of an "associative level" and a "cognitive level," both available to adults. The faster associative mode having precedence over the cognitive mode, and conditions which influence a subject's tempo of response and/or his ability to restrain a first available response will influence which mode determines his response, (p. 216) Luria's and White's quotations show how closely linguistic and cognitive processes are related in that tradition. Development occurs through a dialectical interaction between language and cognition. A striking parallel is found between the work of Jakobson (1971) and Luria's (1973a) neurolinguistic clinical research. Luria (1901-1977) was a Soviet psychologist who first worked closely with Vygotsky and added his training in medicine and neurology to Vygotsky's (1934, 1962, 1965, 1978) ideas. He is now known as the father of a new field in psychology: neuropsychology. Besides his work on the functional organization of the brain described earlier, Luria was interested in understanding the role of language in the functioning of human mind and is well known for his studies of psycho-linguistic problems, especially aphasia. According to Jakobson (1971), Luria's work seems to be the most 55 instructive among the contributions of neurologists, psychiatrists and psychologists. Luria defined two types of aphasia corresponding to the two classes of speech disorders proposed by Jakobson. Paradigmatic aphasia results from lesions in the temporo-parieto-occipital part of the brain, and is identified by the subject's incapacity to name a thing or to understand the meaning of words. The expressions "the brother's father" and "the father's brother" are perceived as identical. According to Luria (1973a): This group of patients, while having severely disturbed phonemes or articulemes and showing a marked deterioration of semantic organization of words and their relations, preserve their fluent syntagmatically organized speech. Syntactic structures can be secondarily disturbed but the basic relations of noun and predicate remain undisturbed, (p. 61) In cases of syntagmatic aphasia, the predicative function of language is disorganized. These cases of aphasia are produced by lesions in fronto-temporal part of the brain. The lack of verbs eliminates transitions between different classes of words. Consequently, a patient with syntagmatic aphasia would have difficulty finding the difference between "the dogs is black" and "the dog is black." Luria (1973a) indicates: "As a rule patients of this group preserve their ability to name objects and to understand meanings of words. Severe disturbances are seen in these patients in their fluent syntactic organization of speech" (p. 63). In summary, the literature on word associations, linguistic development and the neurolinguistic literature on aphasia all suggest that the paradigmatic syntagmatic model or its variations is the dominant one to explain pathologies as well as individual differences in the domain of language. Basic linguistic processes such as Paradigmatic and Syntagmatic 56 Processes, in parallel and independently recognized in linguistic, developmental, and neuropsychological research have to be taken into account in explaining second language production. For these reasons, the present study verified the relation between Paradigmatic and Syntagmatic Processes and types of errors in second language production. From the literature presented on the Paradigmatic and Syntagmatic Linguistic Processes, I hypothesized that individual differences in Vocabulary Diversity and Vocabulary Errors other than Transfer Errors would be correlated with Paradigmatic Linguistic Processes and that individual differences in Grammar Diversity and Grammatical Errors other than Transfer Errors would be correlated with individual differences in Syntagmatic Linguistic Processes. Since the Syntagmatic-Paradigmatic Shift occurs between the age 5 and 8, subjects from Grade 3 should offer the largest variance on these variables. Parallel between Linguistic Processes and Cognitive Processes Disorganization of the Paradigmatic and Syntagmatic systems observed in the cases of aphasia have been associated with impaired underlying cognitive processes. Luria (1973a) explains Paradigmatic and Syntagmatic aphasia by the disorganization of the Simultaneous and Successive Cognitive Processes respectively. These cognitive processes have been observed not only in special populations of aphasic patients, but also in normal populations (for a review of these studies see Das, 57 Kirby, & Jarman, 1975, 1979). These studies of Simultaneous and Successive Cognitive Processes evolved into the PASS model of Naglieri and Das (1988), where individual differences can be identified from a cognitive rather than a purely psychometric point of view. Das, Cummins, Kirby, and Jarman (1979) studied Simultaneous and Successive Processes as they relate to linguistic functions and mental abilities. By using surface structure, underlying structure, and lexical ambiguous sentences they showed that Successive Processing is related to syntax whereas Simultaneous Processing is related to vocabulary. Furthermore, Jarman (1980a) described the link between Simultaneous and Successive Cognitive Processes and Paradigmatic and Syntagmatic Linguistic Processes with a group of 104 children of mean chronological age 7.6 years, using factorial analysis. He concluded his study by saying, "The results of this study are consistent with the premise that the cognitive processes underlying paradigmatic and syntagmatic clustering are simultaneous and successive syntheses respectively," (p. 159). However, in a second study the results were slightly different. The subjects were 52 high IQ children (mean verbal IQ=118.68 and mean performance IQ= 120.01) aged 9. In this study, as in the first study, the Paradigmatic Clustering task loaded on the first factor, Simultaneous Processing. The Syntagmatic Clustering task loaded on the second factor, Successive Processing. However, the loading on this second factor was negative, indicating "a notable inversion of correspondence between competency in successive synthesis and syntagmatic clustering," (p. 161). This 58 ambiguous result is difficult to explain. It could be caused by the small number of subjects or by the high IQ of the subjects. The author explained that it is possible the tasks were not appropriate for the subjects' advanced language development. Syntagmatic Linguistic Processes might be better shown by associations across sentences than by mere contiguous associations in high IQ subjects. My study clarified this finding through replication of Jarman's second study with a larger number of subjects of high IQ. French immersion students have been shown to have higher IQ than children in other programmes (Nielsen, 1983; Trites, 1986). C. Summary In summary, previous studies about the quality of French in French immersion school have provided us with various taxonomies of errors and a global evaluation of the quality of French in these programs compared to other programs and to native French speakers. Almost universally, studies show that French immersion pupils make errors especially of the type identified as negative transfer errors. Models have been proposed to explain errors and second language production. Their contribution has been to show that both developmental errors and transfer errors are part of second language production through overgeneralization. Yet, they do not include subject variables and make only general predictions to the entire population of second language speakers without distinctions among individuals. In the last decade, research in metacognition has had some influence on recent conceptual models of second language. Variables such as planning, linguistic awareness, control of language 59 processing, as well as labels such as " metasystem1 have appeared in different models. However, previous models and empirical studies have neglected to show that there are individual differences in the language production of students in French immersion classrooms and have not analyzed these individual differences in terms of individual differences in cognitive and linguistic processes. The most recent models of second language production and acquisition are not grounded in research in the disciplines of psychology or neuropsychology, nor do they flow from generic models of language learning. As a consequence they do not offer operational definitions of the elements which they comprise. Besides, none of these recent models include all of the metacognitive variables; some variables are included in some models whereas other variables are included in other models. Yet, planning and attention likely play an important role in* negative transfer errors. -> Moreover, Paradigmatic and Syntagmatic Linguistic Processes have been identified as important factors in language in various traditions namely linguistic, developmental, and neuropsychological traditions. In addition,Simultaneous and Successive Cognitive Processes have been shown to correlate with Paradigmatic and Syntagmatic Linguistic Processes in only one study. This relation should be confirmed. I argue, therefore, that an appropriate model for a study of individual differences in second language production is the PASS model, originating in Luria1 s theory and model of the organization of the brain, operationalized in the Cognitive Assessment System of cognitive processes. This assessment battery is presently being standardized and normed. The model comprises four functional units corresponding to the variables identified as playing an important role in second 60 language production: Planning, Attention, Simultaneous and Successive Processes. The literature suggests there is equivalence between the Paradigmatic and Syntagmatic Linguistic Processes and the Simultaneous and Successive more general Cognitive Processes of Luria's and the PASS model. However, the published results are not conclusive. I replicated and extended the results of Jarman's study (1980b) on this question. Furthermore, I hypothesized that the Paradigmatic and Syntagmatic Linguistic Processes which have been discussed in parallel and independently in various disciplines were correlated to Vocabulary Errors and Diversity and to Grammar Errors (other than transfer errors) and Grammar Diversity, respectively. I also hypothesized that Transfer Errors were correlated to Control of Linguistic Processing defined as Planning and Attention. Such relationships have been investigated in the present study. 61 Chapter HI. Statement of the Problem A . Overall Statement of the Problem The foregoing review has suggested a basis for a study of the quality of French spoken by French immersion pupils. A study of the relationship between linguistic and cognitive processes, and second language production had the potential to explain individual differences among French immersion pupils' errors in second language. The investigation of the relationship between the quality of French in French immersion schools, and linguistic and cognitive processes stemmed from several considerations. First, various studies supported the uneasiness of parents and teachers over the quality of French written and spoken by French immersion pupils. Although French immersion pupils speak fluently and are able to communicate, they make errors by comparison to native speakers in speaking French (Hammerly, 1989; Lapkin, 1984; Lyster, 1987; Nemni, 1985; Pellerin & Hammerly; 1986; Spilka, 1976). Diverse teaching methods have been proposed based on inferences about the cause of the problem. However, none of these new ways of teaching took account of the possibility that learner variables could play an important role in second language production. 62 Second, two main models have been proposed to explain errors made in one's second language, namely, the Creative Construction hypothesis and the Contrastive Analysis hypothesis. Both models explained errors by a lack of discrimination in learning a second language. Proponents of the Creative Construction hypothesis claimed that pupils over-generalized the second language rules they had learned, whereas proponents of the Contrastive Analysis hypothesis claimed pupils inappropriately generalized the first language rules they knew to their second language. None of the studies supporting one or the other model was convincing. Rather, Selinker, the originator of the concept of "interlanguage," considered both hypotheses to be right as some errors have an intralingual source whereas other errors have a crosslingual source. However, these models predicted the kinds of errors that could be made by second language speakers and writers as a group. They did not allow prediction of different patterns of errors for different pupils because they did not include subject variables. Third, development of cognitive science and metacognition research has introduced a number of individual variables into recent models of second language acquisition and production such as linguistic awareness, control, monitoring, planning, and consciousness (Bialystok, 1988; Ellis, 1985; Krashen, 1978; Lamendella, 1977). However, none of the models included many metacognitive variables. In addition, operational definitions of the variables were vague. Consequently, the models could not be put to the test. Another weakness of the models was that they were not rooted in generic models of language learning, nor in cognitive psychology, nor in neuropsychology. 63 B. Rationale A review of the literature showed that some variables could be significantly related to the quality of second language production. The investigation of the relationships between cognitive processes as well as linguistic processes and the types of errors made in second language production had the potential to shed light on the quality of French spoken and written by French immersion pupils. The problem caused in second language research by the lack of theoretical foundation in cognitive psychology, language learning, or neuropsychology called for a cognitive model related to language. Luria's theory of brain organization and functions (Luria, 1966, 1973b, 1980) closely related to language in a Vygostkian perspective was useful to the problem. This cognitive and linguistic model based on neuropsychological research contains most of the elements necessary to analyze language production in relation to cognition. Furthermore, the operationalization of Luria's theory in the PASS model (Das, Kirby, & Jarman, 1975, 1979) and in the Das-Naglieri Cognitive Assessment System (1989) was instrumental in the testing of hypotheses about language production and cognition. Thus, it is within this model that I investigated the relationships between three main sets of variables. First, various studies (Das, Cummins, Kirby, & Jarman, 1979; Luria, 1966, 1970, 1973a, 1973b, 1975a, 1975b, 1976,1980; ) indicated that Successive Processes are related to syntax whereas Simultaneous Processes are related to vocabulary. Thus, I hypothesized 64 that Successive Cognitive Processes would be mainly related to Grammar Errors and Diversity and that Simultaneous Cognitive Processes would be mainly related to Vocabulary Errors. Second, recent research in language (Bialystok, 1988; Ellis, 1985; Krashen, 1978; Lamendalla, 1977) included metacognitive variables such as linguistic awareness, control, monitoring, planning, and consciousness. In parallel, Luria (1966) explained that these functions were specific to the third functional unit of the brain which was itself closely related to the first functional unit of the brain: Arousal/Attention. Further, he indicated that this unit was responsible for impulse control and spontaneous speech and that patients with frontal lobe lesions had difficulties inhibiting the production of irrelevant words that came to their mind. This description was reminiscent of the phenomenon of transfer errors in French immersion schools where pupils seem to have difficulties inhibiting English words when they have not fully integrated French vocabulary, or grammatical structures when these are different in the two languages. I therefore hypothesized that Planning and Attention would be related to Vocabulary and Grammar Transfer Errors. Third, Paradigmatic and Syntagmatic Linguistic Processes have repeatedly been shown to play an important role in language, and have respectively been associated with different elements of language. The literature (Hughlings-Jackson, 1879, 1915; Jakobson, 1971) indicates that choice of vocabulary, and classification of concepts are the results of Paradigmatic Processes, whereas the concept of combination of words into sentences which 65 are in turn combined into utterances and paragraphs is related to Syntagmatic Processes. Syntax which is based on contiguity and context is the result of Syntagmatic Processes. Thus, a model of second language acquisition and production had to include these processes. One hypothesis of this study is that Syntagmatic Linguistic Processes would be related to Grammar Errors and Diversity and that Paradigmatic Linguistic Processes would be related to Vocabulary Errors and Diversity. Furthermore, the parallel between Paradigmatic and Syntagmatic Linguistic Processes on the one hand and Simultaneous and Successive Cognitive Processes on the other hand had been widely discussed in the literature (Jakobson, 1971; Luria, 1970, 1973a, 1975a, 1975b, 1976) and tested by Jarman (1980a). However, Jarman's results were not conclusive. For this reason, I replicated Jarman's study with my Grade 3 French immersion students sample to verify the parallel between Cognitive and Linguistic Processes. To recapitulate, the review of the literature indicated that individual differences in Linguistic and Cognitive Processes could explain types of errors in second language production. Simultaneous and Successive Cognitive Processes as well as Paradigmatic and Syntagmatic Linguistic Processes, and Control, defined as Planning and Attention, have been shown to play a large part in language and had to be included in a model of second language production. Finally, the model used had to be rooted in cognitive psychology and neuropsychology and had to have precise operational definitions of variables to allow their measurement. It is with these considerations that I undertook this study. 66 In particular, the relationships between, (1) Successive as well as Syntagmatic Processes, and Grammar Errors and Diversity, (2) Simultaneous as well as Paradigmatic Processes, and Vocabulary Errors and Diversity, and between (3) the level of Control, defined as Planning and Attention, and Negative Transfer Errors deserved exploration. In addition, the parallel between Successive and Syntagmatic Processes as well as Simultaneous and Paradigmatic Processes required confirmation. Figures 1,2, and 3 show the hypothesized relationships between the latent variables under investigation. There are three categories of latent variables and these are represented in oval boxes. First, general cognitive variables as described in Luria1 s functional organization of the brain and as operationalized in the PASS model. These are Successive Processes (Figure 1), Simultaneous Processes (Figure 2), and Planning and Attention (Figure 3). Second, linguistic processes described by Jakobson and by Luria. These are Syntagmatic Processes (Figure 1) and Paradigmatic Processes (Figure 2). Third, French language variables corresponding to classification of errors and quality of French. These are Grammar Errors and Diversity other than Transfer (Figure 1), Vocabulary Errors and Diversity other than Transfer (Figure 2), and Grammar and Vocabulary Transfer Errors (Figure 3). The measures corresponding to each of these latent variables are indicated in square boxes. The arrows pointing back to the square boxes denote error corresponding to the reliability of measurement (see Figures 1,2,3). 67 Figure 1. Path diagram for Syntagmatic Processes, Successive Processes, and Grammar Diversity and Errors other than Transfer. 68 Figure 2 Path diagram for Paradigmatic Processes, Simultaneous Processes, and Vocabulary Diversity and Errors other than Transfer. 6 9 visual planned matching search codes numbers } f f oral trans, grammar errors oral trans, vocabulary errors written trans, grammar errors written trans, vocabulary errors Figure 3 . Path diagram for Planning, Attention, and Vocabulary and Grammar Transfer Errors. 70 C. Questions From the literature review and on the rationale just presented, I proposed to answer the following questions: a. What is the magnitude of the relationship between Paradigmatic and Syntagmatic Linguistic Processes and Simultaneous and Successive Cognitive Processes respectively? b. What is the relative importance of Syntagmatic Linguistic Processes and Successive Cognitive Processes in predicting Grammar Diversity and Grammar Errors other than Transfer Errors? c. What is the relative importance of Paradigmatic Linguistic Processes and of Simultaneous Cognitive Processes in predicting Vocabulary Diversity and Vocabulary Errors other than Transfer? d. To what extent is Control, defined as Planning and Attention, a significant predictor of Transfer Errors? 71 Chapter IV. Method A . Subjects Subjects were 152 anglophone students from grade 3 French-immersion classrooms with 63 boys and 89 girls. The total number was chosen following the rule of thumb of at least 10 subjects per variable for factor analyses (Kerlinger, 1986) and each of the analyses include not more than 10 variables. Learning disabled students and students receiving learning assistance were excluded from the study. Specific details regarding the criteria for selecting the subjects, and the size of the sample prior to the elimination of some subjects, are included in the procedure section of this chapter and in the preliminary analyses in Chapter V. B . Tasks Three sets of tasks were used: first, the tasks developed to measure the cognitive processes as defined in the PASS model; second, two tasks both measuring Paradigmatic and Syntagmatic Linguistic Processes; third, written and oral French language tasks measuring grammar, vocabulary, and transfer language errors. a. Planning, Attention, Simultaneous/Successive Processes (PASS) Tasks The PASS model was developed and validated using Luria's tasks, and other tasks 72 judged similar to Luria's tasks or analyzed as requiring the same Cognitive Processes. After early establishment of the validity of the model in the 1970's and 1980's as explained in chapter II, a new operationalization took form for the development of a standardized test based on the PASS model: the Das-Naglieri Cognitive Assessment System. This system was used in the present study. The Das-Naglieri Cognitive Assessment System is designated to assess Planning, Attention, Simultaneous and Successive Processing of children ranging in age from 5 to 17. There are 12 subtests, three for each process measured. Some of the tasks are directly taken or adapted from Luria, others come from the validation of the PASS model, and some others are new. The Planning subtests are: Visual Search, Planned Codes, and Planned Connections. The Simultaneous Processes subtests are: Simultaneous Verbal, Figure Memory, and Matrices. The Successive Processes subtests are: Word Series, Sentence Repetition, and Sentence Questions. The Attention subtests are: Expressive Attention, Receptive Attention, and Number Finding. A description of the subtests follows (see Appendix A for samples). Visual Search (see Appendix A 1) The child's task is to point to the object, number, or letter in the visual field (area around the centre box) that matches the target (object in the centre box). Each page which comprises one item consists of two searches, one on the top and one on the bottom half of each page. There are 16 pages with two searches on each. For each item (page), the time 73 taken from the point of exposure until both searches are completed, is recorded. The score for this subtest is the total time taken for the 16 items. In this task, none of the subjects made errors. It is important here to note some differences between the various scores obtained on the different tasks in this study. Some of the variables are power tests and others are speed tests. Visual Search is an example of a speed test. This is critical to recognize in an individual differences study because obviously high ability on a speed test results in a low score whereas high ability on a power test corresponds to a high score or to a large number completed on a set time. In this test, the measures that are timed such as Visual Search, Planned Connections, and Expressive Attention are speed tests whereas all of the other measures are power tests. Speed corresponds to time whereas power corresponds to the number right or correct. In order for the measures to have the same directional scale and to be comparable to others, I did a transformation of the speed tests. The transformation was done by changing the sign of the result on the test and adding to it a round number that was larger than the highest limit of the range for that variable. For Visual Search, the range was 58-201; therefore, I chose 250 to add to the inverse of the result. The transformation formula, then, was: T= - X + 250 [X being the result of the test for any subject] After the transformations, a high result is reflected in a high score an a low result is characterized by a low score. 