BILINGUAL INFORMATION PROCESSING: THE EFFECTS OF COMMUNICATION‘ON SEMANTIC STRUCTURE ' Dissertation for the Degree of Ph. D. . MICHIGAN STATE UNIVERSITY GEORGE A. BARNETT 1976 ._ IIIIIIIIIII This is to certify that the thesis entitled Bilingual Information Processing: The Effects Of Communication on Semantic Structure presented by George Aaron Barnett has been accepted towards fulfillment of the requirements for Ph. D. degree in Communication Q Major mores»: Date 3/22/76 0-7639 I I. - ._ o . “QMJJOL 4' OVERDUE FINES ARE 25¢ PER DAY PER ITEM Return to book drop to remove this chec at from your record. . II @633 W”? L33! @1237}? w. . 33 3:53 ‘ W ABSTRACT BILINGUAL INFORMATION PROCESSING: THE EFFECTS OF COMMUNICATION ON SEMANTIC STRUCTURE by GEORGE AARON BARNETT Based upon social scientists and philosophers. ‘writings about the notion of meaning, a paradigm for the measurement of meaning is established. The four basic principles of the paradigm are that the meaning of any lexical item can only be determined through its relation with other symbols. The most important of these relations is semantic similarity. The other principles are that the measured meanings must be consensual, dynamic and empirically fruitful. Next, the existing methods for the measurement of meaning are critiqued. An argument is made that free association, free recall, Osgood's semantic differential and non-metric multidimensional scaling are inadequate because each fails to meet the established principles. However, a modified version of classical metric multidimensional scaling may be adopted as the paradigm for the measurement of meaning. Next, a single substantive problem, the processing of information by bilingual individuals is discussed. This problem is of great social significance because of the bilingual's role in the transfer of information across linguistic boundaries. Bilinguals George Aaron Barnett have been studied by psycholinguists concerned with how these people process linguistic information so that they :may gain insights into how all people process language. By and large, psychologists have not been overly successful. One of the reasons sociolinguists point out is that the bilingual is not viewed as a member of a speech community. The author concludes that both perspectives must be employed in order to describe the communication behavior of the bilingual. Based upon these two lines of research and the presented paradigm, the author suggests and tests eight hypotheses dealing with bilingual information processing. Canadian college students either monolingual or bilingual in French and English made up the sample. Metric M.D.S. was used to measure semantic structure. The results indicated that the semantic properties of lexical items rather than the symbol's language are the primary organizing principles of a bilingual's semantic structure, although the language attribute was of secondary significance; that the semantic space of English and French monolinguals as well as bilinguals in either language are all significantly different.but can be described by the same number of underlying dimensions. The degree of discrepancy between a space produced by a group of bilinguals and a group of monolinguals will be less than the difference between the two monolingual groups was partially supported. George Aaron Barnett The mass media usage patterns of the three groups were significantly different. The English-language group used the mass media significantly more than the French- language group with the exception of television, where the pattern was reversed. The bilinguals total number of media hours exceeds both monolingual groups. Bilinguals used interpersonal channels to a significantly greater degree than monolinguals. Finally, the author was unable to predict aspects of the bilinguals' semantic system from their exposure to the media and interpersonal relations in each of their languages. The results are interpreted in light of the bilingual's potential embeddedness in social networks, the concept of language shift and the theory of bilingualism presented earlier. BILINGUAL INFORMATION PROCESSING: THE EFFECTS OF COMMUNICATION ON SEMANTIC STRUCTURE by 0‘0“ George A. Barnett A Dissertation Submitted to the Michigan State University in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Communication 1976 (9 Copyright by GEORGE AARON BARNETT 1976 ii Accepted by the faculty of the Department of Communication, College of Communication Arts, Michigan State University, in partial fulfillment of the require- ments for the Doctor of Philosophy Degree. QEsgctor of Eyesis' iii To my father, JaCob B. Barnett iv ACKNOWLEDGEMENTS The process of generating scientific knowledge is never carried out by individuals acting in isolation. Explicit in the scientific method is the practice of build- ing upon existing knowledge. Often this involves the synthesis of ideas from many seemingly unrelated disci- plines and lines of research. It is the author's belief that this dissertation is part of a continuing product of a group of scholars all intimately involved in the persuit of knowledge. The author would like to thank a number of individuals for their ideas and feedback to mine. Among them are Donald P. Cushman, for his assistance in the development of the paradigm for the measurement of meaning; James A. Taylor, Kim B. Serota, Rolf T. Wigand, Timothy Maybee, Michael Cody and Franklin J. Boster for their assistance at all stages of the research including the reduction of task related stress. A special note of thanks is due to Thomas McPhail without whom this dissertation would still only be a dream. The author would like to thank his advisory committee members, Richard V. Farace, Daniel Beasley and Jack M. Bain for their guidance throughout the completion of my degree, and committee member Edward L. Fink for his continuous demand for rigor in my intellectual persuits. Finally, a special acknowledgement of gratitude to Joseph Woelfel, my committee chairman and close friend. It was Joe who oversaw my socialization into the role of scientist. He allowed me the maximum academic freedom to persue my intellectual interests and enough direction to keep me in line when my mind wondered to trivia. He was constantly encouraging me and getting me up, when I was down. vi CHAPTER I CHAPTER II CHAPTER III TABLE OF CONTENTS Page OVERVIEW. 0 O O O O O O O O . O O O O O O l A PARADIGM FOR THE MEASUREMENT OF MEANING o o o o o o o o o o o o o o o o 5 Introduction. . . . . . . . . . . . . . 5 The Fundamental Measurement of Meaning . . . . . . . . . . . . . . . 10 The Relational Nature of Meaning. . . . 13 Associationism . . . . . . . . . . l3 Similarity . . . . . . . . . . . . 15 Contiguity . . . . . . . . . . . . 17 Contrast . . . . . . . . . . . . . 17 American Associations. . . . . . . 18 Formal Logicians . . . . . . . . . . 22 Communication. . . . . . . . . . . . 28 The Consensual Nature of Meaning. . . . 31 Philosophy . . . . . . . . . . . . . 32 Sociology. . . . . . . . . . . . . . 33 Linguistics. . . . . . . . . . . . . 35 The Dynamic Nature of Meaning . . . . . 33 Empirical Fruitfulness. . . . . . . . . H2 The Measurement of Meaning. . . . . . N7 Free Association . . . . . . . . . . #7 Free Recall. . . . . . . . . . . . . 53 'The Semantic Differential. . . . . . 59 Multidimensional Scaling (M.D.S.) . . . . . . . . . . . . . . 66 Nonmetric. . . . . . . . . . . 7o Metric . . . . . . . . . . . . . . 7n Summary . . . . . . . . . . . . . . . . 80 Footnotes for Chapter II. . . . . . . . 32 BILINGUAL INFORMATION PROCESSING - THE EFFECTS OF COMMUNICATION ON SEMANTIC STRUCTURE. . . . . . . . . . . 8” Overview. . . . . . . . . . . . . . . an Social Significance of the Bilingual. . 85 Past Research . . . . . . . . . . . . . 90 Psycholinguistics. . . . . . . . . . 91 Sociolinguistics . . . . . . . . . . 103 vii CHAPTER IV CHAPTER V CHAPTER VI APPENDIX Theoretical Hypotheses. . . . . Summary . . . . . . . . . . . Footnotes for Chapter III . . . METHODS O O O O O O O I O O 0 Overview. . . . . . . Operationalization. . . . . . The Setting . . . . . . . . . . Mass Media . . . . . . . . . Instrumentation . . Translation . . . . . . . . Design. . . . . . . . . . . Subjects. . . . . . . . . . . Summary . . . . . . . . . . . Footnotes for Chapter IV. . RESULTS . . . Overview. . . . . Hypothesis One. . . Hypothesis Two. . Hypothesis Three. Hypothesis Four . Hypothesis Five . Hypothesis Six. . Hypothesis Seven. . . . . . . Hypothesis Eight. . . . . . . . Summary . . . . . . . . . . Footnotes for Chapter V . . . . DISCUSSION AND SUMMARY. . . . . Overview. . . . . . . . . . . . The Theory. . . . . . . . . . . Critique of Methods and Future Research. . . . . . . . . . . . Summary . . . . . . . . . . . . INSTRUMENTATION . . . . . . . . English Language Questionnaire. French Language Questionnaire . Mixed Language Version of the Questionnaire . . . . . . . . . REFERENCES. 0 o o o o o o o o o .- 0 viii 118 118 118 12% 125 127 129 131 135 137 138 1H0 1H0 1ND 1H7 1M8 156 15k 172 173 179 182 18” 187 187 187 195 205 207 208 217 227 230 Table 1. Table 2. Table 3. Table A. Table 5. Table 6. Table 7. Table 8. Table 9. Table 10. Table 11. Table 12. Table 13. Table 1a. LIST OF TABLES The Available Mass Media in Ottawa-Hull . . Mean Distance Matrix for Bilinguals in a Mixed Language Condition. . . . . . . . . Spatial Coordinates for Bilinguals in a Mixed Language Condition. . . . . . . . . . Coorelations of Individual Dimensions with Language Vector . . . . . . . . . . . . . . Mean Distance Matrix for Monolingual EngliSh O O O I O O O O O O O O O O O O O 0 Mean Distance Matrix for Monolingual Frencn . O O O O C O O O O I O O O O O O O 0 Spatial Coordinates for Monolingual EngliSh O O O O O O O O O O O O O O O O O 0 Spatial Coordinates for Monolingual PrenCh . C C O I O O I O O O O O O I O O O 0 Differences Between the French and English Monolinguals' Spaces After Least-Square Rotation on All Ten Dimensions. . . . . . . Mean Distance Matrix for Bilinguals in EngliSh O O O O I O O O O O O O O O O 0 O 0 Mean Distance Matrix for Bilinguals in FrenCh O . O O O O O O O O O O O O O O O O O 0 Spatial Coordinates for Bilinguals in EngliSh O O C C I O I O O O O O O O O O 0 Spatial Coordinates for Bilinguals in FrenCh O O O I O O O C O O O O O O O O O O 0 Differences Between the Bilinguals' English and French Spaces After Least-Squares Rotation on all Ten Dimensions. . . . . . . ix 126 1U2 1H3 146 1H9 150 151 152 155 158 159 160 161 165 Table Table Table Table Table Table Table Table 15. 16. 17. 18. 19. 20. 21. 22. Differences Between the English Monolingual and the Bilinguals' English Space After Least-Square Rotation on all Ten Dimensions Differences Between the English Monolingual and the Bilinguals' French Spaces After Least-Square Rotation on All Ten Dimensions Differences Between the French Monolingual and the Bilinguals' English Space After Least-Square Rotation on all Ten Dimensions Differences Between the French Monolingual and the Bilinguals' French Space After Least-Square Rotation on All Ten Dimensions t-tests Performed to Test Hypothesis Five . Matrix of Discrepancies Between all Spaces. . . . . . . . . . . . . . . . . . . Media Use Patterns of Bilinguals, English and French Monolinguals . . . . . . . . . . Comparison of Interpersonal Communication English Monolinguals, Bilinguals and French Monolinguals . . . . . . . . . . . . 167 168 169 170 171 17” 178 180 Figure Figure Figure Figure Figure Figure Figure Figure LIST OF FIGURES Symbolic Account of the Development Sign Processes. . . . . . . . . . . The Compound-Coordinate Distinction Bilingual Mixed-Language Space. . . Graphic Representation of English Monolingual Space . . . . . . . . . Graphic Representation of French Monolingual Space . . . . . . . . . Graphic Representation of English Bilingual Space . . . . . . . . . Graphic Representation of French Bilingual Space . . . . . . . . . . Graphic Representation of Combined Spaces of All Groups . . . . . . . . . . . xi 23 - 100 luv. 153 15” 162 163 ° 175 CHAPTER ONE OVERVIEW The question of how people process information has been a concern of scholars since antiquity. Aristotle wrote that three basic principles —- similarity, contrast and contiguity -- govern reasoning. In the seventeenth century, John Locke postulated the notion that the human mind begins as a blank slate and that all people process ideas based upon their past experiences. In the 19th century, the German psychophysicists, Fechner, Weber and Hemholtz began to apply rigorous laboratory princi- ples to the problem of human sensation and perception. In the 20th century, the focus of the problem shifted away from the human mind and attention was brought to bare on why people behave the way they do. The work of the behaviorists, Watson and Skinner, explicitly ignored the organization of the mind and its impact upon behavior. It is with the work of the early psycholinguists in the 19503, Osgood, Deese and others, that the notion of meaning as a theoretical concept that can have behavioral consequences reappears in psychology. Perhaps the key implication of their work was that meanings can be communicated through linguistic messages and that these meanings can motivate future behavior and the interpreta- tion of future messages. At about the same time, psychometricians (Torgerson, Tucker and others) began to . 1 2 systematically apply Euclidean distance models to the perception of stimuli: colors, sounds, and verbal symbols. It is the intent of this dissertation to build upon the ideas of these individuals. Chapter two establishes a paradigm for the measurement of meaning. The four basic principles of this paradigm are that meanings can only be measured relation- ally, that is, the meaning of any given lexical item can only be determined through its relation with other symbols. The most important of these relations is semantic similar- ity. The other principles are that the measured meanings must be consensual. The measurement system must be dynamic. And, finally, it must be empirically fruitful. Chapter two continues by critiquing the existing methods for the measurement of meaning. An argument is made that free association, free recall and Osgood's semantic differential scale are inadequate because each fails to meet the principles established in the first section of the chapter. Next, the theory of multidimensional scaling (MDS) is presented. However, because of psycho- metricians concern with the individual case, and the methods they employed to compensate for this problem, one form of MDS, non-metric, is rejected also. Chapter two concludes by suggesting that a modified version of classical metric multidimensional scaling be adopted as the paradigm for the measurement of meaning. This scaling procedure would have the advantage of meeting the criteria established early in the chapter. The next chapter (three) discusses a single substantive problem, the processing of information by bilingual individuals. This problem is of great social significance because of the need to transfer certain in- formation across linguistic boundaries. The only way that this information can pass from one language group to an- other is through a bilingual. The problem is of critical importance where national development and social integra- tion are concerned. The bilingual individual has been the object of intensive investigation by the social sciences. He/she has been observed by psycholinguists who are concerned with how these people process linguistic information so that they may gain insights into how all people process language. By and large, the psychologists have not been overly successful at describing the bilingual's linguistic system. One of the reasons sociolinguists point out is that the bilingual is not viewed as a member of a speech community. They raise the general question of how the environment affects the way bilinguals use language. The author concludes that both perspectives must be brought into play if we are going to describe the communication patterns of the bilingual. Based upon these two lines of research, and ideas from contemporary communication theory, chapter three concludes with eight theoretical hypotheses, which constitute the basis for a theory of bilingual information processing. u Chapter four takes the hypotheses in chapter three and Operationalizes them according to the paradigm for the measurement of meaning. Next, a study is de- signed to test these hypotheses in order to gain insights into how bilinguals process information. It is h0ped that the study would also provide data which can be used to demonstrate the utility of the paradigm. Chapter five provides the results of the test of the eight hypotheses. Chapter six provides a discussion and a summary of the investigations presented in this volume. The first section of the chapter discusses the theory of bilingual information processing, as seems warranted based upon the results of the testing of the hypotheses. The second section discusses future research in the area of hi- lingualism. These ideas include a reconceptualization of the operationalization of one of the hypotheses which attempted to relate mass media exposure to semantic structure, a pr0posed time series analysis of emergent bi- linguals and, finally, a network analysis of a bilingual community in order to describe their patterns of inter- action. This chapter ends with a summary of the entire thesis. CHAPTER TWO A PARADIGM FOR THE MEASUREMENT OF MEANING Introduction Historically, philosophers, psychologists, lin- guists and sociologists have all been interested in the theoretical concept of meaning.1 Problems of identifying a precise meaning for the word "meaning" has extended from single morphemes and words and individual sentences to entire code systems. Yet despite this attention, there has been little agreement between fields as to the defini- tion of this construct. Even within each of the disci- plines, the situation is no better. Gilbert Harman describes the current state of affairs in philosophy in the following manner (1968:590). Philosophers approach the theory of meaning in three different ways. (1) Carnap, Ayer, Lewis, Firth, Hempel, Sellars, Quine, etc. take meaning to be connected with evidence and inference, a function of the place an expression has in one's "conceptual scheme" or of its role in some inferential "lan— guage game." (2) Morris, Stevenson, Grice, Katz, etc. take meaning to be a matter of the idea, thought, feeling, or motion that an expression can be used to communicate. (3) Wittgenstein(?), Austin, Hare, Nowell- Smith, Searle, Alston, etc. take meaning to have something to do with the speech acts the expression can be used to perform. Harman goes on to say that theories of the first type ig- ruxre the social aspects of language, i.e., its use as a communication tool. The second type suffers the 5 6 inadequacy of not separating thougfliand language, and the third approach ignores the creative aspects of language. In psychology, there exists the now familiar conflict between the behavioralists Watson (192”, 1930), Skinner (1957) and Bousfield (1953) who discount any notion of an internal meaning and prefer to focus on the actions of individuals without inferring any state of con- sciousness, and the cognitively orientated psycholinguists such as Osgood (1952), Deese (1965), and Cofer (1965). They acknowledge an internal meaning system which mediates between incoming stimuli and behavioral responses. The situation in theoretical linguistics is no better. While Bloomfield (1933) felt that an adequate description of meaning was beyond the scope of the natural and social sciences, Sapir (1921) and Whorf (1956) wrote that language structures thought and therefore contains elements of meaning. Chomsky (1965:“) writes, . . . in a technical sense, linguistic theory is mentalistic, since it is concerned with discovering a mental reality.underlying actual behavior. Observed use of language or hypo- thesized dispositions to respond, habits, and so on, may provide evidence as to the nature of this mental reality, but surely cannot constitute the actual subject matter of linguistics, if this is to be a serious discipline. Chomsky is the father of the linguistic school of generative grammar. He describes it as ". . . simply a system of rules that in some explicit and well-defined way assign structural (syntax) descriptions to sentences . . . 7 generative grammar does not, in itself prescribe the character or functioning of a perceptual model or a model of speech production (1965:8-9)." Compare Chomsky's con- ception of the role of meaning with the following quote from George Lakoff about generative semantics (1971:232). I would like to discuss some questions having to do with the theory of grammar. I assume that a grammar of a language is a system of rules that relates sounds in the language to their corresponding meanings, and that both phonetic and semantic representations are provided in language-independent way. Meaning for the sociologist has two different and seemingly contradictory senses. While meanings must be shared and universal within a group in order to insure the ongoing nature of the social entity (Dewey: 1922), they must at the same time be established as part of a dynamic interactive process between the individuals. This G. H. Mead describes as follows: (193H:81) . . . meaning, as we have seen, is found to be implicit in the structure of the social act, implicit in the relations among its three basic individual components: namely, the triadic relation of a gesture of one individual, a re5ponse to that gesture by a second individual, and completion of the given social act initiated by the gesture of the first individual. The above statement has been taken by Herbert Blumer (1966) and others as an implicit indication that the self (individual) plays the central role in the pro- cess of concept formation. This individualistic approach to sociological concept of meaning may be summarized by the following quote from Blumer (1966:5u8). 8 . . . the meaning of the object is conferred on it by the way in which the individual is prepared to act toward it. Finally, all objects--whether a mountain, a house, a speech, a toothache, a dream, a memory, or a sensation--are located in the individual's environment in the legitimate sense that in being designated by the individual they stand over against him as the designator. From Blumer's conceptualization, it seems nearly impossible for a society to develop what Durkheim (1938) calls the "collective representations" for the meaning of objects. These are shared meanings for linguistic utterances that exist outside the individual's con- sciousness. These signs are common for each member of society. There is normative pressure to act, think or feel in a similar manner in response to these symbols. As Durkheim writes (1961:28-29): If, on the other hand, the categories are, as we believe they are, essentially collec- tive representations, before all else, they should show the mental states of the group; they should depend upon the way in which this is founded and organized, upon its morphology, upon its religion, moral and economic institutions, etc. . . . Collec- tive representations are the result of an immense co-operation, which stretches out not only into space but into time as well; to make them a multitude of minds have associated, united and combined their ideas and sentiments; for them long gen- erations have accumulated their experience and their knowledge. This lack of theoretical unity has led sociology into the schizophrenic nature of concentrating on situa- tion-specific behaviors and at the same time attempting to identify patterns of behavior that are regular through- out society.2 9 The field of communication has its intellectual roots in the above mentioned disciplines. As a result, the lack of a unified approach to the concept of meaning is also manifest in this field. While Berlo (1960) takes an individualistic position by saying that, "Meanings are in people," Cushman and Whiting (1972) have taken the position that meanings are rule governed by their stan- dardized usage. Simply, if the meaning of a signal is to be understood, it must be shared by the interactants. It is the position of this paper that if there is going to be a science of communication then a single unified definition of meaning is a requirement. In order for the field to progress, there must be agreement as to what meaning is so that it can be recognized when it occurs, used as an independent variable which can be manipulated and controlled in order to predict changes in some resultant dependent variable, or as a dependent con- struct which can be predicted and explained as a function of a number of psychological and/or social factors. With a single clearly delineated subject matter focusing on specific aspects of the construct, it becomes possible to replicate research using the concept. Without it, it is impossible to develop a comparable body of knowledge with which to guide future research. This chapter will attempt to develop such a definition for the construct of meaning as it might be used for the field of communication. 10 The Fundamental Measurement of Meaning Since there is a great deal of confusion regard- ing the theoretical concept of meaning, the author pro- poses that an original and fundamental system of measure- ment be deve10ped for the concept's definition. This is known as an operational approach to science. Simply put, the meaning of any concept is nothing more than the opera- tions by which it is measured. The concept may be taken to be the value or series of values that result from the measurement process. The advantages and disadvantages with this approach to theory construction have been dis-. cussed in great depth by G. R. Miller (l97u). This has been the approach adopted by the natural sciences. Woelfel writes (1974:2), ". . . scientific theories have their ultimate roots in certain fundamental or primitive varia- bles which cannot be defined in terms of yet more basic concepts, but rather are defined, by some a priori call to experience of observers." For example, in physics the fundamental variables are usually considered to be distance, time, mass or force and temperature. Of these, only time and distance can be directly observed. The others can only be defined as ratios to time and distance. Einstein (l961:5) describes the measurement of distance in the following manner: For the purpose (the measurement of distance) we.require a "distance" (Rod S) which is to be used once and for all, and which we employ 11 as a standard measure. If, now, A and B are two points on a rigid body, we can construct the line joining them according to the rules of geometry; then, starting from A, we can mark off the distance S time after time until we reach B. The numSer of these Operations required is the numerical measure of the distance AB. This is the basis of all measurement—5f length. Commenting on Einstein, Woelfel writes (197u: Einstein's measurement procedure is two- staged: first, an arbitrary distance (or discrepancy, in the general case) is stipu- lated by the scientist. It is vital to note that rules for the perception or measurement of this initial measurement distance or discrepancy are not stated; rather the scientist must assume the subject and himself/ herself share a common referent for the ordi- nary language symbol "distance" or "difference," and that the subject can make this initial recognition unaided by further definition. Ultimately it is this a riori call to common experience as codified in ordinary language symbols that establishes a link between the everyday experience of the observer and the scientific theory. Secondly, the scientist specifies a rule by which other instances of distance or discre- pancy are to be compared to this unity. In this case, the observer is asked to make ratio comparisons of all other distances or discrepancies to this arbitrary standard. Clearly, fundamental measurement represents a joint activity of scientists and observer. It is the intent of this chapter to propose a fundamental measurement system for the meaning of indivi- dual.lexical items specifically for the science of human communication. What follows is not a totally inclusive system.applicable to all the fields surveyed above, but :pather5 one designed to meet the assumptions and the needs (xf communication. G. R. Miller (1975:6) writes: 12 But the crucial point to be emphasized is that the proponents of Operationism realizes that the boundaries of meaning are dictated by the instrumental procedures employed, that there is a wider focus of meaning not encompassed by this operational definition. The operational position toward definition may be contrasted with the Aristotelian, or essentialist approach, which holds that definitions are statements de- scribing the essence or true meanings of an object. The term to be defined (definiendum) is a name for the essence of an object, while the defining formula (definitions) is the description of the essence. Miller considers this position is unacceptable for the following reasons (G. R. Miller, 197”). One, the notion that there is a correct or true meaning of a term is considered erroneous. Two, this position does not require that the attributes of the definiendum be directly observable in physical or social reality.3 Three, essentialist definitions tend to be tautological. And, four, it implies that the underlying properties of an object are constant. If the description is altered, then the object is somehow not the same. Woelfel and Barnett (197u:5-6) have discussed the point that the perception of a single object as an unique object implies an arbitrary categorization which :renders discrete the continuous process of perceiving the environment, whereby an arbitrary segment of the continuum of stimulation is set aside and identified as an unique object. Thus, an attempt at identifying the unique attri- tnxtes of an object or thought is doomed to failure. 13 The author proposes that any fundamental system of meaning meet the following requirements: 1. The measurement system must be relational. That is, it must meaSure the degree of the relationship between the individual lexical items. 2. The system must take into account the con— sensual nature of meaning. 