74 Planned Codes (see Appendix A 2) The child's task is to fill out a series of boxes marked with the letters A, B, C, or D using the proper sequence of X's and O's (e.g., A = OX, B = XX, etc.). Planned codes consists of two items (pages), each with different codes and different arrangements of response locations. The child is instructed to fill out the boxes as quickly as possible. They are given one minute to fill out correctly as many boxes as possible. The score is the number of correct items. There are 56 items possible for each page; therefore, the maximum score possible is 112. Planned Codes is mainly a power test because it is not the time that is recorded but the number of right answers. The more right answers there are, the higher the score will be. Planned Connections (see Appendix A 3) The child's task is to connect a series of boxes containing numbers or numbers and letters in correct sequence. The child is instructed to connect the boxes as quickly as possible. The first items involve a sequence of numbers only, whereas the last items involve a sequence of numbers and letters (e.g., 1-A-2-B-3-C...). There are four items with different time limits. Appendix A 3 presents the 10th item of the entire test which is actually the 4th and last item of Planned Connections. The score corresponds to the total time taken to complete the four items. There were no errors in this task because if the child makes an error the examiner of the test is supposed to tell the subject that he/she has made a mistake, and to begin again from the previous correct location. The maximum time is 10 minutes and 75 30 seconds. Planned Connections is a speed test since the score is the time taken to do all the items and accuracy does not vary. The scores had to be transformed to be comparable to the other scores as explained for Visual Search. For Planned Connections, the range was 88 - 425, and the transformation formula was: T = - X + 500 [X being the result of the test for any subject] Simultaneous Verbal (see Appendix A 4) The child's task is to choose from among six options the one picture that correctly answers the question read by the examiner. The questions are printed at the bottom of each stimulus page and involve various logico-grammatical relationships. The score is the total time divided by the number of correct items. The maximum number of items is 26, and the test is stopped after four consecutive failures. Figure Memory (see Appendix A 5) The child's task is to locate and outline a geometric figure that is embedded within a more complex design. Each stimulus figure is exposed for 5 seconds, and the child is then instructed to outline the stimulus figure within the more complex design presented on the response page. The score for this subtest is the total of items correct. The maximum score is 24, since there are 24 items. However, the test is stopped after four consecutive failures. Matrices (see Appendix A 6) The child's task is to select one of the six options that best completes the matrix. This 76 subtest uses the standard progressive matrix format and varies from completion of the pattern to a 3 X 3 matrix. Some of the items are coloured and others are in black and white. The score is the number correct. There is a maximum of 31 items, and a maximum score of 31, however the child is stopped after four consecutive failures. Word Series (see Appendix A 7) The child's task is to repeat a series of words in the same order in which the examiner says them. A score of one is given for every perfect recall. There is a possibility of 26 points since there are 26 series of words. The test is discontinued after four consecutive failures. Sentence Repetition (see Appendix A 8) In Sentence Repetition the child's task is to repeat a sentence spoken by the examiner. The sentences contain colour names in place of content words. A score of one is given for every perfect repetition. There is a possibility of 17 points since there are 17 sentences. The test is discontinued after four consecutive failures. Sentence Questions (see Appendix A 9) This subtest is related to the preceding one. In this task the child is to answer a question related to the sentences to be repeated in the previous subtest. There are 17 questions and therefore, a possibility of 17 points. Here again, the test is discontinued after four consecutive failures. 77 Expressive Attention (see Appendix A 10) This test has three parts. First the child is asked to read colour words printed in black as fast as possible. The item is timed. The words are "red", "blue", "green", and "yellow". There are 40 words on the page. In the second part, the child is asked to tell in order the colours printed on the page. This item is also timed. There are 40 colours printed on the page. Finally, in the last part the child is presented with colour words printed in a different ink colour. For example, the word red is printed in blue ink. The children have to tell the colour the words are printed in, as fast as they can. Only the time taken for reading the 40 words is recorded. Only the time part was used to measure expressive attention. Using the time for scoring makes this subtest a speed test and it had to be transformed. For Expressive Attention, the range was 35-121, and the transformation formula was: T = - X + 150 [X being the result of the test for any subject] Receptive Attention (see Appendix A l l ) In Receptive Attention the child's task is to find and underline pairs of letters that are the same on the basis of a physical or a category/name match. In the first item, the child has to underline pairs of letters that are physically the same (capital or minuscule, for example BT or tr or rr). In the second item, they have to underline pairs of letters that have the same name (for example, TT or Tt or tt). Only the second item was used in this study. In this item there are 10 pairs of letters per row and 20 rows (200 pairs of letters in total). Among them, 50 pairs of letters have the same name. The maximum time allowed is 2 minutes and none of the children had underlined all 50 pairs after two minutes had elapsed. 78 The score was the number correct divided by the number attempted. The maximum possible is 50/200 = .25. I assumed that the last pair of letters underlined was equivalent to the last one attempted. However, it is important to note that this way of scoring is only an approximate measure of the real number of items attempted. In reality, the number attempted could be higher, but there is no means to find out whether children had attempted more items nor what they had decided about those items because a written response is not required to all pairs. Here the number of items attempted is only the closest estimation I can make and the same rule was applied to each case. Number Finding (see Appendix A 12) In this subtest, the child's task is to find and underline specific numbers within rows of numbers that contain both targets and distracters. The child has to find and underline the numbers in each row that are the same as the model provided at the top of the page. There are twelve numbers on each row and fifteen rows per page (180 numbers in total). Among them, 45 numbers are identical to the model. The maximum time allowed is 1 minute and 30 seconds and none of the children were finished after that time had elapsed. The score was calculated as the number correct divided by the number attempted. The maximum possible would be 45/180 = .25. As in the preceding subtest I assumed that the last number underlined was the last one attempted. However, this way of scoring is only an approximate measure of the real number of items attempted. The same note as in the previous subtest applies here. 79 b. Paradigmatic and Syntagmatic Linguistic Processes Tasks Among the experimental paradigms used in verbal learning research we can find a variety of paradigms including free word association tasks; multiple choice association tasks, which are a restricted variation of the free word association task; clustering tasks; free recall tasks; and paired-associate recall. Although they all measure some aspect of memory, free recall and paired-associate tasks attempt to measure short-term memory whereas Clustering tasks measure the organization of memory, and free association tasks assess the organization of long term memory. The major source of information about short-term memory is a set of recall tasks such as probe recall and free recall (Glanzer, 1982). In Clustering tasks, on the other hand, there is a basis for making inferences about organization of memory through "the correspondence between output order and the specific categorical relationships defined, a priori" (Murphy & Puff, 1982, p. 119). Murphy and Puff (1982) also indicate that clustering tasks increase our understanding of the nature of the transformational, or organizational, processes. One of the goals of this study was to uncover Paradigmatic and Syntagmatic Processes. I assumed that these processes have an active part in the organization of all forms of memory; therefore, at first I had chosen to use a free word association task and a Clustering task rather than experimental paradigms designed to measure short-term memory. Free word association tasks have been used historically to uncover Paradigmatic and Syntagmatic Processes. They show the preference of subjects for one kind of processes or the other. As mentioned earlier, both processes are available and continually operative in normal verbal 80 behaviour, however people tend to use one or the other depending on their developmental age, their personality, cultural pattern, or verbal style (Jakobson, 1971). In contrast, a Clustering task is more constraining than Free Recall. If the a priori categorization is done along the two kinds of associations, paradigmatic and syntagmatic, subjects should show which of the Paradigmatic or the Syntagmatic Processes are the most efficient for them to organize words for recall. With respect to second language acquisition in a French immersion environment, the way pupils organize the information is crucial. In fact, they receive models of second language by listening and reading and no rules are explicitly given to them. They have to organize the linguistic information they receive under the form of multiple examples in order to speak correctly, in the same way they did when they learned their first language. However, when they learned their first language, everyone around them spoke that language; therefore, they were exposed to a more diversified language, and there were more repetitions of the same grammatical structures or vocabulary. In French immersion, the diversity and quantity of French is restricted compared to a first language, consequently it becomes even more important to have a very efficient way of organizing information in order to learn and remember. Free Word Association reflects the organization of what has been learned and/or the plain preference of output. A Clustering task shows the actual process of organization of information when faced with a learning task. This task includes higher demands than a Free Word Association task because there is an interaction with the material presented. Free Word Association tasks and Clustering task present some advantages for this study. On the one hand, the results on Free Word Association tasks would allow inferences about pupils' memory of linguistic aspects of 81 learned second language-especially linguistic memory organization and preference of output (paradigmatic or syntagmatic). On the other hand the Clustering task allows inferences about pupil's organizational processing at work when receiving linguistic information in the form of examples. Thus, the Free Word Association task should show which of the two processes subjects prefer in the context of production of language, whereas the Clustering task should show which of the two processes they find the most efficient when the task is designed to give both processes an equal value, as when they receive information. For example, they could memorize the list of examples (Syntagmatic Processing) or they could memorize the more abstract rule they inferred from the examples (Paradigmatic Processing). However, a more generalized and abstract way of classifying information (Paradigmatic Processing), is usually more efficient when it comes to memory for production and understanding of language. With respect to the Clustering tasks, they were taken from Jarman (1980a) and consisted of two lists, one for each of Paradigmatic and Syntagmatic Processes. These tasks were developed from measures used by Denney and Ziobrowski (1972). They involved 12 pairs of words. Five pairs of words were paradigmatic associations and five pairs were syntagmatic associations while the first and last words were filler words to reduce serial effects. The 12 pairs of words in each list were listed randomly, with the only restriction that two words in a pair should not be adjacent to one another. The word lists were recorded on cassette tape for administration, with the words spaced at 1 per second. The tests were 82 given in counterbalanced order, using three practice trials of words, followed by the test trials. Subjects were given 2 minutes after each list to recall the words orally. The responses were scored for the number of words recalled in pairs, to form clustering scores that ranged from 0 to 5 for each process. However, trial of the tasks showed that Grade 3 children could easily do the Free Association task, whereas there was a floor effect on the Clustering task. Children could remember four to five items but only one or zero cluster. Cole, Frankel, and Sharp (1971) found similar results. In their experiment the third graders remembered about nine items after five trials, but had a mean clustering score of only 1.4. They note that clustering does not improve with practice and suggest that to study age-related differences, a set of experiments should be designed to teach the skills to younger children. Kirby, Jarman, and Das (1975) found the same results in an unpublished study with grade 4 boys. The mean number of clusters recalled was less than 1 for each of the Paradigmatic and Syntagmatic Processes. In the light of these trials and of the results of these studies, I decided not to use a Clustering task, and instead, to use a Closed Word Association task (see Appendix B 1). Before I explain why this task was my second choice, I will describe it. Closed Word Association Task (see Appendix B 1) The task was taken from Tomlinson (1988), who modified slightly a task used by Ervin 83 (1961). The task has two lists of 10 items each. Each item consists of three words, a stimulus word, a paradigmatic response word and a syntagmatic response word. The first response word is randomly syntagmatic or paradigmatic. The child cannot see the list. The examiner presents each item of the test in the following way. For example, the examiner says, "If I say hot, which word would you choose, warm or fire?" In this study, the lists were administered in a counterbalanced way. Scores could range from 0 to 10. I chose this task mainly for two reasons. First, the memory demand of Closed Association task is considerably lower than the Clustering task. It is merely short-term memory. Hence, the pupils should be able to have an answer for each item of the test, and the bottom effect which happened with the Clustering task should not occur. Second, this task is the closest to the Clustering task in terms of organizational processing of received information. Pupils are given a paradigmatic or a syntagmatic choice of fitting new information (the choice of word) to old one (the stimulus word). They have to do the same thing when receiving examples of second language learning. Free Word Association Task (see Appendix B 2) The Free Word Association task consisted of two lists of 10 stimulus words (see Appendix B 2). One list was scored for Paradigmatic Processes and one list was scored for Paradigmatic Processes. Since it has been demonstrated that nouns produce a predominance of paradigmatic responses (Entwistle, 1966), each list was formed with three stimulus words of four different parts of speech: nouns, verbs, adverbs, and adjectives. Furthermore, each 84 category had three words, except the adverb which had only one word. Among the three words, one was taken from a high frequency list, one was taken from a medium frequency list, and one was taken from a low frequency list (Entwistle, 1966). The lists were given orally in counterbalanced order. A strict scoring was used. Words of the same grammatical category were scored as paradigmatic, and words of different grammatical categories were scored as syntagmatic. In this task, two separate lists were scored to prevent the statistical problems occurring with ipsative measures (Clemans, 1956). The kind of analyses used in this study required independent measures. Scores could range from 0 to 10 for each of Paradigmatic and Syntagmatic Processes. c. Language Tasks French is taught in a natural way in French immersion classrooms. The basic principle in these classrooms is that children learn a second language as they learned their first language, through a communicative approach. Grammar rules are not taught and pupils usually do not learn vocabulary lists as it is often the case in French as a Second Language classes. The method is successful but in the assessment of the level of proficiency, it is difficult to find communicative tests. In such tests there is not only one good answer, but many, which creates reliability and validity problems. For this study, I only needed samples of written and oral language at the Grade 3 French immersion level, and the test had to be similar to a normal activity in French Immersion. Norms were not necessary since I was 85 using my own scoring to correct the tests. There is only one communicative test in Canada for Grade 3 French Immersion. I used that test for the written part. For the oral part, the task was to tell a story that the children made with cards. This activity was presented like a game. The children could create any kind of story and they were involved in a communicative activity by telling their story to the examiner. A more detailed description of the tasks follows. Language Proficiency was measured by the written part of a test developed by the Ontario Institute for Studies in Education (OISE), the French Language Speaking and Writing Evaluation Units (1983). This test is based on the theoretical framework of communicative competence, which corresponds to the pedagogy used in French immersion schools. It has been developed for the Official Minority Language Office of Saskatchewan Education and intended for French immersion and French minority language students. The core component of the kit for Grade 3 is a slide/tape presentation which tells the story of a young boy from Sairit-Boniface, Manitoba who becomes involved in the search for a missing guinea pig when he visits his cousin's school. The written measures include a short question-and-answer exercise, a cloze test with 20 missing words, and a somewhat longer composition exercise (see Appendix C). These tasks took approximately three hours to administer. The test is not standardized but the guide offers a section entitled, "Interpreting the results." Although ranges of expected scores are not given, this section provides information on scores obtained by Saskatchewan classes when the materials were field tested in April/May 1983, for the different facets of communication that are measured in each task. 86 Table 1 recapitulates these norms for all the written tasks. However, I did not use this French written test to compare the language level of my sample with the Saskatchewan immersion population, but rather to analyze individual differences on Transfer, Vocabulary, and Grammar Errors. From the written tasks of the Language proficiency test seven scores of written speech were derived. The questions were not scored; they were only used as a warm up activity. Cloze Test The cloze test was scored to give a general language ability measure. It was used in an informal way to verify whether the results were generally comparable to the Saskatchewan grade 3 population. Composition From the composition six scores were derived, namely Written Grammar Diversity, Written Vocabulary Diversity, Written Vocabulary Errors other than Transfer Errors, Written Vocabulary Transfer Errors, Written Grammar Errors other than Transfer Errors, and Written Grammar Transfer Errors. A type/token ratio was used to measure these errors. Since a ratio tends to decrease as the sample of speech grows longer, the type/token ratio must be applied to samples of uniform size. Thus a sample of 100 words from the written composition was used to derive the six scores measuring language errors. 100 words were chosen because in Grade 3 French Immersion, the children could not write more than this. 87 Up until Grade 3, French immersion pupils work more on the oral than on the written part of language. It already took them some effort to arrive at this length of text. Grammar Diversity was measured by a type/token ratio of the number of sentences made of different grammatical structures to the total number of sentences in the written sample of 100 words. Vocabulary Diversity was scored in this study with a method originally developed by the Ontario Institute for Studies in Education (1983) and also used by Day and Shapson (1987). This type/token ration method for scoring Vocabulary Diversity was the same as the one explained in the test manual. The type was the number of different verbs, in a conjugated or infinitive form, used by a pupil in the 100 word sample. The token was the total number of verbs used in the same 100 word language sample. Grammar Errors (see Appendix H for details) comprised structure errors as well as agreement errors. First, the total number of Grammar Errors was noted. Then, within the Grammar Errors, those which were a direct translation from English were recorded as Grammar Transfer Errors. The list recorded by Lyster (1987) served as a reference. Transfer Grammar Errors (see Appendix H for details) were measured by calculating the ratio of the frequency of those errors to 100 (the number of words in the sample). 88 Grammar Errors other than Transfer Errors (see Appendix H for details) were calculated by deriving a ratio. The number of Transfer Grammar Errors was subtracted from the total number of Grammar Errors and the frequency of Grammar Errors was divided by 100. Vocabulary Errors and Transfer Vocabulary Errors were measured the same way as Grammar Errors and Transfer Grammar Errors (see Appendix H for details). Oral Language Sample The subtest "Picture Arrangement" of the Wechsler Intelligence Scale for Children III (WISC-III) (1991) was used and adapted to obtain an oral sample of French language (see example in Appendix D). This subtest consists of sets of three to seven pictures drawn on separate square cardboard tiles. The cards were presented in a specific standardized order different from the right order. When this subtest is used in the WISC III the child is timed while he/she puts the cards in the right order. If the cards are put in the wrong order the pupil loses the point. For the purpose of this study the pupils were also asked to put the pictures in the right order. However, they were not penalized for a wrong order and they were not timed. Rather they were measured on the extent to which the story made sense to them so they would be able to talk about it. The cards were only a stimulus to help them speak. The test was given individually and the telling of the stories created was recorded on audio-tape. As the pupils were presented the cards they were told that the cards were in a wrong order and that they had to put them in the right order. They were also told that when 89 they were ready, they would have to tell aloud and in French the story that they would have created. Ordering the cards allowed them to think about the ideas and use inner speech while making sense of the story. This task provided the oral language sample. The audio-tapes were transcribed. The first story was used as a warm-up activity to make sure that the pupils understood the task. Then a sample made of the next 100 words was used to measure oral language errors. Based on that sample six scores were derived, namely Oral Grammar Diversity, Oral Vocabulary Diversity, Oral Grammar Errors, Oral Grammar Transfer Errors, Oral Vocabulary Errors, and Oral Vocabulary Transfer Errors. The derivation of the scores was done in the same manner as for the aforementioned scores derived on the basis of the written samples. 90 Table 1 Norms for the Written Tasks of the French Language Speaking and Writing Evaluation Units (1983) T A S K S F A C E T S M E A S U R E D B E L O W A V E R A G E A V E R A G E A B O V E A V E R A G E Incomplete sentences >:1.99 1.98 to 1.37 ^1.36 QUESTIONS Prepositions errors >0.92 0.91 to 0.66 ^0.65 C L O Z E T E S T (maxi=20) Correct responses ^7.94 7.95 to 10.37 >: 10.38 Vocabulary variety ^4.50 4.51 to 5.92 >5.93 COMPOSITION -Vocabulary sophistication ^0.05 0.06 to 0.27 >Q.2S Information communicated ^3.67 3.68 to 3.91 >:3.92 Tenses usage in % of errors >: 15.00% 14.99% to 7.00% ^6.99% C. Procedure The tests were administered to grade 3 pupils from 3 School Districts of the Lower Mainland of Vancouver, involving 9 schools, and 16 classrooms. A letter was sent to the parents explaining the purpose of the study and the time involved in testing their child. The letter requested authorization and some information about the school history and their child (see Appendix E). As soon as authorization forms signed by the parents were returned, testing sessions were organized for each classroom. The letter was sent to the parents of 254 pupils. 88.6% gave permission for their children to be part of the study. Of the 225 left, 15.5 % were eliminated because they could 91 not write 100 words, or had serious learning difficulties. The children with learning difficulties were those attending the Learning Assistance Centre (LAC) and those identified by the teacher as having learning difficulties although they did not attend LAC. In several cases the parents did not want them to be sent outside of the class to receive special education. Furthermore, 10.6% were eliminated because they had a third language other than French or English. Finally, 5.8% were absent on the day of the test. Some of those who had learning difficulties also had a third language. Finally, 59.8% were left in the study from the potential original group. The final sample was 152 pupils, 89 girls, 63 boys, (for more details see the section in preliminary analyses in Chapter V). The language tests were administered first on two consecutive days. On the first day the examiner introduced the story "O— est Nicou" to the whole class, then slides accompanied by a tape were presented to the group. This took approximately 30 minutes after which the pupils were administered the written tasks as a group, the question and answer activity, the cloze test, and the composition. They had morning recess during their test and usually the whole class was finished at 11:00 am. On the second day, those participating in the study were taken one by one out of the classroom for the oral test, the Picture Arrangement subtest of the WISC III. This test took approximately 20 minutes. On a third day the linguistic tasks were administered to each child before the Das-Naglieri test. The linguistic tasks took approximately 15 minutes. The Das-Naglieri 92 Cognitive Assessment System test which took approximately 1 hour and 15 minutes was given last. In total, each child spent two hours being tested in class and one hour and 35 minutes being tested individually outside of the classroom. Four people helped me in the administration of the tests: one registered psychologist and three graduate students. One graduate student was trained to give only the French oral test, another was trained to give the French oral test and the Das-Naglieri test, and the last one was trained to give only the Das-Naglieri test. The persons who administered the French oral test were trained together and those administering the Das-Naglieri test and the linguistic processes tasks were also trained together. This facilitated clarifications and discussions and ensured the homogeneity of testing administration. The trainee had to give 10 tests, the last 3 being supervised, before they could test for the study. During the study they were supervised randomly, to verify that they did not change the way of testing as they were becoming more skilled. 93 Chapter V. Results A . Descriptive Statistics a. Demographic Statistics Subjects were 152 anglophones with a mean age of 104.95 months (standard deviation of 3.9). They included 89 girls (mean age=105.23; standard deviation=4.70) and 63 boys (mean age =104.95; standard deviation=3.92). The sample was taken from three School Districts of the Lower Mainland of Vancouver, British Columbia, with 16 classrooms in 9 schools. The original sample of 254 pupils was reduced to 152 after various eliminations such as if parental permission was missing, if the subject spoke three or more languages, if the subject had learning disabilities, or if the subject was absent on the group testing day. The number and percentage for each category of elimination is shown in Table 2. Table 2 Successive Eliminations from the Original Potential Sample to form the Final Sample Successive Eliminations Number % of Original Sample Original sample 254 100.00 % No parental permission 29 11.40 % 3 languages or more 24 9.40 % Learning disabilities 36 14.20 % Absent on testing day 13 5.10 % Final Sample 152 59.80 % b. Descriptive Statistics of the Cognitive and Linguistic Variables The means and standard deviations of the measures from the PASS model (Das-Naglieri test) and of the linguistic processes variables were calculated, and are given in Table 3. Boys and girls scored differently, with girls scoring higher on almost all the variables, the exceptions being Matrices, Word Series, and Free Association Syntagmatic where boys scored slightly higher than girls. Table 3 Means and Standard Deviations fSD^ of the PASS Model and the Linguistic Processes Variables by Sex and for the Total Sample Variables Boys Girls Total Sample Mean SD Mean SD Mean SD PASS Variables Visual Search 139.51 27 146.11 29.6 143.38 28.8 Planned Codes 50.84 13 55.72 11.9 53.7 12.5 Planned Connections 286.10 59 293.61 56.3 290.49 57.4 Simultaneous Verbal 10.96 2.9 11.86 3.7 11.49 3.4 Figure Memory 11.62 3.3 12.25 2.9 11.99 3.1 Matrices 14.35 4.6 14.17 4.4 14.24 4.5 Word Series 10.59 2.8 10.09 2.7 10.3 2.7 Sentence Repetition 6.09 2.1 6.53 2.1 6.34 2.1 Sentence Questions 6.21 2.9 6.75 2.5 6.53 2.7 Expressive Attention 79.17 17 85.52 15 82.89 16 Receptive Attention 0.79 0.2 0.8 0.1 0.8 0.1 Number Finding 0.84 0.2 0.88 0.1 0.86 0.1 Language Processes Variables Free Assoc. Paradigmatic 5.43 2.3 5.56 2.3 5.51 2.3 Free Assoc. Syntagmatic 4.21 2.3 4.02 2.2 4.1 2.2 Closed Assoc. Paradigmatic 3.57 1.3 3.72 1.5 3.66 1.4 Closed Assoc. Syntagmatic 7.29 1.3 7.53 1.3 7.43 1.3 95 c. Descriptive Statistics of Language Measures The Cloze Test, a general language measure, produced a mean of 10.67 with a standard deviation of 3.9. Compared to the results of the Saskatchewan sample my sample scored in the "Above Average class," defined as 10.38 and higher (see Table 1). Language errors were scored as explained in Chapter IV. The percentage of errors relative to each total of errors is shown in Appendix F and a visual representation of the repartition of the errors is presented in Figures 4 and 5. The results indicate there are more grammar than vocabulary errors in total. In the written part, spelling increased the number of grammar errors. However, in the oral part the number of grammar errors is still higher than the number of oral vocabulary errors. There are more oral transfer errors than written transfer errors; the total of transfer errors is approximately 23% of all errors in oral language production and only 11 % of all errors in written language production. Generally there are more vocabulary transfer errors than grammar transfer errors. Descriptive statistics were calculated for each of the language variables. Table 4 gives the mean and standard deviation by sex and for the total sample for each language variable. In general, girls made fewer errors than boys, and scored higher on measures of Written Vocabulary and Grammar Diversity. 9 6 Figure 4. Distribution of all errors between vocabulary and grammar 97 Table 4 Means and Standard Deviations (SD) of the Language Variables by Sex and for the Total Sample Variables Boy s Girls Total Sample Mean SD Mean SD Mean SD Cloze test 10.67 4 12.67 3 11.82 3.6 Oral Grammar Diversity 0.79 0 0.78 0.1 0.78 0.1 Oral Vocabulary Diversity 0.78 0 0.75 0.1 0.76 0.1 Oral Grammar Errors 8.57 4 7.70 3.7 8.06 3.8 Oral Transfer Grammar Errors 0.86 1 0.62 0.8 0.72 0.9 Oral Vocabulary Errors 4.06 2 3.84 1.9 3.93 1.9 Oral Transfer Vocabulary Errors 3.14 3 2.69 2.1 2.88 2.3 Written Grammar Diversity 0.61 0 0.65 0.2 0.64 0.2 Written Vocabulary Diversity 0.51 0 0.54 0.1 0.53 0.1 Written Vocabulary Errors 7.19 5 5.36 3.8 6.12 4.3 Written Transfer Vocabulary Errors 2.47 3 1.79 1.8 2.07 2.3 Written Grammar Errors 15.80 9 14.30 5.8 14.90 7.1 Written Transfer Grammar Errors 0.86 2 0.54 0.9 0.67 1.2 d. Sex Differences The means by sexes appeared to differ for all three groups of variables, namely the PASS model variables, the linguistic processes variables, and the language variables, with most girls' means being higher than boys' means (see Table 3 and Table 4). It became necessary therefore, to test statistically the differences between the girls' data and the boys' data to pool the two groups and to treat them as a homogeneous group in correlational analyses. 98 To test the differences between sexes, the multivariate test, Hotelling T square was calculated. The result showed that at least one of the means differed between the boys' group and the girls' group (Hotelling T^, square = 70,24, p < 0.008). However this result does not tell us which one(s) of the variables is (are) different. A t test was performed to verify separately the difference of each of the variables between the boys' group and the girls' group. I chose a quasi t test using Welch's formula for degrees of freedom. This test is the "Welch's approximate t" attributed to Smith (1936). This test is more conservative but more appropriate when ns are unequal and when it is not known if there is homogeneity of variance between the two groups. According to Zar (1984) it is an easy, yet reliable, procedure to solve the problem of unequal ns (pp. 130-131). The results are presented in Table 5 for the cognitive and linguistic variables and in Table 6 for the language variables (see Tables5 and 6). Among the cognitive and linguistic variables (Table 5) only Planned Codes and Expressive Attention were significantly higher in the girls' group. Expressive Attention and Planned Codes have been shown to measure Planning. Among the language variables, girls scored significantly higher than boys on three tasks: Cloze Test, Oral Vocabulary Diversity, and Written Vocabulary Errors (Table 6). According to the results of these t tests, then, girls scored higher than boys in Planning, in general language as measured by the Cloze Test, and in Oral Vocabulary Diversity. 99 Table 5 Results of t Tests for Boys and Girls for the Cognitive and Linguistic Variables Variables t value Degrees of Probability Freedom Cognitive Variables Visual Search -1.41 139.58 0.159 Planned Codes -2.38 127.91 .019* Planned Connections -0.79 129.69 0.43 Simultaneous Verbal -1.68 147.49 0.096 Figure Memory -0.76 128.27 0.448 Matrices 0.54 131.18 0.592 Word Series -1.35 131.37 0.18 Sentence Repetition -1.32 133.28 0.189 Sentence Questions -0.97 120 0.335 Expressive Attention -2.4 124.72 .018* Receptive Attention -0.47 118.96 0.638 Number Finding -0.87 132.13 0.386 Linguistic Variables Free Association Paradigmatic -0.34 132.11 0.736 Free Association Syntagmatic 0.49 131.2 0.622 Closed Association Paradigmatic -0.64 141.93 0.521 Closed Association Syntagmatic -1.13 130.17 0.261 * p < .05 100 Table 6 Results of t Tests for Boys and Girls for the language Variables Language Variables t value Degrees of Freedom Probability Cloze test -3.39 112.00 .001* Oral Grammar Diversity 0.36 135.48 .718 Oral Vocabulary Diversity 1.96 139.42 .052* Oral Grammar Errors 1.39 128.86 .165 Oral Transfer Grammar Errors 1.56 109.44 .122 Oral Vocabulary Errors .69 131.51 .488 Oral Transfer Vocab. Errors 1.15 115.24 .251 Written Grammar Diversity -0.92 125.93 .362 Written Vocabulary Diversity -1.59 134.38 .115 Written Vocabulary Errors 2.51 112.39 .013* Written Transf. Vocab. Errors 1.74 100.30 .085 Written Grammar Errors 1.25 101.05 .214 Written Transf. Gram. Errors 1.48 90.80 .142 * p < .05 A second but independent issue with respect to sex differences in the prospective pooling of the two sex groups, is that of the differences between the two groups in correlation patterns. Thus, it became important to compare the equality of the correlation matrices between the two groups. This can generally be done with the test Multicorr (version 2.4) (Steiger, 1987). Multicorr allows a significance test of a correlation pattern hypothesis. Steiger (1987, p. 2) defines a correlation pattern hypothesis as "a null hypothesis that constrains elements of a correlation matrix to be equal to each other, or to specified numerical values or both." Unfortunately, the latest version of Multicorr (2.4) is designed 101 only for a small set of variables, which precluded its use here. A version for a larger set of variables is in preparation. The next best solution was to compare the structure and the equality of the covariance matrices using LISREL VI. LISREL has programmes to compare two matrices on their equality, on the equality of their number of factors, and on the equality of their factor structure. I did these analyses for each of the basic models (Successive, Simultaneous, and Attention/Planning) using raw data. However, in most of those analyses the x 2 could not be calculated, or if it could, it had a very large value. The question of the difference between the boys' and the girls's correlation or covariance matrices, therefore, remains unanswered. In order to proceed to the main analyses, using a pooled sample, I then removed the mean sex differences in the data. To accomplish this, I did a linear transformation of the data to a standard T-scale, so that the boys' and the girls' data would have the same mean (50) and the same standard deviation (10). This is feasible because the addition, subtraction, multiplication, or division of all variables by a constant does not change the correlation. To transform the data into a T-scale, all variables were first changed to z scores using the formula: Z = (X- X / SD) In this transformation the girls' mean was used for the girls' sample and the boys's mean was used for the boys' sample. Then the variables were changed into T scores using the 102 formula: T = 10 Z + 50. Consequently, the confounding effect of mean sex differences was removed to leave only the effect of individual differences in the new correlations, with these correlations as an aggregate of the correlations for each of the sex groups. e. Correlation Matrix As a point of comparison on the effect of the T-score transformations, a correlation matrix was calculated first from the raw data (see Appendix G) and a new correlation matrix was calculated with the transformed variables. Table 7 presents the correlation matrix for all the transformed variables that were to be analyzed in the main analyses. In the new correlation matrix the correlations appear very similar to those in the correlation matrix calculated from the raw data. Table 7 is made up of six groups of correlations, namely the correlations between the cognitive variables, the correlations between the linguistic processes variables, and the correlations between the language variables, accompanied by the correlations between these three groups of variables. The correlations between the PASS model variables ranged in absolute value from a low of .001 (for the correlation between Matrices and Visual Search, and for the correlation between Receptive Attention and Visual Search), to a high of .492 (for the correlation between Sentence Questions and Word Series). The lowest correlations were between one variable measuring Planning and one variable measuring Simultaneous Processes on the one 103 hand, and one variable measuring Attention and the other measuring Planning on the other hand. In contrast, the highest correlation was between two variables measuring the same processes: Successive Processes. The correlations between the linguistic processes ranged in absolute value from a low of .005 to a high of .630. The correlation between Free Association Syntagmatic and Free Association Paradigmatic was -.630, which was expected since this is a correlation between the results of the same task (Free Association) measuring complementary linguistic processes (paradigmatic and syntagmatic). If the Free Association task and the Closed Association task are equivalent we could expect an equivalent magnitude of correlation between the results of these two tasks when they are used to measure the two complementary linguistic processes (paradigmatic and syntagmatic), as we have when only Free Association is used for the same purpose. However, the correlation between Free Association Paradigmatic and Closed Association Syntagmatic was only .005. The correlations between the language tasks ranged in absolute value from a low of .002 (for the correlation between Written Grammar Errors and Written Transfer Grammar Errors), to a high of .273 for the correlation between Oral Grammar Errors and Oral Transfer Vocabulary Errors. In general, there was no clear pattern in the language measures correlations. The correlations were low among oral grammar or vocabulary measures and among written measures of grammar or vocabulary. Some non-zero correlations were found between oral vocabulary and written vocabulary measures. The highest correlations were 104 between written grammar and written vocabulary measures, and between oral grammar and oral or written vocabulary measures. There were also non-zero correlations between oral and written vocabulary measures. Lower correlations were found between oral and written grammar and between written grammar and oral vocabulary measures. The correlations between the PASS variables and the linguistic variables ranged in absolute value from a low of .002 (for the correlation between Sentence Repetition and Closed Association Paradigmatic), to a high of .231 (between Sentence Repetition and Free Association Syntagmatic). Most of the highest correlations were between the PASS variables and the Free Association tasks, whereas there were mostly zero-order correlations between the PASS variables and the Closed Association tasks. In the same way, there were few substantial correlations between the Planning and Attention variables and the linguistic variables. Most of the highest correlations were between the simultaneous and the successive tasks of the PASS model and the Free Association tasks. The correlations between the PASS variables and the language variables ranged in absolute value from a low of .001 (between Sentence Repetition and Oral Vocabulary Errors), to a high of .265 between Sentence Repetition and Written Vocabulary Errors. Most of the highest correlations were between the variables measuring the Simultaneous and Successive Processes of the PASS model, and the language tasks. Planning had more high correlations with oral language tasks than with written language tasks, and conversely, Attention had more high correlations with the written language tasks than with the oral 105 language tasks. The correlations between the linguistic tasks and the language tasks ranged in absolute value from a low of .000 (between Free Association Paradigmatic and Written Transfer Grammar errors), to a high of .236 between Closed Association Paradigmatic and Oral Grammar errors. Most of the highest correlations were between the Oral Language tasks and the linguistic tasks. There were only two correlations greater than .100 between the linguistic tasks and the written language tasks. 106 Table 7 Correlation Matrix of all the Variables calculated from Transformed Data (Sex Differences removed) 107 i l l 383 II 1[ M S I 3 l s § £ 1 1 S £ 3 * i s £ 2 ass? E p g is i H H H 5 U i § H H H H ^ i s s & H £ g I i § § is § s i l i a s i § § § H h I H 5 §s H n i i k k s s I' i § § i S i § *§ § § i 5 1 § i ' 3 i l i i g s i 5 i g c ^ i § § s § § § § i *i it k k i I j§ "i s k i § s i c 5 § 3 i i i § 'i i § i § l I § . t o fa fa J-j g fa g C tj 8! ~J w *3 8 i 5 s s i i 1 i 5 § k 1 1 1 I r— § 8 & § i 5 B * § a is o 8 J § « ft O >— K> _ l/l 00 «J w 00 -3 8 * 8 8 K 2 3 o O if D -1 -« 5' 3 n> -1 ro 3 o re 3 °< 8. LA FT s 8. o 3 3 3 .8. B . Preliminary Analysis Before studying the relationships discussed in previous chapters, I had to verify whether my data fitted the PASS model and the Das-Naglieri test. I did this in two steps, first with a principal components factor analysis (PCA), and then with a confirmatory factor analysis. a. Principal Components Factor Analysis of the PASS Model The first stage in my preliminary analysis was to verify the construct validity of my data against the PASS model and the Das-Naglieri test using a PCA. A PCA is usually necessary when one does not know how many factors will result from a large number of variables. The goal in these circumstances is often merely to summarize many variables in a few factors. In my case, because data collection was based on the PASS theoretical model, and on the Das-Naglieri test, this first PCA was not completely exploratory. The number of factors in the PASS model is known. However, although a model may be correct for the entire population, results from a sample of the population may be different. In this study, in particular, the sample did not represent the general population. French immersion pupils differ from the general population on many variables, such as intelligence, socio-economic status, and parents' achievement orientation. Therefore, the goal of this first analysis was to verify that the factor structure of my sample corresponded to the factor structure of the model. 108 A PCA of the first 12 variables in the correlation matrix in Table 7 was done, followed by an orthogonal rotation of the factors. Eigenvalues greater than one were retained. Table 8 contains the results of this analysis and shows the four factors of the PASS model. Figure 8 shows the results of this Principal Component Analysis. The numbers on the arrows are the loadings or weights, rounded to two decimals. For ease of interpretation, Figure 6 gives only the highest loadings of each task on the factors. The main loadings of Factor I were Word Series, Sentence Repetition, and Sentence Questions. Factor II comprised Planned Codes, Planned Connections, Expressive Attention, and to a lesser extent, Visual Search. Factor III comprised Receptive Attention and Number Finding. Finally, Simultaneous Verbal, Matrices, and to a lesser extent, Figure Memory, were the main variables loading on Factor IV. Consistent with the various factorial validity studies of these tasks or similar tasks (Das, Kirby, and Jarman 1979; Naglieri, Das, Stevens, and Ledbetter, 1991) these factors were respectively, Successive Processes, Planning Processes, Attention Processes, and Simultaneous Processes from the PASS model. Figure Memory, in addition to loading on Factor IV, also loaded to a lesser degree on Factors I and II. In the same way, Visual Search loaded to an equal degree on Factor II and IV. 109 Table 8 PCA with Orthogonal Rotation of the Tasks from the Das-Naglieri Test Factor I Factor II Factor III Factor IV Tasks Successive Planning Attention Simultaneous Processes Processes Visual Search .051 .493 .091 -0.48 Planned Codes -.110 .710 -.170 .077 Planned Connections .149 .803 -0.008 -0.033 Simultaneous Verbal -.004 -0.126 -.114 .700 Figure Memory .349 .431 .229 .454 Matrices .210 .217 .292 .585 Word Series 0.806 .158 .008 -0.004 Sentence Repetition .862 -.009 -.021 .096 Sentence Questions .790 .076 .122 .092 Expressive Attention .163 .704 .045 -0.038 Receptive Attention .110 .010 .765 .064 Number Finding -.037 -.077 .748 -.040 Variance 2.259 2.174 1.351 1.300 % of Total Variance 24.22 15.22 10.38 9.21 This analysis unexpectedly produced different results for Expressive Attention than has been found typically in analyses of the variables in the Das-Naglieri test. In my results, Expressive Attention loaded on Planning rather than on Attention as found in previous studies using the test. However, the scoring of the tasks in this study was done according to instructions in the manual (Das & Naglieri, 1989) and according to a personal communication with Naglieri1. Although the scoring of Expressive Attention was straightforward (total time of the third trial) and done according to Naglieri1 s instructions, 1 Personal e-mail message from Jack Naglieri (#4714497 posted at 6:45 on May 19, 1993). The subject of the message was scoring. It gave information on how to score Visual Search, Planned Connections, Simultaneous Verbal, Sentence Repetition, Receptive Attention, and Expressive Attention. The message indicated that there could be various ways to score Receptive Attention. 110 the task did not load on Attention, but instead loaded on Planning. For Receptive Attention and other tasks for which the scoring had not been finalized, I tried a total of 84 combinations of various kinds of scoring systems to produce adequate measures of the functions/factors that they represented. Various factor analyses were performed to test the factor structures which were extracted from these scoring systems. However, in each of them, Expressive Attention invariably loaded on Planning. Therefore, for this study, Expressive Attention was considered and used as a Planning task. The authors of the test have shown that it is possible that some tasks measure different processes according to the cognitive stage of development. For example, Naglieri (1989) reported that Naglieri and Das (1988) found that Successive Ordering, a test that was thought to measure Successive Processes, loaded on different factors depending on the Grade of the subjects. In the factorial analyses, the test loaded on Successive Processes at Grade 2, on Planning and Successive Processes at Grade 6 111 Attention \ . 80 P l a n n e d f ^ ^ J O C o n n e c t i o n s V i s u a l S e a r c h P l a n n e d C o d e s E x p r e s s i v e Attent ion W o r d Se r i e s r " -86 S e n t e n c e -^ Repet i t ion S e n t e n c e Ques t i ons S imu l t aneous V e r b a l r .45 Figure ^ - ^ 5 8 M e m o r y Mat r i ces R e c e p t i v e Attent ion V - .75 N u m b e r F ind ing Figure 6. Principal Component Analysis of the variables of the PASS model (with an orthogonal rotation). 112 b. Distinction between PCA and Confirmatory Factor Analysis After verifying that the number of factors was the same as that of the PASS model, at least using the eigenvalue greater than one criterion, and whether the tasks loaded on the same or similar factors, a confirmatory maximum likelihood factor analysis (MLCFA) was performed to verify the level of fit of the model and to obtain the level of significance of the loadings of the tasks. It was important here to establish the level of significance of the loadings of the tasks for the variables which loaded on two different factors in the Principal Component Analysis (PCA). There are some important differences between PCA, and confirmatory analyses, especially confirmatory maximum likelihood factor analysis (MLCFA). In a PCA the structure of the model is not controlled, except possibly for the selection of the number of factors, or restrictions on the correlations between factors. A MLCFA gives more precise results by separating error from the total variance of each variable. The part of the variance of the variables due to systematic and random error is distinguished from the part of the variance accounted for by the factors. In contrast, in PCA, systematic and random error is included in the model without being analyzed separately, and therefore, the equation does not contain an error term. Thus, in a PCA "The variables can be directly calculated from the factors by applying the weights" (Gorsuch, 1983, p. 20). A MLCFA estimates the proportion of the variance of the variables that can be accounted for by the common factors as well as the uniqueness. Thus, this distinction between the commonality and the uniqueness of the variables makes a MLCFA more informative than a PCA about the 113 composition of the variance. Further, the parameter estimates of the MLCFA are usually smaller than the weights of the PCA, as is seen here in a comparison of Figures 6 and 7. c. Confirmatory Factor Analysis of the Variables of the PASS Model As I indicated previously, LISREL VI was used to perform the MLCFA. The program calculated four indices of fit of the model: the chi-square ( x 2 ) , the goodness of fit index (GFI), the adjusted goodness of fit index (AGFI), and the root mean square residual (RMSR). There are no direct standards to judge whether these indices are significant, except if they are compared to the fit of another model, which is not the case here. However, the X2 does enable us to make a judgement of the fit of the model as described by a design matrix. In Table 9, a design matrix is given which is based on the configuration of variables and factors found with the PCA. Table 9 Design Matrix for the Confirmatory Analysis of the PASS Model Variables Visual Search 0 100 Planned Codes 0 100 Planned Connections 0 100 Expressive Attention 0 100 Word Series 1000 Sentence Repetition 1000 Sentence Questions 1000 Simultaneous Verbal 000 1 Figure Memory 000 1 Matrices 000 1 Receptive Attention 00 1 0 Number Finding 00 1 0 114 This design matrix was used here in a test of the fit of the data to this matrix, using the four indices of fit described above. Table 10 gives the values of these indices of fit. Table 10 Indices of Fit of the Confirmatory Analysis of the PASS Model Variables. x 2 df GFI AGF RMSR 50.01 48 .953 .878 .054 GFI = goodness-of-fit index; AGFI = adjusted goodness-of-fit; RMSR = root mean square residual. The x 2 was not significant (Prob. level = .394). In this type of analysis, the x 2 should be as small as possible. If the % is significant, it means the population covariance had not been extracted completely; the residual matrix still contains significant covariance. Also the GFI and the AGF should be as close to 1 as possible and the RSMR should be as close to 0 as possible. In Table 10, it can be seen that the data fitted the model since the x 2 was not significant. Also the goodness of fit and the adjusted goodness of fit were close to 1 and the root mean square residual was very small. Figure 7 shows all standardized estimates of the relation between the observed variables (the various tasks) and the latent variables (unobserved variables such as Planning Processes, Successive Processes, Simultaneous Processes and Attention Processes). Estimates indicated on the arrows between the latent variables and the observed variables are regression coefficients for the estimated effect of the latent on the observed. They are standardized, meaning that the coefficient shows the expected magnitude of change in the observed variable for one unit change in the latent variable (Bollen, 1989). The t values for each standardized estimate appear in parentheses. Parameters whose f-values are larger than 115 1.96 in magnitude are normally judged to be different from zero (J"reskog & S"rbom, 1988). If this is the case, it means that the loading of the variable on the factor is statistically significant. The covariances between the latent variables are also indicated. Since these are standardized covariances, they are in effect correlations between the latent variables. Arrows outside the rectangle boxes indicate systematic and random error variance. As shown in Figure 7, most of the standardized estimates were significant (t > 1.96). In that figure, all coefficients have been rounded off to two decimal places. The loadings of Visual Search and Figure Memory on Planning Processes and Simultaneous Processes respectively were significant. Some standardized estimates such as the one between Simultaneous Verbal and Simultaneous Processes, and also between Attention and both of its observed variables, were not significant. This means that Simultaneous Verbal was not the best observed variable to measure Simultaneous Processes, and the two variables chosen to measure Attention left a substantial part of Attention unexplained. As for the correlations between the latent variables, the largest one was between Simultaneous and Successive Processes, followed by the correlation between Planning and Simultaneous Processes (.43). Lower correlations related Planning to Successive Processes and Attention to Simultaneous Processes. Finally, a very low correlation related Planning and Attention, contrary to previous findings in the literature. 