3. It must be dynamic. It must be capable of measuring change over time. And, N. The system must be empirically fruitful. Each of these assertions is based upon the past literature in philOSOphy and the social sciences, and the author's conceptualization of the communication pro- cess. These points will now be discussed separately in light of the above areas. ‘ The Relational Nature of Meaning Associationism The notion that individual events, objects, or ‘words are associationally related in human memory has its roots in antiquity with.the work of Aristotle. In "On [Memory and Reminiscence," he write that there is something systematic in the chain of thought which results in the 'recollection of particular thoughts (Aristotle, 19Hl:61u). This explains why it is that persons are supposed to.recollect sometimes by starting from mnemonic loci. The cause is that they pass swiftly in thought from one point to another, e.g., from milk to white, from in white to mist, and thence to moist, from which one remembers autumn (the "season of mists"), if this be the season he is trying to recollect. In this essay, Aristotle also isolated three sorts of relationships that govern the process of remember- ing. They are similarity, contrast and contiguity of exemplars in past experience. For Aristotle the process of remembering is closely related to the perceptual pro- cess. Thus, while the process of definition or determin- ing the specific meaning of an object for Aristotle was purely descriptive, his notion of memory implies a system of relations between individual objects or words. These three basic principles: similarity, contrast and contiquity were also taken as the basis of British Associative Theory. PhiIOSOphers of the mind who are considered members of this school are John Locke,- Thomas Hobbes, Bishop Berkeley, David Hume, David Hartley (the father of associationism), John and James Mill and Alexander Bain. For in depth discussions of their con- tributions to human information processing see Deese (1965), Schultz (1969) and Anderson and Bower (1973). While the contributions of these individuals to the dis- cussion of human thought were great, for purposes of this 3paper their contribution was largely to bring the ‘Aristotlean paradigm into modern thought by restating and clarifying the relational principles of the mind and 'thereby influencing contemporary psychological theory. 15 Two of the important advancements of this school of thought were that the structure of the mind at any one point in time is contingent on the individual's past experience (as opposed to each thought having an innate and proPer locus in the mind), and Locke's atomistic notion that all knowledge could be derived from discrete simple ideas which through the associational process are combined into complex ideas. Simple ideas are thought to be elemental or unanalyzable. Hume discussed the notion of complex ideas further and said that they do not necessarily resemble simple ideas, since the complex ideas evolve from a combination of several ideas in some new and novel pattern. Complex ideas are compounded from simple ideas by the associational processes. This notion is similar to Barnett's (1975) conception of complex innovations. Similarity.--The principle of similarity states that objects which are perceptually similar will be associated in an individual's mind. For example, thought of lemons may easily lead to thoughts of oranges, although it is likely that either might not be present with the other or appear in immediated succession (the principle (Jf contiguity). What is apparent, however, is that the 'two objects are perceptually similar. They share certain attributes. Among them are shape, texture, internal :physical structure, and color (before dying the skins for 16 marketing purposes). The principle of similarity is a synchronous association which combines simultaneous ideas into more complex ideas; in this case citrus fruits. The notion of similarity is structural because it relates the ideas of thought into a single coherent series of rela- tions or structures. The notion of word-substitution may be used for illustrative purposes in a discussion of the principle of similarity. If two words are synonyms, i.e., they are semantically identical, then the latter can replace the former without any alteration in the interrelationships among the symbols. If they are not, the words are semanti- cally different, then the first symbol cannot be replaced by the second without altering the structure of the rela- tions. The greater the dissimilarity between the terms, the greater the interrelationship among the symbols will be changed. Similarity was taken as an irreducible law of association by Locke, Berkeley, Hume and Bain, although Sml__.____=> RX]. 82 '+ 13$? ————————— '2' Sm2___.—__—_& RX S + r ————————— + s ______i» R X n $1.51 mn n S T rma ““““““ + Sma‘-—__-'T RXa Figure l.--Symbolic Account of the Development of Sign Processes. aCharles Osgood, et. a1., (l957:7). symbols was also evoked by a different school of philoso- phers. They may be labelled the formal logicians. Representative of this tradition are Gottlob Frege and Ludwig Wittgenstein. To Frege (1970) meanings or senses emerge from the relationship between signs. He writes (1970:56-57), Now if we were to regard equality as a relation between that which the names "a" and "b" designate, it would seem that a=b could not differ from a=a (i.e. provided a=b is true). A relation would thereby be expressed of a think to itself, and indeed one in which each thing stands to itself but to no other thing. What is intended to be said by a=b seems to be that the signs or names "a" and "b" designate the same thing, so that those signs themselves would be under discussion; a relation between them would be asserted. But this relation would hold between the 24 names or signs only in so far as they named or designated something. It would be media- ted by the connexion Of each of the two signs with the same designated thing. But this is arbitrary. Nobody can be forbidden to use any arbitrarily producible event or object as a sign for something. In that case the sentence a=b would no longer refer to the subject matter, but only to its mode of designation; we would express no prOper knowledge by its means. But in many cases this is just what we want to do. If the sign "a" is distinguished from the sign "b" only as object (here, by means of its shape), not as sign (i.e. not by the manner in which it designates something), the cognitive value of a=a becomes essen- tially equal to that of a=b, provided a=b is true. A difference can arise only if the difference between the signs corres- ponds to a difference in the mode of presentation of that which is designated. Since the chosen signs and their relationships are arbitrary, there is nothing to keep the connection between A and B from becoming A = B/C, or A =‘CB or any other functional relationship. In this way, the precise pair-wise association between A and B can be specified. The limiting condition to Frege is the truth of the relationship. Thus, for Frege meanings must be empirically testable. When an object is named or designated as a mem- ber of a class, A, it implies the proposition A =NB or A = lel + BZXZ + . . . + Ban, where the X's are the various attributes Of the Object and the B's the specified relationships between the attributes and the object.5 For example, when a lemon is specified as such, it implies ‘that it is yellow, round and has a skin of a certain 25 texture. Likewise, for an orange, it becomes possible to specify the relationship between this object and its attributes. By comparing the relationship between the attributes, it becomes possible to specify the relation- ship between the objects. For purposes of this paper, this relation is similarity. In the case of the lemon, if the object does not possess all the attributes necessary to become a member of the class of lemons (rather than being sour, the lemon is sweet), then A f B. Two moves are then possible. One, the object is not a member of the class of lemons, A = C. Or, two, the class of Objects must be redefined, A = 8*, an altered set Of attributes which now describe the object. Frege's earlier discussion dealt only with pair- wise relations. Later in the same volume he expands this notion to groups of concepts. He writes (1970:159) that precise meaning derives out of the relationships among groups of symbols. A definition of a concept (of a possible predicate) must be complete; it must unambiguously determine, as regards any object, whether or not it falls under the concept . . . . If we represent concepts in extension by areas on a plane, this is admittedly a picture that may be used only with caution, but here it can do us good service. To a concept without sharp boundary there would correspond an area that had not a sharp boundary-line all around, but in places just vaguely faded away into the background. This would not really be an area at all; and like- wise a concept that is not sharply defined 26 is wrongly termed a concept. Such quasi- conceptual constructions cannot be recog- nized as concepts by logic; it is impossible to lay down precise laws for them. The law of excluded middle is really just another form of the require- ment that the concept should have a sharp boundary. In Wittgenstein's Tractatus Logico Philosophicus (1922), he says that meaning or facts exist only as states of affairs which are dynamic and empirically testable. A state of affairs must be expressed by a prOposition, such as, "the lemon is yellow." As such, a state of affairs implies a constellation or a series of relations among a group of objects. Objects make their appearance only with a space of possible states of affairs, as in the case where the lemon is yellow. This is only one of the possible states of affairs. The lemon could have been orange or green as well as yellow. Thus, states of affairs are contingent facts--they only happen to exist. According to C. A. Van Peursen (1970:35-36), Wittgenstein wants to elucidate the logical structure of the world. This is accomplished in the following manner. He therefore tries to present states of affairs or as he also says, the connection of Objects in the form of logical language. Just as in geometry it is possible to project objects on a plane, so a particular state of affairs can be projected in a logical picture. This is done by showing that certain logical elements are mutually related in a certain way. The glass is on the table, for example can be expressed in logical terms "aRb," where "a" and "b" indicate things and "R" indicates a rela- tion. It is this kind of connection that indicates the facts within what Wittgen- stein calls logical space. Logical space 27 covers the field of the possible relations that may Obtain between states of affairs, and that are realized in positive facts. "The glass is on the table" is an indica- tion of such a positive fact, translatable in logical language as "aRb" or in terms of signs representing whole sentences (atomic propositions), a "p." These sen- tences, or, to use the more technical term, these "propositions," may occur in extensive constellations. In Wittgenstein's later work (1953) he abandons his previous position that it is possible by means of logic to reveal the hidden structure of language. Rather than there being one definitie meaning for a single lexical item, there are in fact as many meanings in language as there are ways in which the symbol is employed in ordi- nary use. There are many language-games, each of which is justified within the human situation in which it is applied. Words acquire their meaning with the whole con- text Of their use. The notion that words can take on many different meanings depending on their context of use rather than forcing an abandonment of the earlier conceptualization warrants only a slight modification Of the system of analysis. In any given context, a specific lexical item may be defined by its relationships with the other items that are used in that context. For example, in a fruit store, the lemon becomes defined by its relationship to other fruits, in a restaurant with other foods and drinks, and with regards to automobiles to other descriptors of the quality of vehicles. While it is possible to describe 28 the lemon in all those contexts simultaneously, this would only serve to confuse the issue. The selection of the concepts upon which to base a word's definition should be determined by the context in which it is potentially used. As such, the definition becomes context specific and empirically testable within that situation. Had one attempted to determine the degree of similarity between every lexical item for the entire range of situation in which the symbol is used, it would be nearly impossible to determine the truth-value of the proposition. Communication A similar position is taken from the perspective of contempory communication theory by Woelfel. He writes (l97l:Chapter 5, page H), The process of a definition is a process of relating objects of thought to each other. Fundamentally, this involves tak- ing note of similarities and differences between objects, of identifying the attributes of an Object with similar attributes of different objects, and differentiating the attributes of the object from those attributes of the object "which are different" . . . . The perception of a single Object as a unique object implies a process of categorization in that all the discrete stimuli which constitute the physiological mechanism of that perception are set apart from the totality of stimuli impinging on the individual at the time and designated as a single object of thought. The category renders discrete a continuous process of 29 exposure to the environment, whereby an arbitrary segment of a continuum of stimulation is set aside and identified as a unique object. When an individual identifies two objects as "yellow," for example, it is not implied that they are the same color, only that they are similar enough to be described by the same linguistic symbol. The visual color spectrum covers the range from about 4,000 to 6,N00 AngstrOm units, and research indicates that color differences of only a few AngstrOm units are perceivable (Munsell, Sloan and Godlove, 1933; Halsey, 195N); yet ordinary language does not provide color terms for all these differences. The ordinary language people speak (and for the most part the symbol system of social science) allows only a crudely approximate description of the perception of any object. The continuous set of positive real numbers Offers a potentially error-free language for the defini- tion Of any set of objects with a level of precision far greater than the limits imposed by human perception. Woelfel states that (l972:ll[Chapter 9]): Dissimilarities among objects (whatever those objects may be) may be represented by a continuous numbering system such that two Objects considered to be completely identical are assigned a paired dissimilarity score or distance score of zero(0), and Objects of increasing dissimilarity are represented by numbers of increasing value. Assuming that the definition of an Object or concept is constituted by the pattern of its relationship to other objects, the 30 definition of any object may be represented by an 1 X n vector, 811, 812, 813, . . ., Sln’ where 311 represents the distance or dissimilarity of Object 1 from itself (thus S l = 0 by definition), 812 represents the disEance or dissimilarity between objects 1 and 2, and S n represents the distance between the ls and nth objects. Similarly, the second object may be represented by a second vector, S , S , S , . . ., S , and the definitigg ofzgny ggt of conceggs or Objects may therefore be represented in terms of the matrix 811, Sl2,ooo, Sln 821, 822,000, 821'] Snl’ Sn2,..., Snn where any entry S-- represents the dissimi- . O 1 C . larity or distance between i and 1° This dissimilarity matrix describes the state of affairs (static structure) of the interrelationships among a set of objects at a point in time. The matrix is contingent on an individual's past information. There are an infinite number of possible relationships between the symbols, because the only limiting factor for any cell is that the dissimilarity be a positive real number. Thus, specific definitions are testable. Additionally, it has particularly attractive mathematical or formal properties, including a well—developed set of Operations defined for such matrices. Finally, in order to determine the definition of a given term specific to a particular context, one only has to choose the symbols which describes 31 that context and determine the degree of similarity be- tween these items. In summary, this section Of the chapter has discussed the historical development of the notion that the meaning of an Object can be determined by its rela- tionships to other Objects. Specifically, the relation which most accurately describes a state of affairs is similarity. The structure of meaning at any one point in time is contingent upon an individual's past experi- ences and the context in which the symbol is used. In a later section of this paper, the measurement of similarity will be discussed. The Consensual Nature of Meaning The notion that meanings must be considered as consensual or shared relationships can be justified from philosophy, sociolOgY, linguistics and contemporary thought within the field of communication. Indeed, if meaningful ideas are to be exchanged between two or more people, they must share common definitions for the symbols used in this transfer of information. Deese (1965) takes the position that when investigating "psychological meaning" one must look at normative responses rather than the idiosyncratic responses of individuals. In this way, the results will be more generalizable and indicate the symbol's communication value. 