116 Visual Search .90 Planned Codes .76 Planned Connections .32 Expressive Attention .58 Sentence .30 Repetition Sentence Questions .52 Word .53 Series Simultaneous .79 Verbal Figure Memory .78 Matrices .88 Receptive .05 Attention \ .27 (1.48) Number .93 Finding ^ — Figure 7. Maximum Likelihood Confirmatory Factor Analysis of the PASS model. 117 The analysis also estimated the reliabilities of the PASS model measures: Squared Multiple Correlations (SMC) which are presented in Table 11. These measures of reliability are low and uneven. For example, Simultaneous Verbal and Number Finding are extremely low (.01 and .08 respectively) whereas Receptive Attention has a reliability of .85. A comparison of the reliabilities of the tasks for the general population with those calculated from the sample of the present study would shed some light on the meaning of the latter. Unfortunately, the reliabilities of the tasks of the Das-Naglieri CAS test are not available yet2. However, these reliabilities are specific to the sample and to the data of the present study. Table 11 Reliabilities of the CAS measures (SMC) Visual Search Planned Codes Planned Connections Expressive Attention Word Series Sentence Repetition 0.13 0.27 0.64 0.38 0.51 0.63 Sentence Questions Simultaneous Verbal Figure Memory Matrices Receptive Attention Number Finding 0.51 0.01 0.65 0.27 0.85 0.08 C. Main Analyses a. Structural Equation Modelling with LISREL Structural equation modelling was used in this study with the computer program 2 Personal e-mail message from Jack Naglieri (Sept. 25th, 1995). 118 LISREL 7 (J"reskog & S"rbom, 1989) because this statistical method has special advantages for investigating the questions of this study as presented in Chapter 3. Structural equation modelling combines the analysis possibilities of multivariate analysis, factor analysis, and multiple regression (Abbott & Berninger, 1993). This kind of analysis presents four main advantages as follows: Multiple Measures of the same Concept It allows tests of models in which several measures rather than only one are associated with the same construct. For example, in this study four variables measured vocabulary level, in particular, Oral Vocabulary Errors, Written Vocabulary Errors, Oral Vocabulary Diversity, and Written Vocabulary Diversity. Using multiple variables to represent a construct makes the model more realistic and more accurate, and captures more variance. Thus, it allows a comparison of the relevance of each observed variable to an explanation of the concept. Observed Variables Reliability The reliability of the variables measuring a construct can be compared by contrasting the size of standardized estimates of the relations between one latent variable and each of the observed variables measuring it. For example, three variables are used in the PASS model to measure Simultaneous Processes, namely Simultaneous Verbal, Matrices, and Figure Memory. As shown in Figure 7, Figure Memory were related to Simultaneous Processes by a large and significant standardized estimate, whereas Simultaneous Verbal was related 119 by a low-and non-significant standardized estimate to Simultaneous Processes. Therefore, in this model, Figure Memory was a relatively more reliable measure of Simultaneous Processes than Simultaneous Verbal. Structural Comparisons of Latent Variables In addition, not only means can be compared, but also the structural compositions of the latent variables. For example, the structural composition of the Successive Cognitive Processes and the Syntagmatic Linguistic Processes can be compared, because they are thought to be subsidiary constructs from two different fields of research. As will be seen in Figure 8, Successive Cognitive Processes and Syntagmatic Linguistic Processes were moderately and negatively correlated with each other. Sentence Repetition was the observed variable which best measured Successive Processes, while Free Association Syntagmatic was the observed variable that best measured Syntagmatic Linguistic Processes. Comparison of Models Finally, structural equation modelling makes it possible to test the fit of a theoretical model for a population from sample data. All the relationships in a model can be tested concurrently instead of conducting a series of successive comparisons. For example, my first model on the relations between Successive Processes, Syntagmatic Processes, and Grammar errors could be tested in one operation instead of testing consecutively the relationships between each construct of the model (e.g., testing the correlation between Successive and Syntagmatic Processes first, then the relationships of each of these constructs 120 to grammar level). The synchronized view of the variables of interest found in a structural model is closer to the reality of the world than that given by other statistical techniques. There are two steps in a LISREL analysis. In the first step a measurement model is specified and in the second step a causal model is specified. For example, in this study the hypothesized models (Figures 1, 2 an 3) described in chapter III, were translated into mathematical models with two parts, the measurement models and the causal models. A measurement model is a system of simultaneous equations describing the relation of each latent or unobserved variable to its corresponding observed variables (the results on the various tasks). A measurement model leads to a causal model testing how well the hypothesized model fits the data. b. Cognitive Processes and Language Variables The main analyses comprised, first, the three measurement models corresponding to the hypothesized models in chapter III, and second the corresponding causal models; in particular, the Successive Processes model, the Simultaneous Processes model, and finally the Planning/Attention Processes model. However, none of the causal models converged. To solve this problem a complete verification of the data and a re-analysis was done. This re-analysis is presented in the section designated as "D. Reanalyses" below. 121 c. Parallelism between Cognitive and Linguistic Processes One goal of my study was to analyze the strength of the relationship between cognitive and linguistic processes, or more precisely between Successive and Syntagmatic Processes, and between Simultaneous and Paradigmatic Processes following Jarman's (1980a) study. In this regard it is important to note that although this particular question in the present study was the same as in Jarman's (1980a) study, the design was slightly different. The tasks to measure the cognitive and linguistic processes proficiency were different in two ways from those used by Jarman. First, Jarman used some of the tasks included in the Das-Naglieri CAS test and other tasks used by Luria to measure the cognitive processes proficiency, whereas the tasks used to measure the cognitive processes in the present study were exclusively the ones used in the Das-Naglieri CAS test. Second, Jarman used only one task to measure the linguistic processes proficiency, namely a clustering task. However, a pilot study of the clustering task with my sample resulted in a floor effect. Instead of the clustering task, I used the Closed Word Association task which is the closest to the clustering task in terms of organization of information received. I also used the Free Word Association task in order to have a measure of linguistic memory organization and preference of output (paradigmatic or syntagmatic). Another difference between my study and Jarman's study is the use of measurement structural models as a method of analysis of the relationship between Cognitive and Linguistic Processes. However, I also used factor analysis to examine the data as in Jarman's (1980a) study . The LISREL analysis (see Figure 8) testing the relationships between Cognitive and 122 Linguistic Processes indicated that the data fitted the models. Successive Processes were negatively correlated with Syntagmatic Processes and Simultaneous Processes were also negatively correlated with Paradigmatic Processes. The indices of fit are contained in Table 12. Table 12 Indices of Fit of the Models testing the Relationships between Cognitive and Linguistic Processes. Model x 2 df GFI AGFI RMSR Correlation Successive/Syntagmatic 1.75* 4 0.99 0.99 0.019 Correlation Simultaneous/Paradigmatic 0.65 4 0.99 0.99 0.015 *p. > .05 123 Figure 8. Models testing the correlation between, Simultaneous and Paradigmatic Processes and between Successive and Syntagmatic Processes. 124 For a straightforward comparison of my results with Jarman's (1980a), I factor analyzed the data. I first did a Principal Components analysis forcing the analysis into 2 components, with Varimax rotation. I also set an eigenvalue greater than 1 criterion of the number of factors, as a default option. The results of this analysis is presented in Table 13. Two factors emerged from this analysis, using the number of factors selection criterion, which overrode the fact that 5 eigenvalues were greater than 1. Contrary to Jarman (1980a) who found a Simultaneous/Paradigmatic factor and a Successive/Syntagmatic factor, the first factor I found was composed of all the cognitive variables, and the second one was composed of the linguistic variables. Because there is a large amount of error variance in the data, I decided to do the analysis again using a Common Factor method with Varimax rotation, asking for two factors, but also setting an eigenvalue greater than 1 criterion for retaining factors. The results of this analysis are reported in Table 14. Only one factor emerged from the analysis. In both of my analyses and as in Jarman's (1980a) study, the loadings for Syntagmatic Processes are negative whereas the loadings for Paradigmatic Processes are positive. In addition, the loadings for the two kinds of Linguistic Processes tasks are quite different; the loading for Free Word Association is larger than the loading for Closed Word Association. Thus, I was not able to replicate the results found by Jarman (1980a) with a principal components analysis or with a common factor analysis of the cognitive and linguistic tasks 125 together. I return to this finding in much more detail in subsequent sections. Table 13 Principal Components Factor Analysis of the Simultaneous and Successive Processes Variables and of the Paradigmatic and Syntagmatic Processes Variables Tasks Factor 1 Factor 2 Successive Processes tasks Sentence Repetition .797 .064 Sentence Questions .767 .101 Word Series .750 .019 Simultaneous Processes tasks Figure Memory .641 - .042 Matrices .502 .128 Simultaneous Verbal .163 .022 Syntagmatic Processes tasks Free Word Association (Syntagmatic) - .168 - .886 Closed Word Association (Syntag.) - .030 - .086 Paradigmatic Processes tasks Free Word Association (Paradig.) .086 .885 Closed Word Association (Paradig.) -.038 .159 126 Table 14 Common Factor Analysis of the Simultaneous and Successive Processes Variables and of the Paradigmatic and Syntagmatic Processes Variables Tasks Factor loadings Successive Processes tasks Sentence Repetition 1? .741 Sentence Questions .709 Word Series .646 Simultaneous Processes tasks Figure Memory .470 Matrices .391 Simultaneous Verbal .113 Syntagmatic Processes tasks Free Word Association (Syntagmatic) -.346 Closed Word Association (Syntag.) -.041 Paradigmatic Processes tasks Free Word Association (Paradig.) .277 Closed Word Association (Paradig.) .016 D . Reanalyses Many of the LISREL models did not converge in the main analysis and no solution could be found for them. Failure to converge is probably due to (a) the low level of correlations in my data. Other factors could also explain nonconvergence. These include degree of normality, sample size, outliers, models, and other technical considerations. Although I had checked these factors before starting my analyses, I checked again each of these possibilities. 127 a. Level of the correlations The correlation matrix (Table 7) shows that most of the correlations were low, except for the variables measuring each function in the PASS model. A study of individual differences requires moderately high correlations (McDonald, 1985) since the correlations are the basis of every analysis. This is especially the case with a multivariate method such as LISREL analysis. One reason that could explain these low correlations is the homogeneity of the sample. First, French immersion pupils represent an elite cohort. They have chosen to be in such a programme (or their parents have chosen for them). From a socioeconomic point of view, Burns (1982) found that parents of French immersion pupils have a higher income, a more extensive formal education, and that professionals are heavily represented among French immersion parents. From a cognitive point of view, Lambert (1977) found that in the majority of studies on cognitive ability and bilingualism, bilinguals are adding a socially relevant language to their first language at no cost to their first language competence, scored higher than unilinguals on cognitive ability tests. Furthermore, the successive eliminations from my original sample of French immersion grade 3 pupils made the group even more homogeneous. First, I eliminated those who had no parental permission. Then, I eliminated those who had three languages or more as well as those who had any learning difficulties or could not write a one hundred word composition. These last two eliminations especially, reduced the variance. In general, the students who had three languages or more spoke and wrote better than the others. In contrast, the students who had learning difficulties or could not write a one hundred word composition were making more errors in oral and written French. In summary, the successive eliminations made the group 128 more homogeneous. This could have resulted in a restriction of range and consequently, in low correlations. However, an examination of the standard deviation relative to the mean of each of the different measures (see Table 3 and 4) does not indicate a restriction of range. Another explanation for the low correlations could be the low reliabilities of the PASS measures as seen in Table 11. Such low and uneven reliabilities are not conducive to high correlations. b. Other factors Before examining the usual causes for nonconvergence, I rechecked for errors in the data entry, initial scorings, and transformations (reversal transformations and t-transformations), but none were found. Normality First, a lack of normality or excessive kurtosis in the distributions of the variables can affect the results of the tests. For this reason, I checked all the distributions of the variables for kurtosis and skewness to achieve a better approximation of multivariate normality. Only two variables showed skewness greater than 1. These were Oral Transfer Vocabulary and Oral Transfer Grammar. The most effective transformation for reducing the skewness was a logarithmic square root. I ran the models with these transformed variables and still did not obtain convergence. Sample size Sample size and number of measures for each latent variable can also be a cause for nonconvergence. With regard to minimum sample size, various rules are recommended by 129 different authors; Boomsma (1983) recommends a sample of at least 100, whereas Anderson and Gerbing (1984) recommend a sample of at least 150 to reach convergence. My sample meets either of these criteria and it would appear that the nonconvergence problem does not stem from a too small sample. However, estimating separate models for each sex would not be appropriate, according to either of these criteria. Furthermore, since the differences between sexes have been removed by a t-transformation, an analysis of the models by sex would not give additional information. Number of indicators per latent variable The number of indicators per latent variable is also important in Lisrel analyses. The number of indicators, per latent variable, sufficient for identification of models is two (Bollen, 1989). This number has been met for all the latent variables in my study. Outliers Improper solutions can also be caused by outliers in the various distributions.. These can affect the parameter estimates. Thus, after examining the distributions of the variables and removing the few outliers that existed in the data set, I ran the models again. However, the models still did not converge. Transformations A last examination concerning my data had to do with the linear t-transformation of the variables used to eliminate the mean difference between boys and girls. Using the raw data could sharpen the level of correlations between the variables by removing round-off error due to the transformations, and in turn have an effect on the convergence of the models. I ran the models with the raw data and with the t-transformed data. The results 130 were similar in both cases. Technical causes With LISREL, a solution is found by repeated iterations. The program stops when the difference between two estimated values of the unknown parameter calculated in two consecutive iterations is closer than a set criterion. Two factors which interact with each other are important in reaching convergence: the specified maximum number of iterations, and the starting value for the unknown parameter. Both of these are set according to a default. First the maximum number of iterations can be set higher. Second, if the starting value of the parameter to be estimated is far from the population value, it can be set closer. In general, the choice of the new starting value is the final value of the parameter found in the previous run that did not converge. In my study, I set the number of iterations to a maximum of 5000 and repeated the analyses several times, each time choosing the value of the parameter estimate with the lowest final value in the previous run as the starting value for the next run. These actions were still not successful in achieving convergence. Models Another cause of nonconvergence can be the specification of the model. The cause can range from the faulty inclusion of an indicator to a completely flawed model. It would be meaningless to change the models without a valid reason by trying all the possible combinations of the latent variables and indicators. A re-specification of a model has to make sense. In my study, the models were based on theory and previous research. i However, various model specifications could conform with the theory. I tried to run some of these models. For example, I verified the possibility of using a different model for 131 the latent variable Grammar using a higher order (2nd order) Confirmatory Factor Analysis. I divided Grammar into three latent variables: Grammar Diversity, Grammar Errors (other than transfer), and Transfer Errors. Two measures were associated to each of these variables, a written measure and an oral measure. Anothei;,rnodel divided Grammar into two ', ' '«: latent variables: Oral and Written. Each of these variables were measured with three indicators: Grammar Diversity, Grammar Errors other than transfer, and Transfer Errors. I also tried these models with only the written measures and only the oral measures. And finally , I ran the models without the Grammar Diversity indicator because this indicator had a low discrimination level (SD=.l). However, none of these respecified models converged. Scoring The way variables are scored could obviously affect the results. However, the scoring of the variables included in the CAS test, and the scoring of the results on the linguistic tasks (paradigmatic and syntagmatic), was specified by the authors or in the literature. Thus, although different ways of scoring the language variables could be considered, the classification of errors used in my study was so simple (number of errors or ratio for Grammar Diversity and Vocabulary Diversity) that it would be difficult to find different ways of scoring errors. Another way to score language errors would be to use a more refined classification, but I had no theoretical reason for such a classification and especially for the correspondence between a different classification of language errors and cognitive or linguistic processes. 132 c. Summary From the examination of all the possible factors for nonconvergence, it appears that the main cause of nonconvergence was the low correlations in the data. These were probably due to the low reliabilities (see Table 4) of the cognitive, processes measures. A latent variable model such as LISREL requires moderately high correlations to give a definite solution. Only a method involving observed variables instead of latent variables was appropriate because such model always gives a solution. Multiple regression is one such possible method. d. Regression Analyses As explained in the previous paragraph, the only possible solution to the problem of the data pattern in the present study was to use a different statistical method and multiple regression was the most appropriate method in this respect. The following section gives the results of the various regression analyses which replaced the nonconvergent LISREL analyses. Three groups of regression analyses were done. First, the regression analyses of the successive processes measures on the grammar variables (see Table 15), second, the regression analyses of the simultaneous processes measures on the vocabulary variables (see Table 16), and third, the planning and attention processes measures on the transfer errors (see Table 17). 