32 Philosophy Wittgenstein (1953) takes the position in his later work that meaning does not derive from internal conditions of the individual mind, or from private a priori experiences, but rather, from the entire social convention which is public and based upon the rules that govern language. The meaning of a symbol can be deter- mined by its use and the standard usage of an expression is shown by describing the place it was given (relation- ship to the other symbols) in what Wittgenstein calls a language-game. Language-games may be considered the various activities of life. The use of the symbol may vary between these situations and its meaning will also vary becoming subject to the rules that happen to be evoked in that particular situation. Stephen Toulmin (1969:73) has summarized Wittgenstein's position as follows. Any expression owes its linguistic meaning (Wittgenstein taught) to having been given a standard rule-governed use or uses, in the context of such activities: in isola- tion from any activity Of this sort, the expression itself would lose all linguistic status and would become a mere mark or noise--an "idle wheel," engaging in nothing. Language-games in turn, however, must be understood in their own broader context; and for those contexts Wittgenstein intro- duced the phrase "forms of life." Later in the same essay Toulmin extends Wittgen- stein's conceptualization of language to all non-linguis- tic behavior. However, he makes a minor qualification 33 to distinguish significant human actions from meaningful linguistic expression (1969:85). Quite apart from the actual use made Of an expression on some particular occasion of utterance, we can discuss also its STANDARDIZED (conventional use), which arises out of the STANDARD LANGUAGE games in which it normally figures: this con- ventional use is what determines the "liberal" or dictionary meaning of the expression. Following Toulmin, Cushman and Whiting write (1971:217), We believe there is a class of human actions whose significance is largely dependent on consensually shared rules. These rules control the unfolding of the action over time and constitute its meaning. Sociology For language these rules are embedded at a societal or cultural level. The meaning of a symbol is based upon its standardized usage. Language rules are indistinguishable from society's other norms. When an individual violates the rules of a language, they are subject to the same sanctions that result from the violation of other mores. In the case of language, this sanction is, among other things, non-reciprocated communi- cation. Thus, a necessary condition for communication is that the linguistic rules are shared by the partici- pants Of the interaction. "Certain rules must be consensually held among communicating participants for the communication to occur" (Cushman and Whiting, 1972: 218). an In sociology, it is a given that society possesses properties which cannot be attributed to a single individual. One such social Object is language. Durkheim (1953) takes the position that there are social facts that are in a sense independent of individuals and exterior to individual minds. These he labels collective representa- tion.6 . . . collective representations are ex- terior to individual minds, it means that they do not derive from them as such, but from association of minds, which is a very different thing. NO doubt in the making of the whole each contributes his part, but private sentiments do not become social except by combination under the action of the sui generis forces developed in assoc- iation. In such a combination with the mutual alterations involved, they become something else . . . . The resultant sur- passes the individual as the whole part. It is in the whole as it is by the whole. In this sense it is exterior to the in- dividual. NO doubt each individual con- tains a part, but the whole is found in no one. In order to understand it as it is one must take the aggregate in its totality into consideration (1953:25- 26). Collective representations are formed during the communication process. In this sense, Durkheim shares Wittgenstein's position that meanings emerge out of the use of symbols, and G. H. Mead's (193N) notion that meaning arises out of the interactive process. However, Durkheim takes the position that once the collective representations, are formed they become a prOperty of society. They are independent of the individual. For Mead, collective repre- sentations are possessed by a "generalized other." For 35 Durkheim, the representations are part of the "collective consciousness." As such, they are passed intergenera— tionaly and need not be derived through interaction for each individual as the generalized other. Language like other social facts has very dis- tinctive characteristics: ". . . it consists of ways of action, thinking (meaning), and feeling, external to the individual, and endowed with a power of coercion, . . ." (1938z3). Durkheim notes that (1938:10), . . . Currents of opinion, with an inten- sity vary according to the time and place, impel certain groups either to more marriages, for example, or to more sui- cides, or to a higher or lower birth- rate, etc. . . . Since each of these figures contain all the individual cases indiscriminately, the individual circum- stances which may have had a share in the production of the phenomenon are neu- tralized and, consequently, do not con- tribute to its determination. The average, than expresses a certain state of theggroup mind. Thus, it becomes possible to identify the collective representations or group meaning, by simply taking the average relationship between the symbols pro- cessed by all the users of a language. Linguistics In a sense the language of all individuals and the portion of that language that entails the semantic system is unique to that person. Since no two people use their language in exactly the same manner, there can be as many representations of the language as there are 36 human beings who use that language. Linguists have coined the term "idiolect" to describe the speech of a particular individual. By comparing the features of idiolects, lin- guists have come up with certain regularities for sub— groups of a population which are not found in other sub- groups. These groups linguists label dialects. These sub- groups are usually determined by the clustering of indivi- duals by geographical location Or social class. In order to describe the collective representation of an entire language, one would have to determine the average idiolect for an entire culture. In classical linguistic theory de Saussure (1959) distinguishes between 13 parole, the language of the indi- vidual and 13 langue, the cumulative, dynamic, consensually valid language that is used in communication.‘ L3 langue must be described in terms of the cumulative 13 parole. He writes (1959:15), "Among all the individuals that are linked together by speech, some sort of average will be set up: all will reproduce--not exactly, of course, but approximately-~the same signs united with the same concepts." Commenting on the relationship between the lan- guage of the individual and one possessed by the collec- tivity, Edward Sapir (1921:1N8) writes, . . . there is something like an ideal lin- Aguistic entity dominating the speech habits of the members of each group, that the sense of almost unlimited freedom which each individual feels in the use of his language is held in leash by a tacitly directing 37 norm. One individual plays on the norm in a way peculiar to himself, the next individual is nearer the dead average in that particular respect in which the first speaker most characteristically departs from it but in turn diverges from the average in a way peculiar to himself, and so on. What keeps the individual's varia- tions from rising to dialectic importance is not merely the fact that they are in any event of small moment--there are well- marked dialectic variations that are Of no greater magnitude than individual var- iations within a dialect--it is chiefly that they are silently "corrected" or canceled by the consensus of usage. If all the speakers of a given dialect were arranged in order in accordance with the degree of their conformity to average usage, there is little doubt that they would constitute a very finely inter- grading series clustered about a well- defined center or norm. In summary, this section of the paper has taken the position that the meaning of a lexical item lies in the collective consciousness of the society that uses the language. Meanings must be consensual in order for communication to occur. Language users are forced by society's norms to use the standard usage or meaning for any symbol. This standard meaning may be determined by taking the average representation held by the users of the language. Thus, combining the notions of the last two sections, the meaning of any set of lexical items at any one point in time for a given cultural system may be determined by the "average" dissimilarity matrix, where any entry Sij is some measure of central tendency of the dissimilarity between Objects i and j as seen by the members of a culture. Because of its mathematical 38 properties the chosen measure of central tendency should be the arithmetic mean estimate of a society's members. The Dynamic Nature of Meaning Communication has been defined as "a process of transmission of structure among the parts of a system which are identifiable in time and space" (Krippendorff, 1969:107).7 As such, the communication scientist is obligated to study change in structure over time. Earlier in this paper, the point was made that the meaning of a set of Objects at any one point in time may be taken to be the structure of the interrelationship among a set of symbols used in a particular context. From a communica- tion perspective, the study of meaning would entail Ob- serving the changes in meaning structure over time as a function of the information made available to the social entity under investigation. If the structure of meaning processed by an individual is contingent upon the person's past experiences, then communication researchers are obligated to study the antecedent conditions that pro- duced the relationship. This method has been the choice of the researchers of developmental processes such as language acquisition (Bloom, 1970) or status and occupational attainment (Woelfel and Haller, 1971). An alternative to this approach would be to investigate the change in meaning over a given interval of time and attempt to isolate the parameters which cause the structure 39 to be stable or volatile. The central concept in the con- struct process is a dimension of time and the change in the characteristics or prOperties of the structure be- tween a series of time intervals. The notion of process as a communication concept and its implications for re- search has been discussed in great depth by Arundale (1971). It was pointed out earlier that the association theorists have recognized this dimension of time. While they identified the approximate point in time when the cognitive structure changes, they have only indicated that an association was formed. How the interrelations between the various verbal stimuli were altered was not addressed.8 Thus, from the position of this paper, the associational- behavioralist's notion of time may be considered incomplete. Wittgenstein's (1922) notion that meanings exist only as states of affairs makes explicit the idea that meanings are dynamic and change as the structural rela- tions among a constellation of Objects are altered. States of affairs may be considered events; events which are con- tingent, happening to exist, but which may exist only given certain conditions at a prior point in time. De Saussure (1959) posits two analytical branches of linguistics, synchronic and diachronic. They may be differentiated along two dimensions: one, internal- external, and two, static-historical study of language. Synchronic linguistics is internal and static,--it studies no language as an autonomous system, free from change and influence from external or historical factors. Synchronic linguistics studies the ". . . arrangement of linguistic elements for a given language state" (Hertzler, 1965:12). Diachronic linguistics is concerned with the external aspects of a language. It deals with what is outside the language system, yet affects it, such as the relationship between historical events or a society's in- stitutions and the language. While synchronic linguistics excludes the dimension of time, concentrating only on single states of affairs, diachronic linguistics is con- cerned with the evolutionary study of language. In de Saussure's own terms (1959:1u0), Diachronic linguistics is concerned with the "evolutionary phase," "the stream of language," with the divergences in time and the relations between successive terms that are substituted for each other within the system in time. The synchronic aspects of semantics has already been discussed in great detail. The static internal re- lationship alone, however, cannot adequately describe the meaning of a given symbol. Besides the single symbol being embedded in a language, is the hierarchical arrange- ment that places language as part of a larger social system. Societies are dynamic. They change as a function of social upheavals such as war, natural disasters, and technological innovations. New needs place demands upon the institutions of society forcing them to be altered 1+1 in response to these demands. It is in this context that language is placed. Until the 1950's, the symbol "rock" was only related to the lexical items hard, building material, mineral, etc. Today, it has associations with music, drugs and loud noises. Summarizing these notions Hertzler writes (1965: um, We know, from many well-attested instances that, given time enough, every language in current use, even in the most remote and conservative communities, does succeed, by means of processes and devices, in accommo- dating itself sufficiently to the new elements of the environment, the new popu- lation elements, the new communal experi- ences, the new awarenesses, interests and needs especially in so far as these relate to technical and social achievement-~50 that it is able to serve its basic general and societal functions. At certain times and under certain circumstances, these changes are glacially slow and barely dis- tinguishable, even to expert students of language; at other times and under other circumstances what has been referred to as a "linguistic revolution" is in pro- gress, especially in certain aspects Of the language. To summarize, this section has presented the notion that meanings are part of a dynamic system-- language. Meanings change as a function of new informa- tion being made available to the members of a social system. Communication has been defined as a process, and, as such, researchers in the field are obligated to take into account changes in the structure of relationships over time. H2 Summarizing the chapter so far, the structure of meaning (a state of affairs) of any set of lexical items at any one point in time may be determined, by the average dissimilarity matrix. Process may be recorded in successive matrices §£0’ Stl,...Stn, where the interval between time periods 0,1,2,...n, remain constant and the changes between the matrices calculated.9 These inter- vals can be made as small as desired to increase the isomorphism with the continuous nature of process. The difference St -St would represent the semantic change 0 1 taking place over the interval from t0 to t1. The rate at which any definition is changing can be found by the derivative, 8 -S _d_x_= limit + 0 ti 1:0 Empirical Fruitfulness The notion of the empirical refers to a system of observation by which sensory data is transformed or mapped into a one-to-one correspondence to some score or scale value. By performing the Operations which result in the scale value, it becomes possible to draw Objective evidence allowing one to determine relationships between two or more variables. Once the relationship among a set of variables is specified through the measurement procedures, the researcher can test hypotheses about these relationships. Through the hypothesis testing us procedure, the researcher can then determine the truth— value of certain propositions. For example, if one states the prOposition "the average dissimilarity held by a set of Observers between red and white is 100 units," then, by performing a series of Operations, it is determined that the average dissimilarity is a scale value of 90 units. With this information, the truth-value of the proposition can be rejected. For Frege (1970) the truth value of a relation- ship is a necessary condition for sense or meaning. The theory of meaning is based on verification. If a pro- position has no truth-value, then a statement becomes meaningless. Ayer (l9u6:5) describes the principle of verification as follows: The principle of verification is supposed to furnish a criterion by which it can be determined whether or not a sentence is meaningful. A simple way to formulate it would be to say that a sentence had literal meaning if and only if the pro- position it expressed was either analy- tical or empirically verifiable. Thus, any system of measurement of meaning must allow the researcher to determine whether a statement is true or false. This conceptualization was supported by Wittgenstein (1922), and the logical positivists. They asserted that there is an empirical reality from which one draws the data to test a proposition. While there are an almost infinite number Of other ways to approach the notion of meaning, the author suggests only one. That 1m one is based on empirical measurement. The reasons are twofold. First, since meaning can be considered con- sensual, one must make a series of Observations from a population in order to determine that consensus. But rather than this consensus being considered "real," it may be construed as the average Observation of a group of Observers. In this way, no call to a referrent is made. Rather, the relationship between the Objects is analyti- cal, coming from in some sense inside the Observers heads. Where the observers obtain these conceptions is irrelevant. Verification, then, becomes based upon the agreement rather than reality. Since the relationship is derived from the observer's consciousness, it becomes possible to discuss abstract concepts that have no empirical referrent. Examples of these concepts are democracy, communism, God and good or bad. These terms could not be discussed or entered into relationships in a strict empirical position. Second,is the notion of prediction. The empirical method does not end with the process Of hypothesis testing. The scale value Obtained through a measurement at one point in time allows the researcher to predict the value (Specify the relationship) at a future point in time. The ultimate utility Of any measurement system is its ability to predict the value or values of one or more variables from one or more variables at a previous point in time. The notion Of predictability in some sense U5 parallels the concepts of replication and verification for science. If the actual relationship between the varia- bles is changing, then it would be impossible to replicate the results of the study at a future point in time. Since there are social processes going on around us at all times, one might expect the relationships to be changing as a function of these processes. This is especially true of human phenomena. For example, in the 19H0's, religious affiliation was an excellent predictor of political party identification (Lazarsfeld, Berelson and Gaudet, iguu). The reason was that a majority of a person's interpersonal communication about political topics took place within the context of the church. Today, religion is a poor predictor of political party preferences because the church no longer plays a central role in an individual's life. Thus, when one attempts to replicate or verify this rela- tionship, one fails. This example dealt only with the relationship between two specific variables. Specific variables are those which are particular to an historical period or cultural entity. As such, their relations to other variables are expected to change as historical epochs end or cultures disintegrate. However, even the relationship taken from more general variables such as the relation between age and frequency of sexual inter- course may change over time as a function Of population pressures, food supply and cultural norms. In some sense, one could get around the problem if it could be determined A6 the rate of change in the relationship among the varia— bles. Thus future scientists, rather than attempting to replicate findings, would attempt to predict some re- lationship. Verification becomes the process of predic- tion. After a series of predictions, it would become possible to determine the rate change in the relationship and make adjustments in the theory with considerations for the change. The notion of operationalism is only partly re- lated to the empirical. The operational method allows one to see a phenomenon when one performs a series of operations. It need not result in a value. The empirical method not only requires a series Of operations to obtain a value or series of values, but requires that one look at the relationship among a group of variables, in order to determine the truth—value of some theoretical proposi- tion which led one to investigate the relationship in the first place. Thus, the ultimate criterion for a system of measurement of meaning would rest on its predictive power. Could one determine the relationships among a series of lexical symbols at a future point in time despite by tak- ing into account intervening social events? This predic- tion is based upon the average score which resulted from a group of observers. It is not based on a "real" re- ferrent, but rather, the collective or average conscious- ness of the observers. 1+7 The Measurement of Meaning This section of the chapter asks the question, how has meaning been measured in the past? Since there is such a degree of confusion about what meaning is theoretically, one should adopt the position that the meaning of a lexical item is its measured relationship with other words that are used in the same context. This must be consensual--the average relationship among a series of language users. It must also be dynamic and capable of predicting change in the relationship over time. Traditionally, there have been four major methods for the measurement of meaning of lexical items. They are free association, free recall, the semantic 10 This section differential and multidimensional scaling. of the thesflswill review the brief history of each method, provide an example of each and finally, critique them on the criteria presented above. Finally, it will suggest that a modification of classical multidimensional scaling be adopted as a paradigm for the measurement of meaning. Free Association Free association, the technique of providing a subject with a verbal stimulus and recording his/her immediate response, was developed as a psychological measure by Carl Jung in 190N. Jung's work with associa- tion in absnormal individuals culminated with the publica- tion of Studies in Word Association (1918). Kent and 48 Rosanoff (1910) obtained responses to a hundred common English words from over 1000 abnormal subject and modi— fications of their counts of response frequencies are still used as a normative base for contemporary studies of association. Historically, the majority of the work with free association has dealt with psychoanalytical or psychodynamic problems. They have only indifferently applied themselves to the description of intellectual structure. Deese writes (1965:27), Armed with a number of ideas about the dynamic interpretation of dreams, associa- tions, and the like, they have constructed pictures of personalities and the histories of personalities by linking to- gether the bits and pieces they find in free expressive behavior. Free associa- tions have been a favorite source of data because of the assumption that they, more nearly than any other intellectual product, escape the censoring of the ego. To the extent that dynamic theorists have been interested in intellectual content, they have concentrated on the deviat and un- usual and in some instances, the symbolic— ally interesting aspects of free associa— tion. Few dynamic and almost no psycho- analytical theorist wish to make anything of free associations of high frequency of occurrence. It is of little concern (save for such values as in normative informa- tion) that the most common response to man is woman or that the most common re- 5ponse to woman is man. These are Obvious; they are given by nearly everyone. There- fore, they have not been regarded as pro- viding much information about underlying dynamics. This situation changed in the 1950's. Psycho- logists began to use free association to investigate the organization of the mind, and the psychology of language. U9 The assumption was that the meaning of any stimulus word could be described by the characteristics of the distri- bution of responses obtained from that symbol. Jenkins and Cofer (1957) used free association to assess the overlap between response distributions to compound verbal stimuli as well as the elements of the compound. Rosen and Russell (1957) demonstrated that re— sponse hierarchies obtained from an individual giving successive associations to the same word was similar to the hierarchy obtained from single responses to the same word given by a group. This early work with free assoc- iation is summarized in Marshall and Cofer (1963) and Creelman (1966). Important contributions were also made by Noble (1952), Bousfield, Whitmarsh and Berkowitz (1958), Jenkins, Mind and Russell (1958), Deese (1959), Marshall and Cofer (1963), Rothkopf and Coke (l961a,b). These articles all report attempts at converting the data gathered from associations to some sort of descriptive summary measures of similarity or word relatedness. The measures were all based on associative frequency and associative overlap among the stimuli. One conclusion which these studies agree upon was that words which are more similar tend to elicit each other. Their reasoning was that if the associative meaning of any stimulus is given by the distribution of the responses to that stimulus, then two stimuli may be said to have the same meaning when the distribution of associates is identical. 