133 The results show that three analyses are significant. These are, first the regression analysis of simultaneous processes on Written Vocabulary, F(3, 148) = 3.07, p< .05, MSe=95.41, second the regression analysis of Planning on Oral Transfer Vocabulary Errors, F(4, 147) = 2.80, p< .05, M5e=94.82, and,,;third the regression analysis of Planning on All Oral Transfer Errors, F(4, 147) = 2.43,/>< .05, AfS,=95.73. Regression analyses of the Paradigmatic and Syntagmatic Processes on the vocabulary and grammar language variables were also performed; however, none of them was significant. Although such a regression analysis method has greater statistical power to find significant relationships than the LISREL models used previously , the results were not very different. The level of the correlations in this sample was so low that it was almost impossible to find significant relationships even with more statistical power. The results show that Oral Transfer Errors and especially Oral Transfer Vocabulary Errors are significantly associated with Planning and Attention and that Written Vocabulary Errors are significantly associated with Simultaneous Processes. However, none of the regression analyses relating Grammar Errors to Successive Processes were statistically significant. In summary, in the specific context of very low correlations, three regression analyses showed some weak support for the original models derived from theory and previous studies. Yet, the probability of type I error in these analyses compared to the LISREL 134 analyses is quite high and these results should be taken with caution. Table 15 Regression Analyses of Successive Processes Measures on Grammar Variables. Dep. Var. Ind. Var. Beta t p (2 tail) Adjusted squared Mult. R df F p Oral Grammar Word Series Sentence Rep. Sentence Ques. -0.029 -0.107 -0.040 -0.283 -0.996 -0.394 0.777 0.321 0.694 0.004 3 1.2 0.32 Written Grammar Word Series Sentence Rep. Sentence Ques. -0.013 -0.135 -0.081 -0.132 -1.260 -0.801 0.895 0.210 0.425 0.021 3 2.1 0.11 Oral Grammar Diversity Word Series Sentence Rep. Sentence Ques. 0.057 -0.034 -0.005 0.554 -0.316 -0.047 0.580 0.753 0.963 0.000 3 0.1 0.96 Written Grammar Diversity Word Series Sentence Rep. Sentence Ques. 0.018 -0.181 0.072 0.181 -1.676 0.703 0.857 0.096 0.483 0.001 3 1.1 0.37 Oral Transfer Grammar Word Series Sentence Rep. Sentence Ques. -0.069 0.061 0.140 -0.679 0.567 0.638 0.498 0.575 0.170 0.004 3 1.2 0.32 Written Transfer Grammar Word Series Sentence Rep. Sentence Ques. 0.046 0.132 -0.130 0.450 1.224 -1.271 0.653 0.223 0.206 0.000 3 0.9 0.44 * p < .05 135 Table 16 Regression Analyses of Simultaneous Processes Measures on Vocahnlary Variables. Dep. Var. Ind. Var. Beta t P (2tail) Adjusted Squared Mult. R. df F P Oral Simult. verbal Vocabulary Figures Matrices 0.136 -0.019 0.117 1.675 -0.215 1.310 0.096 0.830 0.192 0.014 3 1.7 0.16 Written Simult. Verbal Vocabulary Figures Matrices -0.114 -0.088 -0.149 -1.422 -0.999 -1.690 0.157 0.319 0.093 0.039 3 3.1 0.03* Oral Simult. Verbal Vocabulary Figures Diversity Matrices 0.053 0.078 0.095 0.650 0.872 1.056 0.517 0.385 0.293 0.006 3 1.3 0.28 Written Simult. Verbal Vocabulary Figures Diversity Matrices -0.033 0.039 -0.085 -0.400 0.431 -0.936 0.690 0.667 0.351 0 3 0.4 0.77 Oral Simult. Verbal Transfer Figures Vocabulary Matrices -0.039 -0.072 -0.113 -0.480 -0.806 -1.260 0.632 0.421 0.210 0.008 3 1.4 0.25 Written Simult. Verbal Transfer Figures Vocabulary Matrices -0.023 -0.110 -0.067 -0.283 -1.228 -0.745 0.777 0.221 0.457 0.004 3 1.2 0.31 * p < .05 136 Table 17 Regression Analyses of Planning and Attention Measures on Transfer Errors Dep. Var.Ind. Var. Beta P Adj. (2 tail) squared Mult. R df Oral Transfer Planned Codes Planned Connections Expressive Attention Receptive Attention 0.071 0.049 0.789 0.496 0.431 0.621 0.007 4 1 0.28 Grammar 0.117 -0.019 1.233 -0.218 0.220 0.827 Written Transfer Planned Codes Planned Connections Expressive Attention Receptive Attention 0.199 0.063 2.217 0.649 0.028 0.518 0.022 4 2 0.12 Grammar -0.083 0.030 -0.883 0.358 0.379 0.721 Oral Transfer Planned Codes Planned Connections Expressive Attention Receptive Attention -0.005 0.208 -0.053 2.149 0.957 0.033 0.046 4 3 0.03* Vocab. -0.151 -0.214 -1.622 -2.546 0.107 0.012 Written Transfer Planned Codes Planned Connections Expressive Attention Receptive Attention 0.030 -0.003 0.330 -0.027 0.742 0.978 0.001 4 1 0.39 Vocab. -0.176 0.031 -1.852 0.357 0.066 0.721 All Oral Planned Codes Planned Connections Expressive Attention Receptive Attention 0.021 0.211 0.237 2.176 0.813 0.031 0.036 4 2.3 0.05* Transfer -0.101 -0.206 -1.079 -2.444 0.282 0.016 All Written Planned Codes Planned Connections Expressive Attention Receptive Attention 0.115 0.030 1.274 0.299 0.205 0.765 0.007 4 1.3 0.29 Transfer -0.190 0.036 -2.012 0.415 0.046 0.679 *p < .05 137 E . Post Hoc Analyses The main analyses did not yield straightforward explanations of the relationships between cognitive processes, linguistic processes, and language variables. Results varied especially by task. Because of the results in the main analyses, and in the reanalyses, I did a series of post hoc analyses. The present study aimed in part to replicate Jarman's (1980a) study but with a larger sample of high IQ subjects than in Jarman (1980a). However, because of the low correlations within the data, it is difficult to compare the results of the present study with those of Jarman's (1980a). In Jarman (1980a) there were 52 subjects of high IQ in one sample, and for this group there was a high negative loading of Syntagmatic Clustering on the Successive Processes factor whereas the loading was positive for the average IQ group. In the present study the loading for the Syntagmatic tasks was also negative. I now propose several explanations for this outcome. I completed more analyses to understand the results and in particular the mechanisms at play producing different effects in low and high IQ subjects. a. Homogeneity and Heterogeneity of the Task Processes One explanation of the results may have to do with the degree of homogeneity or heterogeneity of the tasks. Indeed, the way cognitive and linguistic variables were measured in my study is worthy of note. Cognitive variables were measured by the Das-Naglieri test. 138 In that test, various tasks measure different processes and the results show whether a subject scores high or low on these processes, or in other words, the potential ability of a subject in the various processes. The linguistic processes tasks are different in the sense that the subject has to choose a syntagmatic answer or a paradigmatic answer. Thus, the linguistic tasks measure, at least partly, the subject's preference (or strategy) for using his/her potential ability in Syntagmatic or Paradigmatic Processes. In a linguistic task, an answer which is not paradigmatic is syntagmatic. Hence, a subject could have high syntagmatic ability and prefer to answer paradigmatically. By contrast, with the tasks of the Das-Naglieri test, there is no choice between a simultaneous and a successive answer. A subject scoring high on a successive task does not do it by choice because there is limited choice in the task; the subject can only give an answer via Successive Processes. As a consequence, it is possible for a subject to score high in Successive Processes as well as in Simultaneous Processes. This difference in kind of measurement may, at least partly, explain the results on relationships between the two kinds of tasks. Jarman's (1978) discussion on homogeneity and heterogeneity of tasks suggests how to understand the results found in both studies. Jarman (1978), in discussing Arthur Jensen's "Level I and Level IT theory of mental abilities (1970), indicated there are other dimensions besides "memory," and "reasoning" and "culture free," and "culture loaded" to classify tasks. He distinguished cognitive abilities and strategies, the strategies being connected to the executive control, and as a consequence, to choice. He suggested that more than two levels of classification may be used to classify tasks, and proposed an additional dimension: 139 heterogeneity and homogeneity of tasks. Jarman (1978) explained that: Homogeneity of strategy type refers to tests which are uniformly responded to by different groups of subjects... In contrast, heterogeneity of strategy would refer to those tests for which different abilities are used by different groups in task performance, (p. 262) Homogeneous tasks tap cognitive abilities and correspond to Jensen's tests where there is only one kind of answer. Heterogeneous tasks present a choice of answers according to the strategies used by subjects. The tasks of the Das-Naglieri test are relatively more homogeneous than are linguistic tasks used in the verbal learning tradition. In addition, Jarman (1978) notes that "Jensen appears to view abilities as rather immutable and tied directly to tasks" (p. 260), but that at the same time Jensen remarks that "for example, paired associate learning may be performed by subjects from varying backgrounds through the use of different degrees of Level I and Level II." Thus, Jensen implicitly recognized the role of executive control, and as a consequence, choice, in the way that a task is performed, but had no theoretical way to explain this control. Jarman's distinction between homogeneity and heterogeneity of tasks accomplished this, and is a key element in understanding the results of his study and of mine. As explained earlier, the Das-Naglieri tasks are mainly homogeneous, whereas the linguistic processes tasks are mainly heterogeneous, particularly the Free Word Association task. The concept of homogeneity-heterogeneity of tasks, coupled with the fact the subjects in Jarman's and my study had above average intelligence, then, may explain my results. 140 One may assume children with above average intelligence have high planning ability. That is, they have well developed executive control permitting them to choose the most efficient way to solve problems. If given a word classification task, one may reasonably assume they would more often prefer a paradigmatic way of classifying, even if they had high syntagmatic processing ability, because it is more efficient: it is easier to remember and to think about things classified by categories rather than in random successive order. One may also assume that subjects in the sample probably had a high level of Successive Cognitive Processes, as well as a high level of Simultaneous Cognitive Processes compared to the general population. When the norms of the Das-Naglieri test become available it will be possible to verify this hypothesis. Thus, if the previous assumptions hold true, the correlation between Simultaneous and Paradigmatic Processes would be positive because the subjects who scored high on Simultaneous Cognitive Processes, tended to be those who favoured way of classifying words as Paradigmatic. On the contrary, the correlation between Successive and Syntagmatic Processes would be negative because the subjects who scored high on Successive Processes may have still favoured paradigmatic word classification rather than syntagmatic in these particular types of tasks. This idea may at first examination appear to contradict Luria, who indicated that the contrast of Simultaneous and Successive Cognitive Processes corresponds to the contrast of Paradigmatic and Syntagmatic Linguistic Processes. However, this is not necessarily the 141 case. Luria1 s work was done with people with brain injuries, and who had no access to choice when given linguistic processes tasks. In some cases, the paradigmatic functions of Luria's patients had been damaged, while in others their syntagmatic functions had been damaged. In all those circumstances, Luria found direct equivalence between Syntagmatic and Successive Processes and between Paradigmatic and Simultaneous Processes, because if one was working, the other was injured, and thus blocked. Thus, subjects seen neurologically differ very substantially from subjects viewed educationally. b. High General Ability and Precocious Development Another factor could be that higher general ability leads to advanced development, which in turn may lead to an early Syntagmatic-Paradigmatic Shift. Further, when two tasks of different levels of restriction are presented to children, the more developmentally advanced children answer paradigmatically, especially in the least restricted task. A paradigmatic answer is more abstract in the sense that the classification is done at the conceptual level, as in a semantic net, whereas a syntagmatic answer is more concrete as it follows the general order of words in a sentence. In this study, the age of the subjects ranged from 6.9 to 9.3 years. This age range corresponds to the Syntagmatic-Paradigmatic Shift documented by various authors (Ervin, 1961; Palermo and Jenkins, 1963; Entwisle, Forsyth, and Muus, 1964; Entwisle, 1966; Palermo, 1971). Entwisle, Forsyth and Muus (1964, p. 19) give a clear definition of the Syntagmatic-Paradigmatic Shift: "Briefly, this shift implies that young children often respond with a word normally following the stimulus word in a sentence (GO-HOME), whereas older children and adults frequently respond with 142 a replacement word (GO-WENT)." Thus a syntagmatic or paradigmatic response to a task apparently depends on developmental level. c. Planning and Strategies From Jarman's (1980a) study and his article in 1978, it is seen that the level of Planning may have an effect on results associated with linguistic tasks. However, the nature of the task itself has also an important role to play. Executive control can only be exercised with a task offering a choice. In fact, the character of homogeneity and heterogeneity of task discussed earlier, determines the level at which Planning and strategies can operate in completing the task. Strategies and Planning may not be able to operate with homogeneous tasks because they do not offer much choice of answers. Nevertheless, executive control is an influential variable in the way individuals make choices and choose efficient strategies in linguistic tasks. My study included two linguistic tasks different from each other from the point of view of homogeneity and heterogeneity. In addition, I had a measure of Planning. Therefore, I could study the difference between linguistic type (Syntagmatic X Paradigmatic), task type (Free Association X Closed Association), and level of Planning (High X Low) to find out whether or not there was a triple interaction among the variables. In order to verify the effect of Planning, I divided the sample into a high Planning group and a low Planning group by taking approximately the highest 20% and the lowest 143 20% on Planning factor scores. The method used consisted of ordering the Planning factor scores obtained for each of the 152 subjects from the highest score to the lowest score. Then, I took the 30 highest subjects to form the High Planning group and the lowest 30 subjects to form the Low Planning group. I, then performed various comparisons between the two groups using correlations to find out whether there were qualitative differences in the processes used by each group, and analysis of variance to compare levels of mean performance between the groups. In the correlation analyses, two questions were asked. First, is the correlation between FAS (Free Association Syntagmatic) and FAP (Free Association Paradigmatic) different from the correlation between CAS (Closed Association Syntagmatic) and CAP (Closed Association Paradigmatic) when the entire sample is used? If there was a significant difference, it could be inferred that the two types of tasks measure different things. The second question asked in the analyses of correlations was: do the two types of tasks correlate differently in the high Planning group and in the low Planning group? If there were significant differences, it would indicate that high and low planners use different strategies to respond to tasks. The goal of the ANOVA was to find out whether there were significant mean differences on the performance level between the three groups of variables and if there were interactions among the variables. 144 To answer the first group of questions, the correlations between the linguistic variables were computed. They are presented in Table 18. Table 18 Correlations between the Linguistic Processes Tasks of the Low Planning Group and of the High Planning Group. Variables Correlation df p value CAS/CAP Total sample -.13 150 .050* Low Planning .17 28 .358/w High Planning -.38 27 .040* FAS/FAP Total sample -.63 150 .050* Low Planning -.78 28 .000* High Planning -.60 27 .004* CAS = Closed Association Test Syntagmatic; CAP = Closed Association Paradigmatic; FAS = Free Association Syntagmatic; FAP = Free Association Paradigmatic The table shows that the correlations between the linguistic variables were all significant (p<.05) except the correlation between CAS and CAP in the low Planning group. At the level of the entire sample, results show that the correlation between task types (Syntagmatic or Paradigmatic) were different for each paradigm. Both correlations were negative; however, the correlation between type for Closed Word Association was small (-. 13) whereas the corresponding correlation for Free Word Association was large (-.63). In addition, the two types of tasks correlated differently depending on the level of Planning. Indeed, there was no significant correlation between the CAS and the CAP in the low 145 Planning group whereas the same correlation was significant in the high Planning group. In addition, in the low Planning group, the correlation was positive, whereas it was negative in the high Planning group. In contrast, the correlations between FAS and FAP were both negative and were of a similar magnitude in the low and high planning group. In order to examine further the apparent differences between the paradigms and between the low and high Planning groups, the next step was to find out whether differences between paradigm correlations for the entire sample, and between low Planning and high Planning correlations within paradigm, were statistically significant. The significance of these correlation differences were calculated according to Ferguson's method (1971, p. 170). The difference between CAS/CAP and FAS/FAP for the entire sample was significant (p< .05). This result showed there was a significant difference between the answers on Closed Word Association and Free Word Association. From this it may be inferred the two types of tasks do not measure the same thing. The difference between the correlations CAS/CAP in the low Planning group and in the high Planning group was also significant (p< .05). However, the difference between the correlations FAS/FAP in the low Planning group and in the high Planning group was not significant ip > .05). These results indicated a significant difference in the way high planners and low planners answered the Closed Word Association, whereas there was no significant difference in the way they answered the Free Word Association task. After verifying whether there were qualitative differences in the way high and low 146 planners answered the linguistic tasks, an ANOVA With Repeated Measures analysis was performed to analyze the level of the effects of the variables and of their possible interaction. The computer program used one grouping factor (Planning) and two within-subject factors (paradigm, linguistic type of task). The analysis revealed a significant effect of paradigm, F(l, 58) = 13.20, p< .001, MSC = 1.73, a significant effect of linguistic type of task F(l, 58) = 20.01, /7<.001, MSC = 5.33, and a significant interaction between paradigm and linguistic type of task F(l, 58) = 91.57, p< .001, MS, = 4.78. Both graphs in Figure 9 confirm the interaction between paradigm and linguistic type (task). In the first graph, we see no effect of Planning except a small tendency between the paradigm and the level of Planning. However, the analysis did not show a statistically significant effect of Planning. The individual differences on Planning may not be large enough in the selected sample to show a significant effect of Planning. A distribution of the results on the four linguistic tasks is presented in FigurelO which shows the same trend of results as the ANOVA. In this figure the boxplots show the median and the four quartiles (each containing 25% of the data). The lines outside the CAS boxplot represent outliers. Some important differences between the tasks can be noted (see FigurelO). There was a large difference between the medians of CAP and CAS, CAS having the highest scores, whereas the difference was not so large between FAP and FAS. The median of FAP was higher than the median of FAS and the variance of FAP and FAS was clearly greater than that of CAP and CAS. 147 Thus, in summary, correlational data contain patterns which suggest that low and high planners seem to use different processes in their responses to these tasks but the ANOVA results show no performance differences in effectiveness between the planning groups. In short, the two planning groups have apparently used different means to accomplish similar performances. 148 2 o o cn c CC CD 8 Syntagmatic - tl .» " " u— —-* ti" Paradigmatic — L F r e e W o r d A s s o c i a t i o n C l o s e d W o r d A s s o c i a t i o n Paradigm 10 8 H CD 8 6 H CD ^ 1 2 -0 -Closed Word Association Free Word Association Parad igmat i c Syn t agma t i c Type of Task Figure 9. Graphs showing the effects of high and low Planning, paradigm used (Free Word Association and Closed Word Association), and the linguistic type of task used (Syntagmatic or Paradigmatic). 149 o FAS FAP CAS CAP Test Figure 10. Distribution of the results on the four linguistic tasks, CAP, FAP, CAS, FAS for the total sample (n = 152). 150 Chapter VI. Discussion A. Main Analyses The main purpose of my study was to investigate relationships among three sets of latent variables: linguistic, cognitive, and language production. On the one hand, I analyzed the relationship between the cognitive variables, Simultaneous and Successive Processes, and the linguistic variables, Paradigmatic and Syntagmatic Processes. On the other hand, I analyzed the effect of the same cognitive and linguistic variables on language variables-Grammar and Vocabulary Errors, and Diversity. I also investigated the effect of the Cognitive variables, Attention and Planning, on Transfer Errors. Complementary analyses developed naturally from the results of these three main analyses. a. General considerations In the previous chapter, I discussed some reasons that could explain the low correlations between the variables. The low reliabilities of the cognitive processes measures was probably the main cause of this result. Of course, this result applies only to my sample and cannot be generalized to any population. The subjects of my sample came from a population with special characteristics as described in the previous chapter. They were also at the beginning stages of learning a second language, and they learn all subjects in their second language. Because of this context, it is possible that their cognitive and linguistic processes do not follow regular patterns yet. In such case the measures do not have high 151 reliabilities. Homogeneity of the sample was another factor that was considered as a possible explanation for the low correlations. However, an examination of the data showed that the statistics of the sample did not support that explanation. An important consequence of the homogeneity of the French immersion population concerns the validity of the results of studies showing the benefits of immersion on cognitive ability, or academic results in various domains. These results could be due to the specific characteristics of the population attending French immersion schools rather than to the pedagogy used in such schools. If the positive academic and cognitive results of studies done in French immersion are due more to the characteristics of the students attending these schools than to pedagogy, the question of the effect of such schools on academic achievement for the general population remains unanswered. b. Parallelism between Cognitive and Linguistic Processes An important first question of my study concerned the parallelism between Simultaneous Cognitive Processes and Paradigmatic Linguistic Processes and between Successive Cognitive Processes and Syntagmatic Linguistic Processes. Theorists of verbal learning and language acquisition (e.g., Brown and Berko, 1960; Ervin, 1961; Jenkins, 1954) did research in Paradigmatic and Syntagmatic Processes in the verbal learning tradition, whereas neuropsychologists (e.g., Jakobson, 1971; Luria, 1973a; Pribram, 1971) worked on similar concepts, which they called Simultaneous and Successive Processes, as 152 part of research on aphasia. Jarman (1980a) was the first to give direct support to the parallelism between these two areas of research and to show a direct positive relationship between Simultaneous and Paradigmatic Processes and between Successive and Syntagmatic Processes. His study showed they were parallel concepts described differently in the two areas of verbal learning and neuropsychology. In the present study, I replicated Jarman's study in French Immersion second language learning using different linguistic paradigms to answer the question whether Paradigmatic and Syntagmatic Linguistic Processes were directly associated with Simultaneous and Successive Cognitive Processes respectively. I used two LISREL models to answer this question. Each of them comprised a cognitive and a linguistic latent variable. In one model the latent variables were Simultaneous and Paradigmatic Processes, whereas in the other Successive and Syntagmatic Processes were the latent variables. In the former, Simultaneous Verbal, Figure Memory, and Matrices were the indicators for Simultaneous Processes, whereas Free Word Association Paradigmatic and Closed Word Association Paradigmatic were the indicators of the Paradigmatic Processes. Word Series, Sentence Repetition, and Sentence Questions were the indicators of Successive Processes, whereas Free Word Association Syntagmatic and Closed Word Association Syntagmatic were the indicators of Syntagmatic Processes. Results of fitting the LISREL models to the data showed that the selected combinations of indicators were acceptable measures of the latent variables and that there was a negative relationship between cognitive and linguistic processes in both models. 153 Jarman (1980a) tested the same hypothesis with a two-sample factor analysis study. The first part of his study was done with 7-year-old pupils taken from the general population, and the second with high-IQ 9-year-olds. Jarman's (1980a) second study is the one most comparable to mine since my subjects were on average almost 9 years old (X = 8.7) and probably had a high I.Q. (Nielsen, 1983; Trites, 1986). Some of my results are similar to those of Jarman's (1980a) second study. For subjects taken from the general population, he found that the cognitive processes underlying Paradigmatic and Syntagmatic Clustering were Simultaneous and Successive Syntheses, respectively. For older subjects taken from a high-IQ population, the correspondence between competency in Successive and Syntagmatic Processes was negative. As in Jarman's (1980a) second study, the negative correlation I found between Successive and Syntagmatic Processes is ambiguous and difficult to explain because it contradicts the hypothesis of the neuropsychologists. Furthermore, and in contrast with Jarman (1980a), I found a negative relationship between Simultaneous and Paradigmatic Processes. This discrepancy between his and my might be due to differences in study design. I used two tasks, Free Word Association and Closed Word Association, whereas Jarman used only one, a Clustering task. Differences in the number of indicators used to measure linguistic processes, coupled with differences between the tasks may have influenced the results. An examination of the standardized estimates of the linguistic tasks in the LISREL models testing the correlation between cognitive and linguistic processes (see Figure 8) gives 154 an indication of the role each of the linguistic tasks played in the results. In the Simultaneous-Successive model the standardized estimates of the two linguistic tasks have opposite signs; Free Word Association Paradigmatic is positive and Closed Word Association Paradigmatic is negative. In the Successive-Syntagmatic model the standardized estimates of Free Word Association Syntagmatic is positive and the one of Closed Word Association Syntagmatic is positive but close to zero. The two tasks are supposed to measure the same latent variable. Opposite signs on their loadings suggest that verbal learning paradigms cannot be used indiscriminately since they may not measure the same thing. Differences between the tasks could bring about some confusing results. This leads to the question of whether all linguistic tasks measure the same constructs. Although linguistic tasks have been used indiscriminately in many verbal learning studies, some important characteristics distinguish them. Linguistic tasks have various levels of restriction of responses. They also require different levels of memory from one task to the other. For example, Free Word Association calls for unrestricted responses and does not require any memory other than remembering the stimulus word. In contrast, a Clustering task is associated with a different type of cognitive load; the number and kind of possible responses to the stimulus words presented is restricted. Further, the memory element inherent in the task distinguishes the task from the Free Word Association task. Closed Association is one of the most restricted tasks because the subjects can only choose between two given words to respond to each stimulus word; however, the memory element is not as significant as in a Clustering task. 155 In my study, I chose two tasks, one with a low restriction level, and one with a high restriction level. They were Free Word Association and Closed Word Association respectively. In contrast, Jarman used a Clustering task which requires more memory than any other task, has a higher restriction level than Free Word Association, but a lower restriction level than Closed Word Association. I argue below that tasks with a high level of restriction may elicit a high number of syntagmatic answers. Consequently, a high number of syntagmatic answers for this type of task would influence correlations between cognitive and linguistic tasks and partly explain the various results, especially in analyses relating Simultaneous Processes and Paradigmatic Processes. In short, the main difference between Jarman's second study and mine is that I used two linguistic tasks bearing different restriction levels, whereas Jarman used a sole task that also differed in restriction level from those I used. To examine the role of using two different paradigms in my analysis, I redid the analysis of the relationship between cognitive and linguistic variables using only one indicator at a time as a measure of the linguistic processes. I used Free Word Association in one analysis and Closed Word Association in the other as indicators of linguistic processes in the LISREL model. The purpose of using reduced models (models with only one linguistic processes measure) was to determine what role each of the linguistic tasks played in the results, and to bring the study more in line with that of Jarman who used one linguistic paradigm. However, the analysis did not converge, probably because of an insufficient number of indicators~a circumstance not recommended in LISREL. As an alternative 156 procedure, I factor-analyzed the data as in Jarman's (1980a) study. This time, the results were similar to those found by Jarman (1980a) for the general population and high I.Q. students. The analysis resulted in positive loadings for. the Paradigmatic tasks and in negative loadings for the Syntagmatic tasks. In addition, the loadings for Closed Word Association were relatively low compared to the loadings for Free Word Association and lower for the paradigmatic task than for the syntagmatic task. The previous analysis provided a possible explanation for the discrepancy between my results and Jarman's (1980a) results, by seeking the differences among verbal learning paradigms. Further, a comparison between cognitive tasks and linguistic tasks might help explain the apparently ambiguous results of my study and Jarman's on the relationship between Successive and Syntagmatic Processes. In his discussion of the negative results on the relationship between Successive Processes and Syntagmatic Processes with his high I.Q. sample, Jarman (1980a) suggested that attempts to relate Simultaneous Synthesis to Paradigmatic Clustering and Successive Synthesis to Syntagmatic Clustering should take into account both paradigms and population. I therefore compared verbal learning and neuropsychology paradigms, and I analyzed the effect of special characteristics of the population on the results. The main difference between linguistic tasks used in verbal learning and cognitive tasks used in neuropsychology is that some linguistic tasks offer a choice of response, paradigmatic or syntagmatic, whereas the cognitive tasks do not. Simultaneous and 157 Successive functions are measured by different tasks, and subjects may score high or low independently on each task. That is to say, if subjects function at a high level in both Simultaneous and Successive Processes, their results on the cognitive tasks show their high ability in each of the processes. In contrast, Paradigmatic and Syntagmatic Linguistic Processes are measured by two sets of the same task, and one set of tasks is scored for one kind of processes whereas the other set of tasks is scored for the other kind of processes. Although the measures obtained through the linguistic tasks are not ipsative measures, they are highly related. This is not the case for Simultaneous and Successive Cognitive Processes which are measured by different and independent tasks. As a consequence, the results of the cognitive measures and of the linguistic measures have different meanings. Consider Free Word Association for example. Subjects who function at a high level in Syntagmatic Processes as well as in Paradigmatic Processes could choose to answer in a mostly paradigmatic way. In this case, their high ability in Syntagmatic Processes would not appear in the results of the linguistic tasks but it would in the results of the cognitive tasks. Thus, responses on linguistic tasks indicate both a choice of mode and a level of ability in the chosen mode, whereas results on the cognitive tasks indicate only the ability of the subjects in the cognitive processes measured. In summary, results on linguistic tasks show a choice of strategy as well as a level of ability in the chosen strategy, paradigmatic or syntagmatic, whereas results on cognitive tasks show subjects' ability in each of the two processes, simultaneous and successive. Earlier research has not previously made this distinction because researchers on verbal 158 learning usually work with 'normal' subjects, whereas neuropsychologists mainly work with subjects who have lost one of their coding functions. Choice is not possible for the latter subjects; they use their intact coding function (paradigmatic/simultaneous or syntagmatic/successive) to answer linguistic or cognitive tasks. In these cases, there is no ambiguity in the meaning of the results of studies analysing the relationship between the concepts in the two fields. In a study of the parallelism between cognitive and linguistic processes, subjects' characteristics within each population are important if they bear on the strategies subjects choose to respond to linguistic tasks. The degree to which subjects' characteristics may affect task results is governed by the degree of homogeneity or heterogeneity of tasks (Jarman, 1978). This measure refers to the degree to which group-population-specific strategies can be used in solving the tasks and particularly the degree to which the use of these strategies influences the results on the tasks. The Das-Naglieri tasks are relatively homogeneous in the sense that there is little choice of strategic pattern in the answers. The subjects' answers to a task indicate their ability on the function measured by the task. In contrast, the linguistic tasks are rather heterogeneous, some tasks being more heterogeneous than others. The level of heterogeneity of a linguistic task varies with the level of constraint of the task. For example, a low constraint task such as Free Word Association is highly heterogeneous since it offers more possibilities of using strategies, whereas a high constraint task such as Paired Associate, which offers very little possibilities of using strategies, has a low level of heterogeneity. Scores on linguistic tasks, therefore, indicate a choice of 159 strategy and not only the subjects abilities in the functions measured by the tasks. This is true in direct proportion to the level of heterogeneity of the task. Pupils whose abilities are high in both linguistic functions are, in principle, equally likely to choose a paradigmatic over a syntagmatic answer to questions. However, when tasks have a high level of restriction of strategy choice and are less heterogeneous, different strategies used by subjects will have varying degrees of success. Subjects with high planning ability, for example, will be notably more successful than subjects with low planning ability because they are better able to adapt to the optimal strategy requirements of the tasks. Conversely, when tasks are of low restriction in strategy choice, that is, are more heterogeneous, low and high planners will differ less in their performances because of the open-adaptive possibilities that exist in the test for all types of subjects. The strategy patterns of each subject also depend on his or her individual characteristics. In a study such as the present one, the strategy patterns of the sample depend on the general characteristics of the population to which they belong. Thus, whereas the results of Jarman's first study corresponded to the hypothesis independently held by previous theorists in the verbal learning and neuropsychology fields, ambiguous results were found in both Jarman's second study (1980) in my study. An important difference between these studies and their opposite results, is the characteristics of their samples. The subjects in Jarman's first study were taken from the general population and had a mean chronological age of 7.6 years, whereas the subjects of his second study 160 were older and had a higher I.Q. as did subjects in my study. In my study, subjects were almost 9 years old ( X = 8.7). Various studies had previously shown that children older than 8 give mostly paradigmatic answers to linguistic tasks (Ervin, 1961; Entwistle, 1966; Entwistle, Forsyth, & Muus, 1964; Palermo, 1971; Palermo & Jenkins, 1963; White, 1965). In addition, the I.Q. of French immersion students is typically superior to the average (Nielsen, 1983; Trites, 1986). These findings help to explain the apparently ambiguous results in Jarman's and my studies. The subjects with high general ability scored high on cognitive tasks. They scored high both in Simultaneous Processes and in Successive Processes. However, their score on the linguistic tasks depended upon the level of restriction of the paradigm, and thus, the extent to which they could use different strategies to answer the question. In the case of Free Word Association, the least restricted task, they had the opportunity to choose the way they answered. Therefore, they generally demonstrated more strength in the paradigmatic category, as seen in Figure 9. In that case, Simultaneous and Paradigmatic Processes should be positively correlated and Successive and Syntagmatic Processes should be negatively correlated. On the other hand, with Closed Word Association which has a higher level of restriction, their score on Paradigmatic Processes decreased somewhat and their score in Syntagmatic Processes increased (see Figure 9). This time, the correlation between Simultaneous and Paradigmatic Processes should be negative and the correlation between Successive Processes and Syntagmatic Processes should be positive, as found by Jarman (1980a) and in my study. 161 In summary, the parallelism between cognitive and linguistic processes is not straightforward. It depends on the paradigms used to measure the linguistic processes, and on the characteristics of the population from which the sample is selected. When subjects have high ability, as was the case in my study, the parallelism between Simultaneous and Paradigmatic Processes is clear whereas it is less clear between Successive and Syntagmatic Processes. This is due to the range in level of heterogeneity among the linguistic tasks, and the relative homogeneity of the cognitive tasks. This question is studied and discussed in more detail in section "B." of this chapter entitled: "Ad Hoc Analyses". c. Results about Language Variables Descriptive Statistics Descriptive statistics on language errors throw new light on published research on transfer errors, and on the methodology of grammar teaching in French immersion. The results on the relative number of Transfer Errors were surprising. Although most writers give great importance to these errors, analysis of the results in this study showed they represented only 16% of the total number of errors. This percentage agrees with Dulay's and Burt's research (1973), who found them to represent 15% of all errors. Further, there were individual differences among pupils in number of Transfer Errors, whereas writings on that topic leave the impression that all French immersion students make the same kinds of transfer errors with the same frequency (Hammerly, 1989; Lyster, 1987; Pellerin and 162 Hammerly, 1986). The standard deviations were .9, 2.3, 2.3, and 1.2, for Oral Transfer Grammar Errors, Oral Transfer Vocabulary Errors, Written Transfer Vocabulary Errors, and Written Transfer Grammar Errors respectively. Among Transfer errors, there were more Vocabulary Transfer Errors than Grammar Transfer Errors. Seventy eight percent of the Transfer Errors were Vocabulary Errors and 58% of these were Oral Vocabulary Transfer Errors. As Luria (1966) indicated, these kinds of errors are produced at the time of word selection. These are mostly Vocabulary Errors and would occur more often in oral situations which are constrained by time than in written situations. Within Grammar Transfer Errors, Oral and Written Transfer Errors were equivalent in numbers. These errors have a different origin than the vocabulary errors described in Luria. Such errors may be due to a lack of knowledge of syntax structure in French which is compensated for by an English structure, rather than being the result of a lack of control in sentence or morphology selection. These descriptive results call for recommendations concerning language pedagogy in French Immersion. Transfer Errors may be corrected by intense practice and repetition. However, the results of this study showed that the greatest number of errors are Grammar Errors (57% of all errors in oral production and 65% of all errors in written production). Most of these grammar errors were verb tenses or verb conjugations. Because there is usually one verb per sentence or clause, the probability of making grammar errors may be larger than the probability of making vocabulary errors. Nevertheless, there is no grammar curriculum in French Immersion because one principle of the communicative approach is that grammar is learned unconsciously in context as it is in the learning of a first language. 163 Communication of a "natural" and authentic sort is the main principle of French learning in French immersion schools. To teach grammar systematically would seem regressive to teachers as it implies a translation method leading from simplest to most complex structures. One advantage of creating a grammar curriculum that would set a minimum number of grammar objectives for mastery at each grade level would be the prevention of repetitive teaching of plurals and gender agreement from Grade 1 to Grade 7, as is often now the case. Further, grammar could be taught with a communicative purpose. Some specific language structures could be taught by showing their function for the purpose of precise and accurate communication depending on the topic, the medium, and speaker's intention. For example, the knowledge of the imperative mode is necessary to read and write recipes, or for advertisement. Statistical Analyses Three questions concerned the relationship between Cognitive and Linguistic Processes and language variables. The first question had to do with the relative importance of Simultaneous Cognitive Processes and of Paradigmatic Linguistic Processes in predicting Vocabulary Errors other than Transfer Errors. The second question was about the relative importance of Successive Cognitive Processes and Syntagmatic Linguistic Processes in predicting Grammar Diversity and Grammar Errors other than Transfer Errors. And finally, based on Luria's theory, the third question asked whether Control, defined as Planning and Attention, was a significant predictor of Transfer Errors. 164 The LISREL models used to answer these questions comprised three latent variables: Cognitive Processes, Linguistic Processes, and language errors as represented in Figures 1, 2, and 3. These LISREL models did not converge, most likely because of the low level of correlations. A reanalysis using three groups of six multiple regressions, corresponding to the three questions, was conducted to answer these questions with more statistical power than with LISREL. However, multiple regression methodology is also based on correlations and only three of the eighteen analyses were significant. These were the regressions of Simultaneous Processes on Written Vocabulary, of Planning and Attention Processes on Oral Transfer Vocabulary Errors, and on All Oral Transfer Vocabulary Errors. The lack of variability in the sample prevented a clear conclusion about the relationship between Cognitive and Linguistic Processes and language variables. The strongest relation was between Transfer Errors and Planning and Attention. Yet, the results were not significant for all the different kinds of transfer errors. They were only significant for Oral Vocabulary Transfer Errors and not for Grammar Transfer Errors or Written Transfer Errors. In addition, none of the regression analyses of Linguistic Processes on Vocabulary and Grammar Errors were significant. One interesting result of the regression analyses is the significant relationship between Planning and Attention Processes and Oral Transfer Errors. Luria often stated that one key symptom of frontal lobe patients was that, although they knew a rule and were able to repeat it and verbally admit their errors, they were unable to correct them (Damasio, 1985). Planning and Attention are most necessary when pupils communicate aloud because the time 165 for thinking and producing an answer is limited. The more Planning and Attention, the more control on the production and the fewer errors. In contrast, writing is more leisurely and there is less pressure to communicate quickly. In this case, individual differences in Planning and Attention may be less related to the number of errors in the production of a text than it is in oral communication. In summary, Vocabulary Errors constituted the main element in Transfer Errors and were the most numerous in oral situations. Results of regression analyses indicated that, in oral production, these errors were related to Planning and Attention Processes. B . A d Hoc Analyses The literature in neuropsychology and in aphasia is not easily reconciled with Jarman's and my findings. On one side, researchers in verbal learning use many kinds of paradigms to measure linguistic processes and consider that they all measure the same thing, whereas I found that using different tasks to measure linguistic processes resulted in markedly different conclusions. On the other side, researchers in neuropsychology have declared that there is a direct isomorphism between Cognitive and Linguistic processes. However, the findings of Jarman's and my study about this equivalence were mixed. Results varied according to the tasks used for measuring linguistic processes, and according to subjects' characteristics. Such mixed results invite an explanation. 166 Individual differences on the Planning variable may have had a role to play in the way pupils responded to the tasks. According to Luria (1970), the main function of the third block of the brain (the frontal lobes) is the formation of intentions and their translation into programs for behaviour. In addition, language plays a key role in the regulation of behaviour as well as in the formation of higher processes (Luria, 1966; Vocate, 1990). Damasio (1985, p. 362) indicated that people with frontal lobe injuries often showed impaired fluency, as measured by verbal association tests, although their memory was not affected and no change could be detected in their speech output. Thus, Planning regulates verbal behaviour in verbal tasks through the mediation of language. In addition to Planning, which is a subject variable, the concept of homogeneity and heterogeneity, which is a task variable, may be of central importance in explaining the unexpected results of the mixed results on parallelism between Cognitive and Linguistic Processes. A task is homogeneous when everyone answers it in the same way because of the low possibility of using variations in strategies to answer the tasks. Inversely, a task is heterogeneous when individual differences among subjects result in different ways of answering the task because of the possibility to use various strategies to answer that task. Thus, high and low planners may use quite similar strategies to answer a constrained task such as Closed Word Association which offers subjects a choice between only two words and allows few strategies to enter into this choice, whereas Free Word Association task, in which various routes to answer are possible, leaves more possibilities in variation in response strategy. In the latter case, individual differences in strategy choice have little effect on the 167 accuracy of response. Thus, the more constrained a task is, the less strategies may be used to deal with these constraints, particularly in subjects with weaker strategy repertoires. Therefore, although it appears initially counter-intuitive, executive control only becomes a significant source of performance differences as task restrictions increase, and as tasks become less heterogeneous. To verify the possible effect of the task variable, homogeneity-heterogeneity, and the individual difference variable, Planning, as well as the possible effect of their interaction, I did two kinds of analyses. First, I analyzed the correlation patterns between the variables to see whether the processes were qualitatively the same according to the level of Planning, and to the level of restriction of the task used. Second, I did an analysis of variance to verify the degree of significance of the differences on mean performance on the processes according to the level of Planning and the kind of linguistic task used. The results of the correlations showed that there was a correlation between the linguistic types of each task except for Closed Word Association in the low Planning group. The difference between the correlations CAS/CAP and FAS/FAP for the entire sample was significant (p< .01). Furthermore, the differences between low Planning and high Planning correlations were significant only for Closed Word Association (p< .05). The analysis of the mean differences between the variables revealed significant effects of linguistic type and task, a significant interaction between task and type (p<.0001); however, there was no 168 significant effect of individuals' Planning efficiency. The significant difference between the correlations CAS/CAP and FAS/FAP for the entire sample showed that there was a qualitative difference between the two tasks Closed Word Association and Free Word Association. The high correlation between Free Word Association Paradigmatic and Syntagmatic showed that the subjects responded very similarly to the task, be it Paradigmatic or Syntagmatic. On the contrary, the correlation between Closed Word Association Paradigmatic and Syntagmatic was low. It showed that on that task the subjects' patterns of paradigmatic and syntagmatic answers were different. This difference between Free Word Association and Closed Word Association was significant and showed that the degree of heterogeneity of the two tasks is different. Free Word Association is relatively more heterogeneous and Closed Word Association is relatively less heterogeneous. Thus Planning, which is a subject characteristic, should have a different effect on each paradigm. Indeed, the differences between the correlations of the high and low Planning groups showed that subjects with different characteristics responded similarly to some tasks and not to others. From my discussion on homogeneity and heterogeneity of tasks, and the above finding, a difference between the way high planners and low planners responded to Closed Word Association should be expected if the task is less heterogeneous, and no significant difference should be expected for Free Word Association if the task is more heterogeneous. The results confirmed this expectation. Low and high planners used different strategies with 169 the Closed Word Association task whereas this was not the case for the Free Word Association task. Differences among variables on mean performance was analyzed with an ANOVA with Repeated Measures. This analysis was performed mainly to show the effect of Planning on the level of performance on the various paradigms and linguistic types of tasks. However, the analysis revealed no significant difference between high and low Planning, although the lowest and the highest 20% were chosen from the total sample to make up the two groups. The non-significance of Planning could be explained by the sample characteristics of this particular study. First, French immersion pupils have a higher mean IQ than pupils in regular schools. Although neuropsychologists (Damasio, 1985; Zangwill, 1966) have shown that standard intelligence tests do not measure frontal lobes functioning level and that patients with frontal lobe lesions still achieve at an average level on IQ tests, I assume that in general, a high IQ is associated with high Planning ability. Further, the range in ability was further restricted by the choice of the sample among French immersion pupils. All pupils getting help from the Learning Assistance Teacher, or those who could not write 100 words for the composition, or those the teacher identified as "having problems", were eliminated from the sample. In consequence the final group was quite homogeneous. They most probably had a high academic standing and their level of Planning was likely to be uniform. Thus, the final sample had a restricted range of intellectual ability and probably of Planning also, at the top 170 of the general distribution of the Planning variable. There was no significant difference between the low and high Planning group, and the effect of Planning on tasks and paradigms could not be shown. The specific selection of the sample results in an