50 Two stimuli overlap or resemble one another in associative meaning to the extent they have the same distribution of associates. Or, in other words, the most common seman- tic relationship used in free association is synonymy. This notion of synonymy led Bousfield, Cohen and Whitmarch (1958) to the concept of implicit associated responses-- a word elicits itself as an implicit response 100 percent of the time. Deese (1962) performed a free association study used 19 words about nature as stimuli.ll A matrix of associative overlap was constructed from the number of times a stimulus elicited one another or a response outside the set but common to both. The percentage of common responses was placed in each cell. On the diagonal was placed the number of subjects under the assumption of the implicit associated response, thus 100 percent. The matrix was then converted to overlap coefficients, rather than corre- lations, by taking the ratio Of the sum Of the overlapping response frequencies to any pair Of stimuli to the maximum possible sum. The reason why correlation was rejected as a measure of association were because subjects get only one response and correlation would yield a measure of distribution rather than the extent to which frequencies are in common. Also the distributions are too steep to differentiate the words. This new matrix was then factor analyzed to reveal six factors. When orthogonally rotated, the 51 resultant structure found the animate creatures on the first factor (moth, insect, wing, bird, fly, bug, cocoon, bees and butterfly). The second factor was composed of the words which did not have anything to do with the con- cepts on factor one (yellow, flower, color, blue, summer, sunshine, garden, sky, nature and spring). Factor three was made up of the concepts which loaded on the first dimension, but it split up bug, cocoon, butterfly and moth from the rest of the concepts. Factor four performed the same operation upon the second dimension, separating the color terms from the seasons. One important thing to notice was that due to the structure of the data input into the factor analysis, the loadings were either highly positive or negative or zero. The factors did not vary in magnitude. Thus, free association can provide an indication of semantic structure. Concepts which are more similar tend to load on the same factors due to their degree of associative overlap. However, this is a poor measure of structure at best. One of the problems with free association as a measure of meaning is a result of the method's structure. The reliability of a response is inversely related to the degree of structure in the instrument. Free association is highly unstructured. As a result, the responses to a given stimulus are very unreliable. Unreliability may be somewhat ameliorated by increasing the number of sub- jects or the frequency of individual responses to a given 52 stimulus. The use of large numbers of subjects is very costly and time consuming. Additionally, the reSpondents free associations are very situation specific. The demand characteristics of the setting often determine the specific response. The instrument was developed in this manner so that psychodynamic researchers could investigate the volatile nature of personality states. The later approach of having a subject respond a number of times to a given stimulus is also unacceptable because it leads to chaining (Deese, 1965; Rommetveit, 1968). Chaining occurs when the response at one time mediates the next response to the stimulus. Thus, it becomes impossible to determine the domain of associa- tions an individual stimulus word possesses and its rela— tionship to a given set of lexical items. An example of chaining would be when the stimulus word "lake" leads to "cabin" which mediates the future response "hiking," "hunting" and so on. Had the first response been "water" then the next could have been "drink," "swim," "bathe," and so on. Chaining has led a number of investigators to network models of language (Anderson and Bower, 1973; Collins and Quillen, 1972). In terms of the criteria established in the first part of the thesis free association is unacceptable as a measure of meaning. It does not describe the degree of association between a given set of concepts at any one time. Rather, it only indicates that an association 53 exists. Also, the basis of the associative relation may also be one of a number Of types rather than similarity (Deese, 1965). Other semantic relations may be incom- patibility (the set of colors), antonymy (big-small), hyponymy (tulip-flower), converse (buy-sell), or conse— quence (fire-smoke) Kintsch, 1972). However, as in the case of the principle of contrast, these associations may also be considered as special cases of similarity. They are similar on the dimension specified by Kintsch. As Deese (1962) pointed out, the structure of data gathered from association are too steep to determine magnitude Of association. Thus, only clusters or groupings of symbols are revealed. Inherent in this method is the idea that the associations are idiosyncratic for an individual. It is not designed to be a measure of consensual meaning. Additionally, since the structure of the results are very unstable, it would be next to impossible to separate true change from random fluxuations (unreliability) in the data. Thus the method is incapable of measuring change in meaning over time. Finally, since it is unable to measure change over time, it cannot be used as a predic- tive measure. Free Recall Free recall as an experimental method of study- ing memory is a decendent of the method of retained members. The method was first used by several investigators 54 prior to 1900. It involved either simultaneous or suc— cessive presentation to subjects a series of items (objects, pictures, words, syllables, geometric figures, letters or digits) either once or more and the recall by the sub- jects of as many of the responses corresponding to the presentated items as possible, usually immediately after the completion of the presentation. The order in which the subject recalled the items was not specified by the experimenter. It was quite a popular method. These early studies were conducted with recall of individual items as a function of variables such as the characteristics of individual items, amount of material presented, sensory modality of presentation, serial position, length of re- tention interval, nature Of activities between presenta- tion and recall and the demographic characteristics of the subjects. The popularity of the method was short lived because Ebbinghaus (1902) felt that it had only limited utility and that his measures of retention were more sensitive. This early history is reviewed in great depth by Tulving (1968). The method regained its respectability as a. measure of associative strength in the l9u0's with the work of Postman, Egan and Davis (l9u8). Free recall, as the method was now labelled, was regarded as being sensi- tive to intermediate degrees of association. Also, its use became restricted to verbal learning research (Post- man, Adams and Phillips, 1955; Bush and Mosteller, 1955; 55 Miller and McGill, 1952; Waugh and Smith, 1962) and the organization of verbal symbols in the mind as a function of past experiences (Bousfield, 1953; Bousfield and Cohen, 1953; Jenkins and Russell, 1952; Bousfield, gt a1., 196”). The method of free recall operates as follows. A list of words (usually H0) is presented to a subject in serial order and the subject is told to recall the items in the order he/she thinks of them. The stimulus may be a list of random words (Tulving, 1962), groups of re- lated words in random order (Cofer, 1959; Lambert, Ingnatow and Krauthamer, 1968), or related words blocked, some nonrandom order in which the stimuli are grouped for research purposes (Cofer, 1967; Puff, 1966). The list is presented either once (Bousfield and Cohen, 1953) or a number of times (Underwood, 1969; Tulving, l96u; Bousfield, 196”). The multitrial design is better for measuring the associative processes (Tulving, 1968). The two major dependent variables in this research have been the number of words recalled and the organization of recall. Tulving (1962) has shown that the measure of subjective organization is positively correlated with the number of words recalled. By subjective organization, Tulving means clustering. One of the most interesting findings of this research is that the order of presentation is not a good predictor of the recall order. There is a tendency for 56 items which are related to one another to be recalled to- gether even though the words were not contiguous during presentation. In single trial recall there is a tendency for subjects to recall late input items first, regardless of meaning (Bousfield, 1953). Also, there is a tendency for clustering to be the greatest in the middle output period. In attempting to explain this phenomenon, Shuell has written (1969:353), This discrepency between the order in which the items were presented and the order in which they were recalled is presumed to represent a tendency on the part Of the subject to organize his re- call on the basis of various second-order habits, that is, preexperimental associ- ations or conceptual relationship. This tendency for related items to be recalled together has been termed clustering. Clustering or subjective organization has been taken to indicate the notion of categorical arrangement of meaning. Each item within a given subset or cluster is assumed to be more similar in meaning to other words within the cluster than it is to any other item in other subsets. The measures of clustering in recall are re- viewed in Tulving (1968). They are derived from the assumption that if the items in the total list are pre- sented in random order, organization or clustering occurs when the items from a subset are recalled in immediately adjacent output positions more frequently than one would expect by chance. Thus structure in free recall is measured in terms of a series of probability statements. 57 Clustering has been shown to be a function of the number of repetitions of the material (Bousfield, Puff and Cowan, 196”; Tulving, 1962), strength of assoc- iative relations among words (Jenkins and Russell, 1952), number of categories presented in the material (G. A. Miller, 1956; Handler, 1967), and the developmental level of the subjects (Jablonski, 197”). Reviews of the re- search using free recall as an indicator of the structure of associative meaning have been presented by Cofer (1965), Tulving (1968) and Shuell (1969). They conclude that the words cluster due to their similarity in meaning. How- ever, rather than discussing these conclusions in light of an association theOry, they have used the results to justify a notion of organization by categories, or what Tulving (1968) calls the principle of superordination. It should be noted that clustering will occur in the free recall if either method of organization (similarity or superordination) is used. Thus Cofer (1965:271) concludes ". . . such a contrast is neither useful or heuristic. In free recall, our evidence suggests sub- jects will find ways to organize recall even though the experimenter has not provided means in the list he pre- sents." Since the method of organization used by the subject cannot be differentiated, this author feels that the notion of superordination be dropped and that only the principle of similarity be used. It is mathematically more powerful and provides a more holistic view of the structure of meaning. 58 Free recall provides a marked improvement over free association as a measure of meaning. This is due in part to its ability to sense intermediate degrees Of structure rather than the nominal identification of association. In addition, because it prespecifies the domain of concepts to be recalled, it is much more struc- tured and therefore a more reliable measure. However, it is still inadequate as a measure of the structure of meaning. The reason is that the measure is ordinal and therefore unable to provide an accurate configuration of association and the change in structure over time. As a result of these two problems, the method is not empiri- cally fruitful because it cannot provide accurate pre- dictions. The results from a free recall trial produces a vector with each stimulus ranked according to its order of recall. Consensus is not a problem because given enough cases, the average order of recall can be deter- mined. Clusters are next determined based upon these rank orders and a summary probability statement deter- mined to describe the departure from randomness or the associative structure present in the list. Also order of presentation confounds recall order and therefore structure. Although controllable, it requires considera- ble expense to do so. Thus, while one can determine if associative structure exists and even the degree to which structure occurs, it cannot measure the strength 59 of the relationship between individual pairs of concepts. One only knows their frequency of occurrence in pairs not the degree Of similarity between the items. One can only say that symbol "X" occurs more frequently with "Y" than it does with "Z." From this similarity must be inferred. Because the items are only ordinally scaled, change as a function of time cannot be determined. Suppose recall trials are performed at two points in time with an intervening message designed to alter the relations among the stimuli words. Recall measures can only determine if a change in structure did occur. Concept "X" no long- er clusters with "Y,"'but now with "Z." The degree to which the meaning changed cannot be calculated. Also, the measure is too insensitive to change within a cluster. It is conceivable that a change might go undetected if a cluster were strongly associated at the first trial. It is doubtful that concepts will disintegrate past associations due to some manipulation. However, this is not to say that some change in the strength Of the assoc— iation was not altered. The ordinal measure is incapable of determining this change. Due to this lack of sensitivity to change, free recall cannot be used as an accurate pre- dictive measure. Thus, it is not empirically fruitful. The Semantic Differential Developed by Charles Osgood and his associates 60 explicitly as a measure of meaning in the 1950's, the se- mantic differential scale has obtained wide acceptance as a measurement technique in the fields of psychology and communication. The development and the early history of the method is reviewed by Osgood, 33.211’ (1957). It was developed in conjunction with Osgood's mediational theory of meaning. Thus, it was intended to measure the strength of associations or the degree of similarity or difference between individual words at one point in time. These relationships were considered contingent on an indi- vidual's past experience. They postulated that meaning can be described in terms of a semantic space of unknown dimensionality and Euclidean in nature. Each semantic scale, defined by a pair of polar adjectives, was assumed to represent a linear function that passes through the origin of this space, and a sample of such scales comprised a multidimensional space. The more bipolar scales that are used, the better the definition Of the space. However, to describe this space with maximum efficiency, one needs to determine the minimum number of orthogonal dimensions which describes the space. "Difference in the meaning of two concepts is then merely a function of the differences in their re- spective allocations within the same space, i.e., it is a function of the multidimensional distance between two points" (1957:26). The individual concept's locus in the space has two prOperties--direction from the origin, and 61 distance from the origin. These may be identified as the quality and intensity of meaning. Another way to describe these two properties is, "direction in the space being equated to what mediators are evoked and distance from the origin being equated to how intensely these are evoked" (1957:29-30). The method operates in the following manner. On a paper and pencil instrument, a series of bipolar ad? jectives are presented along with single word. The dis- tance between the ends of each scale is broken into seven supposedly equal intervals. The subjects then rate a series of words on these set of scales. A typical seman- tic differential scale looks like this: Father Happy : : : : : Sad Hard : : : : : : Soft Slow : : : : : : Fast Once the data is gathered, it generates a three- dimentional matrix of subject x scales x concepts. Sub- jects are summed, generating a single, square matrix Of the scales for each concept. This matrix is then inter- correlated and factor analyzed revealing the dimensions of the semantic space. The three most prominent dimensions which have been found are evaluation, potency and activity. Osgood (197”) reports research into the semantic structure of some 27 different language-cultural communities. The 62 results produced loadings in the .8 to .9 range on the evaluative dimension, loadings from .” to .7 for the potency factor and for the activity dimension, .3 to .7. From these results, he concludes, "This is rather convincing evidence for the universality of the affective meaning system" (33- 3”). He concludes that evaluation, potency and activity are universal dimensions used by all people to define