analysis having the smallest probability of showing the effect of Planning on performance in the various tasks. For this reason, any trend in the results of the analysis should be considered, since a more heterogeneous sample for Planning would in any case show a larger effect and larger differences between the means. Differences among the performance means on the linguistic tasks between the high Planning and the low Planning groups showed that low planners scored generally higher on Syntagmatic Processes and high planners score generally higher on Paradigmatic Processes. If this difference were significant it could mean that high planners have a precocious development and pass the Syntagmatic-Paradigmatic Shift earlier than others and therefore have more practice in it. The other noticeable difference among the variables between low and high planners was on Closed Word Association Paradigmatic, where high planners scored rather higher than low planners whereas there was no difference between the two groups on Free Word Association. Here again, if this difference were significant, it would show that high planners use processes and strategies with less heterogeneous tasks that are different and more successful than those used by low planners. However, it is important to keep in mind that these differences were not significant and that further analyses are needed to confirm the significance of the trends seen in the data of my study. In addition to the non-significant effect of Planning, the ANOVA showed a main effect 171 of paradigm, a main effect of linguistic type, and a significant interaction between the two. The main effect of paradigm was significant but not interesting in itself. The main effect of linguistic type showed that there was a significant difference between syntagmatic and paradigmatic performance when nothing else was taken into account. Since the subjects were 9 years old, they were still at the developmental period corresponding to the Syntagmatic-Paradigmatic Shift, thus they were expected to score higher in the well-practised Syntagmatic Processes than in Paradigmatic Processes. An examination of the means confirmed this expectation. The most interesting result of the analysis of variance was the interaction effect between paradigm and linguistic type. Considering the heterogeneous quality and the low level of constraint of Free Word Association relative to Closed Word Association, and the Syntagmatic-Paradigmatic Shift, one should expect a similar performance on Syntagmatic and on Paradigmatic Free Word Association and a greater difference between the Syntagmatic and Paradigmatic performance of Closed Word Association, Syntagmatic being higher than Paradigmatic. An analysis of means showed that the subjects scored slightly higher on Free Word Association Paradigmatic than on Free Word Association Syntagmatic. On this relatively heterogeneous task the subjects were able to use their recently acquired Paradigmatic Processes because of the low constraint of the task. Meanwhile, subjects scored higher on Closed Word Association Syntagmatic than on Closed Word Association Paradigmatic. As to the level of performance between Free Word Association and Closed Word Association, the performance on Paradigmatic Processes went slightly down from Free 172 Word Association to Closed Word Association whereas the performance on Syntagmatic Processes almost doubled. Because the subjects were at the point of the Syntagmatic-Paradigmatic Shift, they used equally well the two linguistic processes on the task without constraint and reverted to their most practised kind of linguistic processes on the task with more constraints. Figure 9 shows the results previously described, especially the interaction effect between paradigm and linguistic type of task. Some trends could be derived from the graph (see Figure 9 ) showing the effects of Planning, paradigm used, and linguistic type. Note that these data are only suggestive and should be treated cautiously. Concerning the linguistic tasks of the Paradigmatic type, high planners and low planners scored almost at the same level on Free Word Association whereas high planners scored higher than lower planners on Closed Word Association. An explanation can be found in the difference between the two tasks. Free Word Association is a heterogeneous task, with no constraint. No feedback was given to the subjects, and therefore, no planned strategy could be applied to the task by any of the two groups. On the opposite, Closed Word Association is a less heterogeneous task and high planners could develop a more successful strategy for doing it than low planners and achieve better . For the linguistic tasks of the paradigmatic type the level of constraint of the task made a difference between high and low planners. In that case, the former were able to develop strategies helping them to achieve better than the low-level planners. This hypothesis remains to be verified in further 173 research. Results were different on the linguistic tasks of the syntagmatic type. Low planners scored the same as high planners on Free Word Association Syntagmatic and on Closed Word Association Syntagmatic. For the linguistic tasks of the syntagmatic type, the level of constraint of the task does not make any difference between high and low planners because Syntagmatic Processes are well practised in both groups. Figure 9 shows that for the total sample, performance increased on the syntagmatic type of task, and decreased on the paradigmatic type of task as the constraint of the task augmented. This is also visible in FigurelO, which shows the distribution of each paradigm by linguistic type. This profile on type of task corresponds to the Syntagmatic-Paradigmatic Shift observed in numerous studies (Ervin, 1961; Entwisle, 1966; Entwistle, Forsyth, & Muus, 1964; Palermo, 1971; Palermo & Jenkins, 1963; White, 1965). The subjects of my study were about 9 years old and had a high IQ. Thus, they were over the shift point or just passing it. Their Syntagmatic Processes were well developed whereas their Paradigmatic Processes were still developing. There were less individual differences on Syntagmatic Processes than on Paradigmatic Processes. For each task, the standard deviation of the paradigmatic type of tasks was slightly greater than the standard deviation of the syntagmatic type of task (FAP: SD=2.3; FAS: SD=2.2; CAP: SD = 1.4; CAS: SD = 1.3). The correlation and standard deviation patterns as well as the results of the analysis of variance fit the above explanation of the trends in the results presented in Figure 9. Yet the 174 data do not permit a strong conclusion, but rather, invite hypotheses to be tested in further studies. In my study, it was not possible to find differential effects between high and low planners because of the homogeneity of the total sample on the Planning variable. A model of the general grade 3 population should show a full range of abilities and individual differences and would probably show a differential effect between high and low planners. I have tried to summarize in a theoretical model (see Figure 11) the hypotheses I have developed from my results on Planning, linguistic type of task, and level of constraint of task. This model applies to the general population of 9-year-old grade 3 students. This model shows the theoretical interaction between Planning (high and low), paradigm (more or less heterogeneous), and linguistic type of task (syntagmatic or paradigmatic). The paradigms represented are currently used to measure linguistic processes. Those paradigms are placed on the abscissa according to an ordinal scale, from the least constrained task (Free Word Association) to most constrained (Paired Association Task). In this ordering of the tasks, memory is confounded in amount and type. The tasks go from Free Word Association to Paired Associate, from the most heterogeneous to the least heterogeneous tasks. Strategy differences are highest on heterogeneous tasks of the paradigmatic type. 175 CD O cr CC E € CD C L w .2 5 o H Paradigmatic ... * '.. L -— H Syntagmatic # L Low Constraint High Heterogeneity High Constraint Low Heterogeneity Figure 11. Theoretical model showing the level of performance on various paradigmatic and syntagmatic linguistic tasks depending on the level of Planning. 176 The level of constraint from low to high has to do with the number of possible strategies that subjects may use to create a response in each task. In the least constrained tasks, the subjects can answer using a great variety of strategies, whereas in the most constrained task, subjects are restricted to one possible answer and have access to only few strategies to answer the task. This is shown in the following description of linguistic tasks. Free Word Association is the most heterogeneous task. There is no constraint in that task. As a result, subjects with different abilities tend to complete the task the same way, with similar efficiencies. Stimuli words are given one after the other and subjects have to give the first word that comes to their mind. They can choose any word. Mediated Free Word Association is similar to Free Word Association except that the subjects receive individual training in paradigmatic response. This training constrains the task in that it influences the subjects' responses, and therefore they will tend to respond with their own strategies. Free Recall asks subjects to remember as many words as they can from two lists presented one after the other in random order: a list of clusterable nouns and a list of non-clusterable nouns. For example, a clusterable list could be 12 nouns that could be classified in 3 general categories and a non-clusterable list could be 12 nouns with no common characteristics. In this task, the number and kind of possible answers are restricted to the words presented by the experimenter. 177 In a Clustering Task two lists of words paired paradigmatically or syntagmatically are prepared. The words are put in a random order with the restriction that two words in a pair are not adjacent. In addition, the first and the last word are filler words. The score takes into account the number of pairs of words recalled together. This task is more restricted than free recall because subjects have to recall words in pairs rather than a total number of words in a random order. Multiple Choice Task comprises a list of words presented one after the other with various choices of answers, some syntagmatic and some paradigmatic. Subjects have to choose one among multiple choices. In this task, subjects' answers are restricted to only three or four choices of words. Closed Word Association is like a multiple choice task with only two words to chose from, a paradigmatic word and a syntagmatic word. In this case, subjects have only the choice between two words. Finally, the most constrained task is the Paired Association Task because there is only one possible answer for each item. In that task two lists of words are prepared, a paradigmatic word-pair list and a syntagmatic word-pair list. Each list consists of 10 pairs of words. The examiner reads the paired list to the subject. Then, the first word of each pair is presented in random order and the subject has to remember the second word of the pair. 178 In summary, when an infinite number of answers is possible as in Free Word Association Task, strategies are not particularly helpful especially in the absence of feedback. In the least heterogeneous tasks, on the other hand, a subject may apply strategies in order to perform better according to the given constraints. As the level of constraint increases, the number of appropriate possible strategies decreases. For syntagmatic types of tasks, high planners and low planners perform almost at the same level, essentially because 9 year old subjects have well integrated the syntagmatic structures. For paradigmatic kinds of tasks, larger individual differences still pertain among 9 year old subjects and the effects of different levels of strategies are apparent. These features are expressed in Figure 11, which could serve as the basis for future research. C. Summary of Main Findings First, I found that the ratio of Transfer Errors relative to the total number of errors is smaller than Grammar Errors or Vocabulary Errors. Grammar Errors represented more than 50% of all errors made whereas Transfer Errors accounted for only 5 %. This counters recent research , although Dulay and Burt had found a similar result in 1974. Second, I found that the tasks of the Das-Naglieri CAS test did not measure the processes of the PASS model as clearly as the literature shows. According to the authors of the test, Expressive Attention , for example, should load on Attention. Instead, I found that for my sample it loaded on Planning. It has been shown (Naglieri & Das, 1988) that 179 some measures loaded on different processes depending on the cognitive stage of development of the subjects. Third, I found that Written Vocabulary Errors were predicted by Simultaneous Processes and that Oral Transfer Errors were predicted by Planning and Attention. Fourth, I found that the relationship between Cognitive and Linguistic Processes was far more complex than ever thought before. As Jarman (1980a), I found that Paradigmatic Processes were positively related to Simultaneous Processes and that Syntagmatic Processes were negatively related to Successive Processes. I explained that this result was different from what was proposed by Luria (1975) and Jakobson (1971) because of the difference in the population studied and in the tasks used to measure Linguistic Processes and Cognitive Processes. Finally, I also found that not all linguistic tasks were equivalent and I proposed a hypothetical model about the various linguistic tasks for future research. D . Limitations of the Study The main limitation of this study is the low reliability of the tests for Cognitive Processes and its consequences on the various analyses as discussed in the Results Chapter. 180 Research of an experimental nature as proposed hereafter would be more suited to the specific population attending French immersion schools. Furthermore, the sample in this study comes from French immersion classrooms, and is unrepresentative of the general Grade 3 population. Some results may be specific to French immersion and may generalize only to that population. E . Suggestions for Future Research The present study had an exploratory goal because, in the literature, there did not seem to be studies relating second language production in French immersion schools to cognitive and linguistic processes. This work invites research on further questions. a. Second Language Learning This study classified errors in three groups: vocabulary, grammar, and transfer. The classification was based on an hypothesis of correspondence between cognitive and linguistic processes on the one hand, and these three categories of errors on the other. More refinement in classification of errors, especially in grammar errors, would produce higher precision in results. The data could also be analyzed from a different perspective, comparing written and oral errors in relation to linguistic and cognitive processes. Finally, true experiments bearing directly on various ways to learn a second language could be done. These three possible further lines of research deserve explication. 181 First, a more precise classification of language errors would result in a greater understanding of the relationships between errors and cognitive and linguistic processes. Grammar, for example, is not only a successive function. One might argue that some elements derive from Successive Processes and others from Simultaneous Processes. Morphological rules of agreement, such as subject-verb, or gender and number agreements usually apply to words placed close to each other in a succession, and thus correspond to Successive Processing. On the other hand, rules governing the sentence or paragraph structure take into account the entire sentence or paragraph, and the implied processing is Simultaneous. Luria (1975a) distinguished between patients "experiencing considerable difficulty in the analysis of logico-grammatical constructions expressing relationships [and those who could not easily] cope with the analysis of syntactical mistakes in contextual speech" (p. 49). A separation of grammar errors according to Luria* s distinction could test whether sentence construction errors correlate with Paradigmatic or/and Simultaneous Processing and whether morphological errors in contextual speech correlate with Syntagmatic or/and Successive Processing. Second, an analysis of data to compare oral and written errors in their correlation with Successive and Simultaneous Processes would be interesting. Since oral language is produced in a linear fashion, it may correlate with Syntagmatic and Successive Processes, whereas written language may correlate with Paradigmatic and Simultaneous Processes. Third, true experiments could be conducted to test the effect of remediation. One such 182 remediation might be training in Successive, Simultaneous, or Planning Processes, or any combinations of these, while a control group would receive no training. The effect of these various kinds of training on errors would be calculated and tested for significance. A second type of training might be the introduction of a grammar curriculum with a pilot group starting in grade one. This group would be compared to a regular French immersion group of pupils. The pilot group would receive systematic and explicit grammar teaching according to a specific grammar curriculum. A strictly communicative approach would remain the preferred one for teaching and learning French in non-experimental French immersion schools. b. Planning Planning was an important variable in the present study. However, research interest in this variable has developed only recently in psychological, neurological, and linguistic circles. Planning is still not well understood. Some main difficulties have been to define it as different from any other processes, to find its effect on other variables, and to measure it. Some of these difficulties arose in the present study. The definition of Planning came from Luria's neurolinguistic research. However, the use of Das-Naglieri tests to measure Planning did not produce the expected results. In the PASS model, Expressive Attention measures Attention. However, in this study that task invariably loaded on Planning. Thus, I used it for the rest of the study as a task measuring Planning. This begs the question of the stability of the PASS tasks. Is it possible that different tasks load on different processes according to the sample used to validate the kind of processes they measure? One hypothesis 183 to explain this loading would be that the test measures Attention up to a certain level, and Planning after that basic level. According to Luria (Das, Naglieri & Kirby, 1994) the first and third functional units of the brain are closely related. The organization of the Attention system helps to understand this relationship. Selective Attention has various components, the main ones being Arousal and Planning. In Luria's terminology general attention is similar to Arousal and is necessary for Selective Attention. It seems obvious that a high level of Planning implies a threshold level of Arousal. Similarly, Selective Attention cannot function without the mediation of the Executive Control. Das, Naglieri, and Kirby (1994, p. 49) indicate that the Stroop test measuring Expressive Attention involves selective inhibitory processes which are closely related to the control of cognitive processes. In the present study, most students displayed a high level of Planning. One could imagine a threshold beyond which the achievement on the Stroop test would reflect so much the involvement of the Executive Control that Expressive Attention would load more on Planning than Attention. In this study, only a few students made mistakes on the test. The variation among them was the time taken to accomplish the task. A study investigating the loading of Expressive Attention according to Planning level could verify this hypothesis. In regard to Planning, another difficulty in the present study was the homogeneity of the subjects on that variable. A study analysing the interaction of Planning, tasks, and paradigms would be worth conducting with two groups presenting a larger difference on Planning or with one group with a larger variance on that variable. In this way, interaction between Planning and tasks could be more easily verified. 184 c. Task Specificity The various linguistic tasks found in the literature are not equivalent measures, as this study shows. Further, different results could be obtained with different tasks according to level of Planning. This study did not show any significant difference between high and low planning groups for the results by paradigm and type of tasks because of the homogeneity among the subjects on Planning. However, I have proposed a theoretical model that would apply to the general population of 9-year-olds where the range of Planning would not be restricted as it was in my study (see Figure 11). This model remains to be tested. Future research should test this model. It would show precisely whether the paradigms are equivalent measures or not. This would already be a significant step in research, since they are now considered as equivalent despite the absence of any demonstration of this. It would also show whether other variables, such as Planning, interact with the various paradigms in different ways. Knowledge among researchers about the differences between linguistic paradigms would indisputably enhance linguistic research. Researchers could in consequence choose the paradigm appropriate to their research question instead of choosing any paradigm, by assuming the empirical equivalence of all linguistic paradigms. 185 References Abbott, R. D., & Beminger, V. W. (1993). Structural equation modelling of relationships among developmental skills and writing skills in Primary - and intermediate grade writers. Journal of Educational Psychology, 85, 478-508. Anderson, J., & Gerbing D. W. (1984). The effects of sampling error on convergence, improper solutions and goodness-of-fit indices for maximum likelihood confirmatory factor analysis. Psychometrika, 49r 155-173. Baron, J. (1987). Tools for studying human intelligence. Contemporary Psychology, 32t 135-136. Beauvois, J. L., & Ghiglione, R. (1981). L'homme et son langage [Man and his language]. Paris: Presses Universitaires de France. Berko Gleason, J. (1985). The development of language. Toronto: Charles E. Merrill Publishing Company. Bialystok, E. (1978). A theoretical model of second language learning. Language Learning, 28. 69-83. Bialystok, E. (1988). Levels of bilingualism and levels of linguistic awareness. Developmental Psychology, 24, 560-567. Bialystok, E. (1990). Communication strategies: A psychological analysis of second-language use. Oxford: Basil Blackwell, Inc. Bollen, K. A. (1989). Structural equations with latent variables. Toronto: Wiley & sons. Boomsma, A. (1983). On the robustness of LISREL rMaximum Likelihood Estimation') against Small Sample Size and Nonnormality. Amsterdam: Sociometric Research Foundation. Bournot-Trites, M. (1991). French immersion: Myths and reality [Review of French immersion: Myths and reality by Hector Hammerly]. Canadian Journal of Education, 469-471. Bournot-Trites, M., Jarman, R. F., & Das, J. P. (1995). Luria's language theory within a cognitive theory: A Canadian perspective. Aphasiology, 9, 123-135. 186 Brooks, N. (1964). Language and language learning. New York: Harcourt, Brace & World. Brown, R. W., & Berko, J. (1960). Word association and the acquisition of grammar. Child Development, 31, 1-14. Brown, R., & McNeill, D. (1966). The "tip of the tongue" phenomenon. Journal of Verbal Learning and Behaviour, 5, 325-337. Carroll, J. B. (1988). Cognitive abilities, factors, and processes. Intelligence, 12, 101-109. Chomsky, N. (1959). Review of Verbal Behaviour by B. F. Skinner. Language, 35, 26-58. Clemans, W. V. (1956). An analytical and empirical examination of some properties of ipsative measures. Psychometric Monographs, 14 (Serial No. 14). Cole, M., Frankel, F., & Sharp, D. (1971). Development of free recall learning in children. Developmental Psychology, 4r 109-123. Corder, S. P. (1967). The significance of the learner's errors. IRAL, 4, 161-170. Cronbach, L. J. (1957). The two disciplines of scientific psychology. American Psychologist, 12, 671-684. Crookes, G. (1988). Planning, monitoring, and second language development: a review (Report No. 6). Manoa: Centre for Second Language Classroom Research, Social Science Research Institute, University of Hawaii. Damasio, A. R. (1985). The frontal lobes. In K. M. Heilman & E. Valenstein (Eds.), Clinical neurospsychology (pp. 339-375). New York: Oxford University Press. Das, J. P. (1984). Intelligence and information integration. In J. R. Kirby (Ed.), Cognitive strategies and educational performance (pp. 13-31). New York: Academic. Das, J. P., Cummins, J., Kirby, J., & Jarman, R. F. (1979). Simultaneous and successive processes, language and mental abilities. Canadian Psychological review, 2Q(1), 1-11. Das, J. P., Kirby, J., & Jarman, R. F. (1975). Simultaneous and successive syntheses: An alternative model for cognitive abilities. Psychology Bulletin, 82, 87-103. Das, J. P., Kirby, J., & Jarman, R. F. (1979). Simultaneous and successive cognitive processes. New-York: Academic Press. 187 Das, J. P., & Naglieri, J. (1989). Das-Naglieri Cognitive Assessment System, Administration manual, (Tryout edition). USA: The Psychological Corporation Harcourt Brace Jovanovich, Inc. Das, J. P., Naglieri, J., & Kirby, J. R. (1994). Assessment for cognitive processes. Boston: Allyn and Bacon. Day, E. M., & Shapson, S. M. (1987). Assessment of oral communicative skills in early French immersion programmes. Journal of Multicultural Development, 8. 237-260. Denney, N. W., & Ziobrowski, M. (1972). Developmental changes in clustering criteria. Journal of Experimental Child Psychology, 13, 275-282. Dixon, W. J. (Ed.). (1988). BMDP Statistical software manual. Los Angeles: The Regents of the University of California. Dulay, H., & Burt, M. K. (1973). Should we teach children syntax? Language Learning, 23^245-258. Dulay, H., & Burt, M. K. (1974). You can't learn without goofing: An analysis of children's second language 'errors'. In J. C. Richards (Ed.), Error analysis: Perspectives on second language acquisition (pp. 95-123). Essex: Longman Group. Dulay, H., Burt, M,. & Krashen, S. (1982). language Two. New York: Oxford University Press. Ellis, R. (1985). A variable competence model of second language acquisition. IRAL, 23, 47-59. Ellis, R. (1989). Understanding second language acquisition. Oxford: Oxford University Press. Entwistle, D. R. (1966). The word associations of young children. Baltimore: John Hopkins University Press. Entwistle, D. R., Forsyth, D. F., & Muus, R. (1964). The syntagmatic-paradigmatic shift in children's word associations. Journal of Verbal Learning and Verbal Behaviour, 3. 