the connative meaning Of a word. Just as Osgood's mediational theory was a sign- ificant improvement over behavioral associationism, so is the semantic differential scale an advancement over free recall and free association. The method is very reliable, and does not have the intervening variable Of being attached to a serial learning task. Additionally, it was designed to measure meaning rather than learning or psychological states. This is accomplished in terms of a multidimensional space, which allows the various attributes which determine meaning to be revealed. Finally, the differences in meaning among words can be described in terms Of their similarity or distance from one another. However, there are a number of serious problems which render Osgood's semantic differential as less than ideal as a measure of meaning. At the scale's very foundation rests the assumption that the semantic space can be generated from a series of bipolar adjectives. Osgood asks that we intuitively accept these adjectives as being Opposite in meaning. Wishner (1960:110) reports, 63 however, that results, "indicate that grammatical antonyms do not necessarily correspond to psychological opposites." Danes and Woelfel (1975) present evidence that the bipolar descriptors used in the semantic differential are in fact 293 bipolar. Using metric multidimensional scaling, they determined that "good" was ”5.39 units from the origin and "bad" 57.1”; "favorable," 66.32 and "un- favorable," 51.98; "positive," 51.73, and "negative," 57.68. The differences in separation from the origin are substantial. Thus, rather than the concepts being polars, they are, in fact, far from it. The angles between these terms were next determined. If the concepts were truly opposites, the angle between them would be 180° and the cosine -l.0. The Obtained cosine for good-bad = -.98 and its corresponding degrees = 168°; the obtained cosine for positive-negative = -.7” and its corresponding degrees = 152°; and the cosine Obtained for favorable-unfavorable = -.88 and its corresponding angle = 138°. Thus, the assumptions upon which the semantic differential are tenuous. Similar findings are reported by Anderson (1970). Another consequence of the use Of the bipolar opposites to generate the dimensions is bounded semantic space. The subjects are forced to respond with a very limited range, +3 to -3. This has the effect of limiting the Variance between responses. Thus, it appears as if the language is more homogeneous than might be otherwise. Because the space is bounded, it limits the instrument's 51+ use as a dynamic measure. If the members Of a culture evaluate a particular object very strong (+ or -3) at one point in time and later the symbol becomes increasing- ly positive or negative, the semantic differential will be incapable of showing that change. This is called the floor and ceiling effect. Because the range is so small, only seven points, this shortcoming becomes a major problem. The semantic differential is only an ordinal scale. As with the free recall procedures this greatly limits its utility as an indicator Of change. One can only say that symbol "X" became more or less on any Of the three dimensions without any indication of how much more or less. It then becomes impOssible to discuss the rate at which a concept is changing. As is the case of all factor analytical solu- tions, the dimensions which emerge are a function of the variables which are input into the analysis. This problem is compounded because the semantic differential is generated with a series of scales bounded with gram- matically bipolar opposite adjectives. Subjects are forced to describe the Objects on the scales provided which may or may not be the dimensions and individual used to eval- uate the symbol. Because Osgood (197”) always uses the same (or functionally translated equivalents) set of ad- jectives, it is no wonder that the same set of dimensions always emerge with varying degrees of explained variance. 65 It would cause some concern if this did not occur. The instrument has the further property of being a discontinuous space. It goes from -3 to +3 with some mysterious properties occurring at the zero point. Osgood describes the origin, the point where all three dimen- sions are zero, as follows: "Like all self-respecting spaces, this one has a central origin--a locus of com- plete meaningless analogous to the neutral grey center of the color space" (197”:22). It seems that a concept at this point has connotative meaning but not on the three dimensions Osgood suggests. Other dimensions may be used by the subject to discriminate the object. Finally, the semantic differential breaks down the total meaning into a large number of separate quali- tative aspects, rather than integrating the dimensions into a holistic description. The reason for this is because with this method the stimuli are presented to the subjects one at a time. This leads to the stimuli being evaluated in isolation without benefit of the other Ob- jects which are to be scaled. Thus, rather than describ- ing the interrelations among a set Of lexical items, the semantic differential only locates the symbol along a series of bounded scales, which when factor analyzed can be described on three orthogonal dimensions. In summary, despite the apparent generality of Osgood's affective semantic space as a theoretical and methodological device, there are serious shortcomings 66 which render it less than optimal for the study of meaning. First, it does not directly measure the relations among a set of lexical items. ‘Second, the scale upon which the words are rated are not bipolar and only ordinal. These problems make it an inaccurate measure of change. Because it cannot accurately measure change, it becomes not empirically fruitful due to its inability to make accurate predictions. Multidimensional Scaling (M.D.S.) The logic behind M.D.S. has been described by Helm, Messick and Tucker (l959:l”): . . . the fundamental concept in multi- dimensional scaling is psychological dis- tance, which is usually estimated in terms of judgments of similarity among stimuli; i.e., two stimuli judged to be very similar are considered to be psychologically closer together than two stimuli judged to be very different. Given judgments of similarity among all the stimuli in a set, mathemati- cal models exist which provide an inter- pretation of this psychological distance in terms of Euclidean geometry. The stimuli are treated as points in a Euclidean space, and analytical techniques are available to obtain the dimensional- ity of the space as well as stimulus scale values determined within a rotation and translation. Multidimensional scaling models can be used in situations where the stimuli may vary simultaneously with respect to several underlying dimensions or attri- bute as in the case of the meaning Of lexical items.12 Some attributes are unidimensional, among them, weight and width. Spatial position on the surface Of the earth 67 is also measured with three dimensions; altitude, longi- tude and latitude. Knowledge of the position of an ob- ject along any one or two dimensions will not locate it precisely in the space. All three dimensions must be known. Likewise, color is said to possess several underlying qualities. The color green can vary simultaneously accord- ing to hue, chroma and any other of several different qualities. According to Torgerson (l958:2”8): . . . the notion of a single unidimensional, underlying continuum is replaced by the no- tion Of an underlying multidimensional space. Instead of considering the stimuli to be represented by points along a single dimen- sion, the stimuli are represented by points in a space of several dimensions. Instead of a551gning a single number (scale value) to represent the position of the point along the dimension, as many numbers are assigned to each stimulus as there are independent dimensions in the relevant multidimensional space. Each number corre- 8ponds to the projections (scale value) Of the points on one of the axes (dimensions) of the space. The process is analogous to converting a matrix Of city to city mileages to a graphic representation such as the map itself. In this special case, an n x n matrix of cities would be reduced to a 2-dimensional configura- tion. While the similarity structure among a set of lexical symbols can be accurately described by a dissi- milarity matrix, it is extremely cumbersome due to its size. The matrix is order n, where n equals the numbers of concepts described. N-l is the maximum total number of 68 dimensions used by the sum of the individuals of a society to differentiate the objects, not the ones shared by the members of social system.13 This matrix may be reduced to usable proportions and the uniquely shared underlying cultural dimensions identified.11+ This task can be per- formed by multidimensional scaling. The development of M.D.S. has been described in great depth by Serota (197”:””-51). M.D.S. as a psycho- logical measurement technique can be attributed primarily to the work of Torgerson (1951, 1958). Serota points out, (197”:2”) "It also draws heavily on the theoretical con- struction Of Gulliksen (19”6) and Thurstone (1927), and the mathematic contributions of Hotelling (1933), Young and Householder (1939), and Garnett (1919)." Since Torgerson, there has been one major adjustment in the technique. That was the development of non-metric M.D.S. by Shepard (1962) and Kruskal (196”). The mathematics of the procedure by which an underlying vector space (the number of independent dimen- sions and the concepts scaled values on these dimensions) is as follows. A dissimilarity matrix is gathered from a group of peOple, yielding a three dimensional concepts x concepts x subjects matrix. There are a number of ways to generate this matrix and they will be discussed later. This matrix is averaged across the n_persons into a two dimensional concepts x concepts square symmetric matrix S where any entry Si' represents the average distance 3 69 between concepts i and i as seen by the observers. This matrix S is transformed into a scalar products matrix B (Young and Householder, 1939), by premultiplying it by its transpose. However, it is generally the practice to "double-center" this matrix by establishing an origin for the space at the centroid of the distribution. This can be done simply during the construction of the scalar pro- ducts matrix, and the transformation for any cell bij is given by the equation n n n n .2 d2ij 2 d2ij 2 2 d2ij - d2ij b--=1/2°.-_=.].:___+l'-_'_J:_____1=1 J=l ) l] n n n2 which is a straight-forward linear transformation that sacrifices none of the information present in the original matrix D (Torgerson, 1958). This new centroid scalar products matrix is such that any entry: pi = the length of vector i bij = pipjcosaij where pj = the length of vector i Gij = the angle between i and 1. Consequently, when matrix B is reduced to its base by routine factorization (i.e., the application Of any standard eigen routine, such as principal axis or Jacobi), the result is a factor matrix F, whose columns F1, F2, ... Fk are orthogonal vectors with their origin at the cen- troid of the vector space spanned by F and where any entry Fij represents the projection (loading) of the ith 70 variable on the jth factor. This resultant matrix pro- vides an accurate description of the structure possessed by a set of concepts at any one point in time. However, as mentioned above, there are a number of methods that one can use to generate the original dissimilarity matrix. The particular method chosen affects the ability of the technique to measure change in structure over time and the quality of the structural description. Nonmetric.--There have been a number of ways to generate the basic dissimilarity matrix. Among them are sorting, triad combinations, and direct pair compari- son, either bounded Or unbounded. G. A. Miller (1969) had subjects sort ”8 English nouns which are printed on 3 x 5 index cards. The subjects were instructed to sort the cards "on the basis of similarity of meaning." The subject was allowed to put them into piles which Miller took to represent associative clusters. This data was converted into an incidence matrix. The matrix was ”8 x ”8, and cell i, j represented the particular pair of nouns i and j. Cell i,j was assigned a one if that pair Of nouns was put into the same cluster and a zero if they were in separate groups. The matrices, one for each judge, were added together and the resulting matrix was taken as the basis for further analysis. This matrix was symmetrical and could be regarded as a measure of similarity or proximity, where N-- is a measure Of the 1] semantic proximity of noun i to noun j. It can be 71 converted into a distance matrix by using the formula Dij = N-Nij. While this method is quite reliable (Mandler and Pearlstone, 1966), it provides an inaccurate description of the data's structure. The reason is that the distance matrix is taken from the co-occurance of a particular pair within a given cluster. The estimate of the dissim- ilarity of two concepts from different clusters becomes inaccurate because the concepts are treated as if members of separate and discontinuous categories rather than within a single continuous domain. The actual distance between the terms is never determined. In addition, the matrix produces a bounded space with the distance between the terms limited to the number of subjects. Thus, the problems which limit the semantic differential's use as a measure of meaning also apply to M.D.S. with data gen- erated from sorting procedures. The method of triad combinations has been used by Henley (1969) and Szalay and Bryson (197”) to investi- gate the meaning Of individual lexical items. Triad combinations operates in the following manner. Subjects are presented three words and asked to judge which two are most similar and which two are most different. Matrices of interstimulus similarities are obtained for each indi- vidual and converted to dissimilarities by subtracting from a constant. Henley (1969:180) reports reliability coefficients for the individual of .85 using triad 72 combinations. Also, results from this method correlate highly with other methods of data collection. This method is an improvement over sorting because it allows the respondent to directly compare the stimuli on similarity. However, it only allows the respondent to say that one concept is more similar than another, rather than how much more similar. The result is an ordinal scale with its inherent limitations. This method also results in a bounded space which restricts the dissimilarity among concepts and makes the method incapable of measuring change over time. Henley (1969) also had her subjects rate pairs of stimulus words in the following manner. Subjects were presented all possible pairs of words, one pair at a time and were asked to make a judgment of the amount of dissimilarity between the two, on a scale from 0 (no difference) to 10 (maximal difference). A week later she had her subjects made the judgments again, this time with the pairs in Opposite (within pair) order from the first session. The reliability coefficients for 21 subjects ranged from .29 to .85, with a median Of .71. The relia- bility estimates obtained for the mean dissimilarity _matrix was .97, "indicating both that order of presenta- tion of stimuli does not affect reliability, and that 88 showing low reliability in their ratings do not cause low reliability in the overall matrix" (1969:178). The correlation of data Obtained with this method and triad combinations was .90, indicating high agreement between the scaling methods. 