19-29. Ervin, S. M. (1961). Changes with age in the verbal determinants of word association. American Journal of Psychology, 74, 361-372. 188 Faerch, C , & Kasper, G. (1987). Perspectives on Language Transfer. Applied Linguistics, 111-136. Ferguson, G. A. (1971). Statistical analysis in psychology & education. New York: McGraw-Hill Book Company. Galton, F. (1883). Inquiries into human faculty and its development. London: Macmillan. Glanzer, M. (1982). Short-term memory. In C. R. Puff (Ed.), Handbook of research Methods in human memory and cognition (pp. 63-98). New York: Academic Press. Gorsuch, R. L. (1983). Factor analysis. London: Lawrence Erlbaum Associates, Publishers. Hammerly, H. (1982). Synthesis in second language teaching. North Burnaby, B.C.: Second Language Publications. Hammerly, H. (1989). French immersion: Myths and reality. Calgary: Detselig Enterprises Limited. Hayduk, L. A. (1987). Structural equation modelling with LISREL: Essentials and advances. Baltimore: The Johns Hopkins University Press. Hoelter, J. W. (1983). The analysis of covariance structures: Goodness-of-fit indices. Sociological Methods and Research, l l r 325-344. Hughlings-Jackson, J. (1879). On affections of speech from disease of the brain. Brain, 1T 304-330. Hughlings-Jackson, J. (1915). On the nature of the duality of the brain. Brain, 38, 80-95. [Reprinted from Medical Press and Circular, January 14.1874, vol. i. p. 19]. Hynd, G. W., & Willis, G. W. (1985). Neurological foundations of intelligence. In B. B. Wolman (Ed.), Handbook of intelligence: Theories, measurement and applications (pp. 119-157). New York: Wiley. Ianco-Worrall, A. (1972). Bilingualism and cognitive development. Child Development, 43_* 1390-1400. Jakobson, R. (1971). Studies on child language and aphasia. Paris: Mouton. Jarman, R. F. (1978). Level I and Level II abilities: Some theoretical reinterpretations. British Journal of Psychology. 69, 257-269. 189 Jarman, R. F. (1980a). Cognitive processes and syntactical structures: Analysis of paradigmatic and syntagmatic associations. Psychological Research, 41, 153-167. Jarman, R. F. (1980b). Modality-Specific information processing and intellectual ability. Intelligence, 4, 201-217. Jarman, R. F., Vavrik, J., & Walton, P.D. (in press). Metacognitive and frontal lobe processes: At the interface of cognitive psychology and neuropsychology. Genetic, Social, and General Psychology Monographs. Jenkins, J. J. (1954). Transitional organization: Association techniques. In C. E. Osgood & T. A. Sebeok (Eds.) Psycholinguistics: A survey of theory and research problems (pp. 112-118). Baltimore: Waverly. Jensen, A. R. (1970). Hierarchical Theories of Mental Ability. In W.B. Dockrell (Ed.), On intelligence (pp. 119-190). Toronto: Ontario, Institute for Studies in Education. Johnson, K. (1988). Mistake correction. ELT Journal, 42, 89-96. J"reskog, K. G., & S"rbom, D. (1989). LISREL VII: A guide to the program and applications. Chicago: SPSS Inc. Kerlinger, F. N. (1986). Foundations of behavioral research. New York: Holt, Rinehart and Winston. Kirby, J. (1984). Strategies and processes. In J.R. Kirby (Ed.), Cognitive strategies and educational performance (pp. 3-12). Orlando: Academic Press. Kirby, J. R., & Das, J. P. (1990). A cognitive approach to intelligence: Attention, Coding and Planning. Canadian Psychology. 31. 320-333. Kirby, J. R., Jarman, R. F., & Das, J. P. (1975). Paradigmatic and syntagmatic clustering: Relation to simultaneous and successive processing. Centre for the Study of Mental Retardation, University of Alberta, Edmonton, Alberta, Canada. Krashen, S. D. (1978). The Monitor Model for second language acquisition. In R.C. Gingras (Ed.), Second language acquisition and foreign language teaching, (pp. 1-26). Arlington VA: Centre for Applied Linguistics. Krashen, S. D. (1985). The input hypothesis: issues and implications. London: Longman. Lado, R. (1957). Linguistics across cultures: Applied linguistics for language teachers. Ann arbor, Michigan: University of Michigan. 190 Lai, C. (1990). UBC SPSS: Statistical package for social sciences - Extended version release 3.0 (Under MTS). Vancouver: University of British Columbia. Lamendella, J. T. (1977). General principles of neurofunctional organization and their manifestation in primary and nonprimary language acquisition. Language Learning, ZL 155-196. Lapkin, S. (1984). How Well Do Immersion Students Speak and Write French. The Canadian Modern Language Review, 40r 575-585. Luria, A. R. (1966). Human brain and psychological processes. New York: Harper & Row. Luria, A. R. (1970). Traumatic aphasia: Its syndromes, psychology and treatment. The Hague: Mouton. Luria, A. R. (1973a). Two basic kinds of aphasic disorders. Linguistics, 115, 57-66. Luria, A. R. (1973b). The working brain. London: Penguin. Luria, A. R. (1975a). Two kinds of disorders in the comprehension of grammatical constructions. Linguistics, 154, 47-56. Luria, A. R. (1975b). A note on the organization of fluent speech in a semantic kind of amnesic aphasia. Linguistics, 155, 57-60. Luria, A. R. (1976). A neuropsychological analysis of speech communication. In J. Prucha (Ed.), Soviet studies in languages & language behaviour, (pp. 191-207). Amsterdam: North-Holland Publishing Company. Luria, A. R. (1980). Higher cortical functions in man. New York: Basic Books. Lyster, R. (1987). Speaking immersion. The Canadian Modern Language Review, 43, 701-717. McLaughlin, B. (1987). Theories of second-language learning. London: Edward Arnold. Mefferd, R. B., JR., Dufilho, J.H., & Dawson, N. E. (1979). Word association and formal training of the written language: Consolidation of the response base. Psychological Reports. 45, 615-728. McDonald, R. P. (1985). Factor analysis and related methods. Hillsdale, NJ: Laurence Erlbaum Associates. 191 Murphy, M. D., & Puff, C. R. (1982). Free recall: Basic methodology and analyses. In C. R. Puff (Ed.), Handbook of research methods in human memory and cognition (pp. 99-128). New York: Academic Press. Naglieri, J. A. (1988). A cognitive processing theory for the measurement of intelligence. Educational Psychologist, 24r 185-206. Naglieri, J. A., & Das, J. P. (1988). Planning-Arousal-Simultaneous-Successive (PASS): A model for assessment. Journal of School Psychology, 26, 35-48. Naglieri, J. A., Das, J. P., Stevens, J. L, & Ledbetter, M. F. (1991). Confirmatory factor analysis of planning, attention, simultaneous, and successive cognitive processing tasks. Journal of School Psychology, 29, 1-28. Nelson, T. O. & Narens, L. (1990). Metamemory: A theoretical framework and new findings. In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory: Vol 26 (pp. 125-173). New York: Academic Press. Nemni, M. (1985). Si communication savait...Si grammaire pouvait.... The Canadian Modern Language Review, 42, 288-306. Nielsen, B. (1983). Validation of the WISC-R for grade two French immersion students. Unpublished master's thesis, University of British Columbia, Vancouver. Ochs, E. (1979). Planned and unplanned discourse. In T. Givon (Ed.), Syntax and Semantics, 12, 51-80. Odlin, T. (1989). Language transfer. Cambridge: Cambridge University Press. Ontario Institute for Studies in Education. (1983). Saskatchewan project, French language speaking and writing evaluation units. Toronto: Modern Language Centre, Ontario Institute for Studies in Education. Palermo, D. S. (1971). Characteristics of word association responses obtained from children in grades one through four. Developmental Psychology, 5, 118-123. Palermo, D. S., & Jenkins, J. J. (1963). Frequency of superordinate responses to a word association test as a function of age. Journal of Verbal Learning and Verbal Behaviour, 1, 378-383. Pellerin, M., & Hammerly, H. (1986). L'expression orale aprSs treize ans d'immersion fran^aise. The Canadian Modern Language Review, 42, 592-606. 192 Piaget, J. (1929). The child's conception of the world. London: Routledge & Kegan Paul. Pons, L. & L'homme, M.F. (1983). Paradigmatism and syntagmatism of word responses of normals and patients. Psychological Reports, 52, 412-414. Pribram, K.H. (1971). Languages of the brain: Experimental paradoxes and principles in neuropsychology. Englewood Cliffs, N. J.: Prentice Hall. Province of British Columbia, Ministry of Education. (1990). Year 2000: a framework for learning. Richards, J. C. (1973). Error analysis in a second language environment. In T. W. Oiler & J. Richards (Eds.), Focus on the learner, (pp. 114-135). Rowley, Massachussets: Newbury House Publishers Inc. Segalowitz, N. (1986). Skilled reading in the second language. In J. Vaid (Ed.), Language processing in bilinguals: Psycholinguistic and neuropsychological perspectives, (pp. 3-19). Hillsdale: Lawrence Erlbaum Associates. Selinker, L. (1969). Language transfer. General Linguistics, 9, 67-92. Selinker, L. (1972). Interlanguage. IRAL, 10, 219-231. Smith, H. (1936). The problem of comparing the results of two experiments with unequal means. Journal of Council Scientific and Industrial Research, 9, 211-212. Spilka, I. (1976). Assessment of second language performance in immersion programs. The Canadian Modern Language Review, 32, 543-561. Steiger, J. (1987). How to use Multicorr (Version 2.4). University of British Columbia: unpublished manuscript. Sternberg, R. J. (1977). Intelligence, information processing, and analogical reasoning: The componential analysis of human abilities. Hillsdale, NJ: Erlbaum. Sternberg, R. J. (1979). The nature of mental abilities. American Psychologist, 34, 214-230. Taylor, B. P. (1975). Adult language learning strategies and their pedagogical implications. TESOL Quarterly, 9, 391-399. The Psychological Corporation. (1991). WISC-III, Picture arrangement. New York: Harcourt Brace Jovanovich, Inc. 193 Tomasello, M., & Herron C. (1989). Feedback for language transfer errors, the Garden Path Technique. Studies in Second Language Acquisition, 1 l f 385-395. Tomlinson, J., L. (1988). Language processing abilities of reading disabled, low achieving, and normal readers. Unpublished Ph.D. thesis. University of Alberta, Edmonton. Trites, M. (1986). Bilingualism and reasoning ability. Unpublished master's thesis, University of British Columbia, Vancouver. Vocate, D. R. (1990). Luria on language and mind. Language and Communication, 10. 267-284. Vygotsky, L. (1934). Thought and speech. Moscow: Sotsekgiz. Vygotsky, L. (1962). Thought and language. Cambridge, MA: M.I.T. Press. Vygotsky, L. (1965). Psychology and localization of functions. Neurospychologia, 3, 387-392. Vygotsky, L. (1978). Mind in society: The development of higher psychological processes. Cambridge, M.A.: Harvard University Press. White, S. (1965). Evidence for a hierarchical arrangement of learning processes. In L. P. Lipsitt & C. C. Spiker (Eds.), Advances in child development and behaviour, vol 2 (pp. 219-220). New York: Academic Press. Zangwill, O. L. (1966). Psychological deficits associated with frontal lobe lesions. International Journal of Neurology, 5, 395-402. Zar, J.H. (1984). Biostatistical analysis . Englewood, N. J. : Prentice Hall, Inc 194 Appendices 195 A 1. Visual Search 196 SAMPLE A 196A A 2. Planned Codes 197 X o B O O c X X D O X B D B C D D B D B D B D B B D B D B D B D D A B D B D B D B I T E M 2 197A A 3. Planned Connections 198 ITEM 10 198A A 4. Simultaneous Verbal 199 199A A 5. Figure Memory 200 200A i ITEM 13 200B A 6. Matrices 201 A 7. Word Series 202 Hord Series A d m i n i s t e r » t r « t . o f 1 word p . r s econd D i i c o n t i n u . » f t « r -i c o n i . c u t i v . f . i l u r . i ( e x c l u d i n g j w i o l e i ) U » . » c h e c k M t o I n d i c a t e . c o r r e c t w o r d . R . c o r d .n i n c o r r . c t - o r d I n t h . o o « b e l o w t h . c o r r e c t w o r d . U i * » d a . h (-> t o I n d i c t , c m K H o n i t t t h . .nd .of t h . i . n . i . I f - o r d S a r . » d d . d t o t h . . n d o f t h . s e r i e i . r e c o r d t h e " t o t h . r i c h t o f t h . l i l t w o r d . S c o r e e i c h I t e m » J P » i « O ) ° r F » i l ( 0 ) &2 E°i t c ° r t iirar I Y P ° * Score (1.0) Error Type e: BooK ! Cir i Ojajijcrv:.^ .: i: ' jaT:;:~« K P V 1 Shoe I mi —— tr^rli*. - -T Car 1 Wall 1 mm Hon 1 G1rl BooK mm Han I Cow 1 Kev . Shoe i r pnq | Man ! | Book 1 i Wall 1 fnv 1 Car Girl BooK Wall i Kev ! Ooa : i : 1 Girl i i • f a r 1 Shoe ' Cow J I Han Wall 1 Car i Shoe Cow Kev SooK | Poq G1rl ! mm Shoe Car- Doa 1 Man K.ev mm fcrr=> -..-1 Ha l l Man Book. Girl i Cow BOOK Shoe Girl Ooa Car Cow Kev Man Wall Dog BooK Cow Car Wall Man Girl Kev Shoe Doa Wall Book £j;|Jir.Jii Man G i r l Shoe Car COK Ha l l noq i Kev 1 Man Cow Shoe .. Gin 202A A 8. Sentence Repetition 203 The white Ms IbluT 2 3 4.1 5 64 7. 8. The I vellow Igreened I the Iblue. J L The minks |are I veil owl no [the I tans The jblue [whitesjthe aravjthe!red, Thelaurole lis Iblack1ng Ito I tan the Iqrav The ivellow land iqreen ihrown 'theIpurple. I I I I ! ' Theired Iblued | theIqreeniwith ia|vel low The'purple |vellows lare green and jtheIrpqs iareIwhlte. 9,j Thelred who I blues I vel 1 owl browned I on I theigreen JL 10 11 12 13 14 IS . 16 17 Thelpurp1e|bluediat J_ the i greeni when lite gray* vel lowed la 11 the ; p1 nk .1 Sreenireds jthe ;blueland vellows Itheibrown I to white,I Red POt hiuesI to I pink (theiqreenIpurplel but I I the brown purple The red. greened that Itannerl jr'hp|blue 'to ithe IblackT before;p1nk yellowed l gray.| Brown Ibiues|green, j i n k WJTC. reds Ii n fthi TheIbrown [thatjoraved Ithe (blue green blues 1M purple jwhite whitel pjnked ItheIqreen jto| black ja The |purple'blacked Ithe tanIgreened red 1 qraviwithi|a oink when! vellcwI to brown TheIblue j reds |a arelpurple ore en|ve11owloflpinks that! In jthe |brown then qravs jthe [tan. 203A A 9. Sentence Questions 204 Correct Response Score (1.0) 9. Green reds the blue and yellows and the reds are white. What does green do to the blue? reds thera 10. The purple yellows are green and the rads are white. Who are green? the purple yellows 11. The purple blued at the green when the gray yellowed at the pink. Who was blued at? the green 12. The red that tanned the blue to the black greened before pink yellowed gray. What was tanned? the blue 13. Red blues to pink the green purple but not the brown purple. Which purple does red blue? the green one (or green purple) 14. Brown blues green, but green blues the pink who reds in the purple white. Who does green blue? the pink 15. The purple blacked the gray with a pink when the tan greened the yellow to brown. When did the purple black the gray? when the tan greened the yellow to brown 16. The brown that grayed the blue white pinked the green to black a red. Whv did the brown gray the blue white? to black a red 204A A 10. Expressive Attention 205 h-YELLOW BLUE BLUE GREEN YELLOW BLUE YELLOW GREEN RED YELLOW GREEN RED RED YELLOW GREEN RED YELLOW RED BLUE YELLOW BLUE YELLOW GREEN BLUE RED YELLOW GREEN GREEN RED YELLOW 205A A 11. Receptive Attention 206 " R f i) i i A o P. (i A r R A n A a >i E b N r R e T Tb t T E N b T e E n e r t R r t a T r B R .5 E N E B b T e R n R b n r a T N n a E A a ; r N e e E b r e r A r t n t T n a r B n N a r r T b t r a E e t e r ,M n B a R N t Tt E a t B r b e t B b T n T t n R E e T E N b b B a A r n B r A e R E r A. N r a A B e T N A B t E N n B n E r a b r R B A t T e b A R E n b n a B e E a N A E N T b R e N t R N n E e T a N B R r A t t N E R n N e a t T E T b B R T t b A b B N B b r e A a n t a t B E R a A T e n N a B t a n n T E A T A a e n N R N B a A a N A r E R r b e R r A N T B E b B T R b b a t T R t t a a A b A R e n E E e e A B b b E N E T e R n T t R b n r a T a E t r n N e E N e b r e r A r r R t n T t B e n a r B E t a r r T bt r a t e r N b B n B a R N t N n E a e B a A t B r b e t N R I T E M 6 206A A 12. Number Finding 207 Find the numbers that look like this: 1 2 3 4 5 6 4 3 1 5 1 4 2 2 5 1 3 . 4 1 4 2 6 3 6 4 5 3 5 3 6 2 4 1 6 6 2 3 6 1 • ^ 3 5 1 5 5 4 6 3 1 4 5 6 2 2 1 4 6 3 4 5 3 2 3 4 6 1 2 5 2 1 6 4 3 6 5 3 5 6 6 6 4 5 3 2 6 1 2 4 3 1 1 4 6 2 3 4 6 5 2 3 1 5 3 1 4 3 5 2 6 2 1 3 5 6 2 5 2 1 1 6 5 6 3 4 4 2 4 1 6 4 2 5 3 4 1 2 6 5 3 3 5 S 6 4 1 2 6 4 2 2 6 1 4 2 4 3 5 1 6 3 5 4 5 4 3 2 6 • 5 '1 4 2 5 1 6 2 1 5 1 3 4 2 5 3 1 4 4 ITEM 4 207A B 1. Closed Word Association Task 208 CLOSED ASSOCIATION TASK PUPIL'S NUMBER: DATE OF BIRTH: DATE OF TEST: CHRONOLOGICAL AGE: SCHOOL: M OR F PARADIGMATIC LIST SYNTAGMATIC LIST 1. hot - waxm/fire 2. high - up/sky 3. dark - black/night 4. tell - secret/talk 1. wicked - witch/bad 5. him 6. run he/her fast/walk 7. harder - softer/stone 8. desk - write/table 9. back - behind/front 10. sheep - woolly/animal 2. cheat lie/bad 3. grass - trees/green 4. music - sing/song 5. float - light/sink 6. night - dark/day 7. eating - supper/drinking 8. worked - played/hard 9. soft - pillow/hard 10. bee - sting/insect TOTAL PARADIGMATIC: TOTAL SYNTAGMATIC: 208A B 2. Free Word Association Task 209 FREE WORD ASSOCIATION TASK PUPIL'S NUMBER: DATE OF BIRTH: DATE OF TEST: CHRONOLOGICAL AGE: SCHOOL: M OR F PARADIGMATIC LIST SYNTAGMATIC LIST 1. carry 2. bird 3. salt 4. clean 5. slow 6. galop 7. alow 8. black 9. gently 10. moth 1. move 2. table 3. music 4. yellow 5. sour 6. obey 7. listen 8. pretty 9. lovely 10. needle TOTAL PARADIGMATIC: TOTAL SYNTAGMATIC: 209A Appendix C. Written Language Tasks French Language Speaking and Writing Evaluation Units 210 1. Pendanc quelle saison se passe cette hiscoire? Cecce h i s t o i r e . . . 2. Ou se crouve la cage de Nicou? Pourquoi Marc v e u t - i l rentrer chez sa tance? •4. Ou Nicou se cache-c-il? 5. Quel animal ainerais-tu avoir le plus chez coi? Pourquoi? 210A Marc trouveunami Marc esc assis au fond de la salle de classe avec un l±vre de lecture. II trouve l a premiere histoire tellemenc passionnante qu'il la l i t du debut jusqu'a La f i n . Quand 1 1 a f i n i de l i r e , i l commence a regarder autour de l u i . Juste a. cote de son pupitre, 1 1 volt une cage. A l'interieur i l 7 a une petite boule de fourrure. C'est un cochon d'Inde. Marc s'approcfae de l a cage. ' regarde 1*animal et i l veut ^ toucher. II fait un petit bruit sa bouche et 1'animal le . <Viens, Nicou, dit Marc, viens me .3) Nicou s'approche du doigt du . <Qu'il est mignon,$> pease Marc. 3 e demande s i Nicou a faim. • de la cage, i l y a assiette avec quelques feuilles de laitue. les prend pour les doimer a. animal. II glisse toutes les feuilles • la cage. On voit que Nicou bien la salade parce qu'il les feuilles de laitue, l'une 1'autre. <^J'avals raison, pease Marc, avait tres faim.^ > Mais i l ne ._ plus de salade. Marc regarde autour [ l a cage pour voir s ' i l . ; 7 a P a s d'autre nourriture a donner. Tout a. coup i l entend nom. C'est le professeur qui l'appelle. «$Marc, viens nous aider. Nous commencons la lecou de mathematiques et nous avons besoin de t o i . ^ Marc ref erme tres vite • la porte de l a cage et s'en va au tableau. II n'a meme pas le temps de dire au revoir a Nicou. 210B 210C Appendix D. Oral Language Task Subtest "Picture Arrangement" Of the Wise m Intelligence Test Example of One Item In the Order Presented to the Pupils 211 211A E 1. Letter of Contact to the Parents 212 T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A Department of Educational Psychology and Special Education Faculty of Education 2125 Main Mall Vancouver, B.C. Canada V6T 1Z4 Tel: (604) 822-8229 Fax: (604) 822-3302 Dear Parent or Guardian: I am writing to you to request your permission for your son or daughter to participate in a research project I am conducting in his/her school entitled "Relationships between cognitive and linguistic processes and second language production in French immersion programmes." The purpose of this study is to discover what is the relationship between the ways French immersion pupils process information, particularly linguistic information and the errors they make when speaking French. This study should provide a theoretical basis for devising a more efficient pedagogy to teach second language in immersion setting so that pupils speak more accurate French when coming out of French immersion programmes. The project requires the cooperation of 180 grade 3 French immersion pupils to take tests which are widely used. Some of the tests are group tests and will be given to the entire classroom and others will be individual tests taken outside of the c lassroom. The entire set of tests will take approximately 3 hours. Participation in the study is entirely voluntary and your son or daughter may withdraw from the research study at anytime without penalty. Withdrawal from the research study or refusal to participate will not jeopardize class standing in any way. Pupils who do not participate will be given an activity related to the topic being covered in their class. Al l information collected will be strictly confidential and will not be available to pupils ' parents, teachers, or other school personnel. Pupils ' names will not be placed on the answer sheets to ensure anonymity. Needless to say, I would be extremely pleased if your son/daughter does decide to participate and, if you are willing, to give him/her permission to do so. If you have any questions and wish to further discuss this project, feel free to call me at (604) 228 -1880 or my supervisor, Dr. Ron Jarman at (604) 822 -6296. Please keep a copy of this request and all attachments for your records. I would appreciate it if you would indicate on the slip provided on the attached page whether or not your son/daughter has permission to participate. Would you then kindly sign and date the slip and have your son/daughter return it to school tomorrow? Furthermore, if you agree for your son/daughter to participate, would you please complete the Background Questionnaire attached and send it back to school along with the permission slip. Thank you very much for considering this request. Sincerely, 212A Monique Bournot-Trites, M.A. Page 1 of 3 E 2. Parent Consent Form 213 E 3. Background Information Questionnaire 214 BACKGROUND INFORMATION QUESTIONNAIRE Your assistance in providing the following information would be very helpful in making this a meaningful study: 1. What is the date of birth of your child? 2. What is the main language spoken in your home? 3. Do you sometimes use another language in the home? 4. Has your child enrolled in the French immersion program continuously from Kindergarten through to the present time? Yes No (check one) If not, when did your child start French immersion? 5. Has your child any learning difficulties now or has she/he had any in the past ? Yes No (check one) If yes, specify: If yes, did she/he require learning assistance or special help of other kind (specify) Page 3 of 3 214A Appendix F. Repartition of Errors in Oral and Written French Language Production. (Percentage of each Category of Errors relative to the Total Number of Errors, of Transfer Errors, of Oral Errors, and of Written Errors) Errors Total Transfer Oral Written (5984) (963) (2369) (3615) Total number % % % % A L L ERRORS (5984) 100 -- Oral (2369) 39.59 - 100 -- Written (3615) 60.41 - - 100 - Grammar (3704) 61.90 - - -. Oral (1334) 22.29 - 56.31 -. Written (2370) 39.61 - - 65.56 - Vocabulary (2280) 38.10 - - -. Oral (1035) 17.30 - 43.70 -. Written (1245) 20.81 _ _ 34.44 A L L ERRORS MINUS TRANSFER (5021) 83.91 - - -- Oral (1823) 30.46 - 76.95 --Written (3198) 53.44 - - 88.46 - Grammar (3493) 58.37 - - -. Oral (1225) 20.47 - 51.71 -. Written (2268) 37.90 - - 62.74 - Vocabulary (1528) 25.53 - - -. Oral (598) 9.99 - 25.24 -. Written (930^ 15.54 _ 25.73 TRANSFER ERRORS (963) 16.09 100 - -- Oral (546) 9.12 56.70 23.04 --Written (417) 6.96 43.30 - 11.54 - Grammar (211) 3.53 21.92 - -. Oral (109) 1.82 11.32 4.6 . Written (102) 1.70 10.59 - 2.82 - Vocabulary (752) 12.57 78.08 - -. Oral (437) 7.30 45.38 18.45 -. Written (315) 5.26 32.71 _ 8.71 Numbers in parentheses are raw. Percentages have been calculated only when appropriate. 215 Appendix G. Correlation Matrix of All the Data (Calculated From Raw Data) 216 a t e o > o I 3 81 ST 85 o E § g o u a w 5 § 5 S 2 § § S S 8 a s H § s 8 2 S 2 ^ ^ 5 r "! ^ ^ ^ 1 9 —< • * i t i i i § 8 3 M 2 S 3 § ? ~+ * i ( i ** i * i* ' ; M M § I i 8 M H — • I I I { % { ' { ' ^ I I I # * t * i* i* i* * § q q 1 q " q q S 5 - - « o 8 9 o o o § — • • ' •' ' ' i ' i ' " ' i* r • - , .. I » • i * i i i i § S S 3 n § i § § : S : H « ? ^ § s 3 § 3 s s U § 3 ! § § S § ? § U M § ? n § § § 3 i i s s n H i — , • • > i • • * • * * • " * «" • i t i i i * i i i i % § § § % § s § § % i i i i. i § i § 5 i § g 5 i § i *"* • • > i • r •' «* * i* *~* < i i # i i t* i* * • • * • " " " • • * < " * U M M H M U 8 S a 8 8 S M = S S & S •3 Q | 1 -S 3 5 5 5 5 5 6? Si U B? § 2; 1 I l l It ! S E E J S , | * It II II || 12 „ 11 * » g y a » J. § Appendix H . Examples of classification of language Errors 1. Grammar Errors • All prepositions • Gender errors if agreement is incorrect • "II sont ven," : 2 grammar errors (incorrect agreement of pronoun and wrong past participle. • "lis vient": 2 grammar errors (incorrect agreement of pronoun and incorrect agreement of verb. • In a sentence in the past tense, each wrong tense counts for one error. • All incorrect agreements • Wrong conjugation, wrong tense or mode • Wrong pronoun 2. Vocabulary Errors •All gender errors if agreement is correct with the chosen gender •Pronominal verbs without reflexive pronoun • Wrong word which is not an English word or missing word 3. Transfer Errors Transfer Grammar • "Elle a donn," : wrong place of the pronoun • "Qu'est-ce que" instead of "ce que" • "Ses pieds" instead of "les pieds" ^Transfer Vocabulary •English word ("le bicycle") •English verb conjugated like French verbs ("checker") •direct translation of word ("les hommes de feu" for "pompiers") 217 Errors not counted • Inverted letters: Example: "vior" instead of "voir", or "peau" for "beau" • Wrong spelling ("mons" for "raon") except if the word means something else. In that case it would be a vocabulary error ("en" for "on"). • Wrong spelling ("je vai" for "je vais") except if the spelling represent another grammatical form ("ont" for "on"). • Accents, elisions, capitalization • style ("toutes les choses sont pas 1..." instead of "rien n'est 1..." 218 


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items