73 While this method is an improvement over sorting because it has the subject make direct pair comparisons and the method of triads because it allows for variable similarity between the pairs of concepts, it suffers the drawback of forcing subjects to respond with a restricted range of 11 points. Thus, it cannot be used to measure change over time and may in fact provide a distorted picture of the structure at one point in time. If a sub- ject rates one pair early in the task as a 10 and later comes to a pair that is even more dissimilar, he/she has no alternative but to settle for a rating value of 10 again. In response to this problem, psychometricians Shepard (1962) and Kruskal (196”), have developed what has come to be known as nonmetric multidimensional scaling. These procedures take the Euclidean space resulting from the classical M.D.S. approach of Togerson (1958) and per- form monotonic transformations based upon the rank order of the stimuli on a subset Of the n-l dimensions. If the resultant space is non-metric, the concepts loci are ". . . simply moves the points in such a way as to stretch those distances that are too small and compress those distances that are too large" (Shepard, 1962:128). The result is that the final space is ordinal and cannot be transformed back into the original dissimilarity matrix. This transformation, while providing perhaps a better description of the structure of the data at one point in 7t+ time, renders the method as useless as a dynamic measure. The reason is that the matrix has been destroyed for the description of the structure. This form of M.D.S. has been used in the past to investigate the dimensions underlying the meaning of lexical items (Cliff, Pennell and Young, 1966; Henley, 1969; G. A. Miller, 1969; Fillenbaum and RapOport, 1971), the perception of speech sounds (Miller and Nicely, 1955; Degerman, 1972), and the processing of information (Schroder, Driver and Streufert, 1967; Rips, Shoben and Smith, 1973; Rumelhart and Abrahamson, 1973; Szalay and Bryson, 197”). Metric.——An alternative method of data collec- tion exists which facilitates the study of dynamic pro- cesses. It can be called unbounded direct-pair compari- sons. By using this method of a procedure, it is possible to compare metric M.D.S. structures across time intervals. It Operates in the following manner. Sub- jects are presented a criterion pair which serves as a comparative standard along with the concepts to be scaled. The distance between the criterion pair is speci- fied and the subjects are told to compare the dissimilarity between two concepts in relation to this standard. This can be accomplished by a question worded in the form: "If g and y are 3 units apart, how far apart are 3 and b?" 75 Such an item wording requests a dissimilarities judgment from a respondent (". . . how far apart are 3 and 2?"). An example of this procedure would be, "If RED and WHITE are 100 units apart, how far apart are: Applies and Oranges Apples and Lemons Oranges and Lemons I This is done until the entire n(n-1)/2 pairs have been estimated. The lower triangle of a n x n data matrix is filled. In making the decision as to which pair of con- cepts to choose as the comparative standard, one should follow Woelfel's advise (197”:12), First, the standard should be relatively stable. Changes in the standard over time can confound time series measurements and prevent meaningful comparisons of measure- ments made at different times. Secondly, the standard should be the same for all observers regardless of reference points, i.e., two independent Observers must both agree on the length, for example, of a meter of a kilometer. Less important, but nonetheless worthy of consideration, good practice for minimum error suggests using a standard approximately midway between the largest and smallest measurement likely to be encountered (measurement Of astronomical distances in miles, for example, is cumber- some, as would be measure Of terrestrial distances in fractions of light-years). These criteria, however, are never achieved in any science. NO distance, for example, is truly invariant, no clock emits signals so that ". . . the duration between any two signals is (exactl ) the same. . ." Thirdly, at least Within the framework of relativistic physics, viewers in referent systems moving at differential velocities with regard to one another will not agree 76 on distances or durations of time when view- ing the same events. Whatever consequences failures to meet these criteria exactl may be for philosophy, they are not insuperable barriers to science. Criterion distances which have been used are Red and White are 10 units apart (Barnett, 1972, 197”; Woelfel and Barnett, 197”; Barnett and Wigand, 1975) Dwight D. Eisenhower and John F. Kennedy are 50 units apart (Bar- nett, 23.21;: 197”) and Dwight D. Eisenhower and John F. Kennedy are 10 units apart (Taylor, Barnett and Serota, 1975). According to Woelfel (197”:13) this technique has several key advantages: First and foremost, no restrictions are placed upon the respondent, who may report any posi- tive real value whatever for any pair. Thus the scale is unbounded at the high end and continuous across its entire range. Second- ly, because the unit of measure is always the same (i.e., the unit is provided by the investigator in the conditional, "If x and are 3 units apart," and thus every Ecale unit is l/u units), and because the condi- tion of zero distance represents identity between concepts and is hence a true zero, not at all arbitrary, this scale is what social scientists usually call a ratio scale, which allows the full range of stan- dard arithmetic Operations. Third, since the unit Of measure is provided by the experimenter it is possible to maintain the same unit Of measure from one measurement to another, both across samples and across time periods, which is crucially important since time is one Of the primitive varia- bles of scientific theory. These three characteristics taken together provide the capacity for comparative and time-series analyses at very high levels of precision. While the technique suggested meets the scaling criterion, quite exactly, and in fact will be the technique 77 of choice in the measurement of aggregate cultural patterns, problems of unreliability may make it unsuitable for the measurement of individual's meaning. Since reliability of any scale is approximately proportional to the complexity of the judgmental task required of the respondent, and the technique of direct paired distance estimates requires a highly complex set of judgments from the respondent while providing virtually no structure, it is consequently un- reliable for the individual (typical test-retest relia- bility correlations range in the .70's for individuals). Barnett (1972) and Gillham (1972) have shown that the for- mat is extremely reliable On large samples and that con- sistency of measure increases as a function of sample size. The reason for this is that the error which occurs in measurement is random error rather than systematic bias producing invalidity. Such random error will be normally distributed in a series of measure. The law of large numbers and the central-limit theorem assure that the scores obtained will be normally distributed and that the sample mean will converge on the population mean as the sample size increases. Barnett (1972) reports test- retest reliability coefficients in the range of .90 with 75 caseslwahus} reliable measurement becomes only a function of the cost of gathering additional cases. Once the dissimilarity matrix is generated, the matrix is reduced accOrding to the procedures presented earlier and outlined in Torgerson (1958). The reduced spatial manifold 78 or multidimensional space provides an undistorted picture of the structure at one point in time. Process can be recorded in the following manner. A series of spatial manifolds are generated at separate points in time. They can then be rotated to a least-square best fit congruence in order to calculate the change over time. Recent re- search indicates that a solution exact to within theore- tical assumptions, by rotation to stable criterion exists (Woelfel, Saltiel, McPhee, Danes, Cody, Barnett, A. D. and Serota, 1975). This has the advantage over a simple least-square solution of increasing the stability of the vector space despite the extreme movement of a single concept. This technique is analogous to the heliocen— tric theory Of motion of celestial bodies and its notion of measurement. In this system fixed reference points are established, against which all change is measured. It would Operate by rotating the concepts whose meaning is theoretically assumed to be stable to a least-square best fit and then by simply subtracting the coordinate values for the non-stable concepts to determine their change over time. This change may be expressed as velocities as given by: a: "M (aij - bij)2 D. v.=i=j1 where, 79 Vi = the velocity of concept i di = the distance concept i has moved across the interval of time 3 t = time aj = the coordinate value of concept i on the jig factor Of the t0 space b. = the coordinate value of concept i’on the J jtg factor of the t1 space. Given multiple time periods, the accelerations Of the Objects in space may also be calculated: Or, for instantaneous accelerations, 9.: dt A = These velocities and accelerations, necessary components Of process (Arundale, 1971, 1973) are unmis- takably measures Of cultural change of very high pre- cision. This is so since the culturally shared defini- tion Of any Object is given by its location in the multi- dimensional space, and changes in the terms locus re- present changes in its definition.15 The study Of process was made possible with metric scaling because the data is ratio level and was not al- tered in a nonfunctional manner anywhere in its analysis. Metric scaling provides the investigator Of meaning a system of measurement which meets all four cri- teria presented in the first section of this paper. One, 80 a matrix of the interrelations among a set of lexical items is described, using the dissimilarity among the words as the basis. This relationship is not inferred from the terms association to some separate criterion. The degree of association amongst the items is ratio level and there is no loss of information by forcing the estimates into some prespecified system. Two, the consensual nature of meaning is taken into account at two points in the analysis. First, an average dissimilarity matrix is created. Second, the matrix is reduced through M.D.S. to those dimensions that are shared by the population of language users. Third, because the data generated are not bounded or ordinal, the dynamic nature of meaning may be taken into account. Finally, and perhaps most importantly, this system has demonstrated its use as a predictive tool. Taylor §£_§l; (1975) used this system of measurement to predict change in political attitudes during an election campaign with extreme accuracy. Marlier (197”) found this method very accurate for a test of social judgment theory. Summary In summary, this chapter has proposed that a fundamental system of the measurement meet the following four requirements. This system must take into account the degree Of relationship among a set of lexical items. It must be consensual and dynamic, and finally, be 81 empirical fruitful. Next, four general systems for measur- ing meaning were described and critiqued. Free associa- tion, free recall, the semantic differential and non- metric multidimensional scaling were rejected for failure on one or more criteria. Finally, a modified version of classical multidimensional scaling was suggested because of is acceptability on all four criteria. 82 FOOTNOTES FOR CHAPTER TWO 1The enormous and varied literature on theories of meaning has been surveyed in great detail by Haney (1970). The author does not intend to review the entire literature, but to attempt to reconcile some of the con- flict in the area for the science of communication. 2This controversy manifested itself in an ex- change of articles by Hurbert Blumer and Robert F. Bales in the American Journal of Sociology,March, 1966. For an example of research using the individualistic perspective see Goffman (1959) and for an example of the social see Haller and Woelfel with Fink (1968). 3While this position is quite acceptable to philOSOphers interested in metaphysical problems and ones who deny the notion of a shared or perceived reality, it is unacceptable to scientists who consider agreement of this reality as an epistimological requirement. “This is an Obvious oversimplification. The author acknowledges that other theoretical perspectives have been evoked by American psychology. The example is intended to show one example of psychological research in the area of language with the potential of significant findings for the notion of meaning which ignore the or- ganization of lexical items used in the research. 5For Frege this proposition must be in the form of a sentence. Words have sense (sinn) as opposed to reference (bedeutung) only if the conditions necessary to fulfill the truth—value of the proposition are met. False propositions, although conceptually possible, are meaning- less. 6For an indepth discussion Of Durkheim's notion of collective representations see Gillham (1972). 7The author does not suggest that this is in anyway an ideal definition of communication. It is only intended to serve as an example of the field's awareness of the notion of process. It could be improved by chang- ing the words "transmission" to "exchange" and "parts" to "components." In this way, it would take into account the transactive nature of communication and the simul- taneous exchange of information. Other definitions of communication which share the notion of process are Berlo (1960) and G. R. Miller (1966). 83 8It seems reasonable to assume that the dissi- miliarity between the Objects is reduced when an associa— tion is formed, although this point is not made explicit. 9If the time intervals between measurements are unequal, but they are known, the same information may be gained. However, a more complicated analysis is required. This problem is discussed in depth by Coleman (1968:”37). 10This is not to take the position that these four are the only ways meaning can be measured, but only a convenient grouping under which most systems can be sub- sumed. ' 11The words were: moth, insect, wing, bird, fly, yellow, flower, bug, spring, cocoon, color, blue, bees, summer, sunshine, garden, sky, nature and butterfly. 12The author does not mean to imply that the notion of dimension and attributes are isomorphic. Attribute refers only to the systematic alignment Of concepts where some property which they have commonly in varying degrees can be identified. Dimension refers 'Only to the orthonomal reference vector which results from the mathematical procedures of orthogonal decomposition. 13Any two points (Objects) may be connected by a line, yielding a single dimension differentiating the Objects. Three objects may be connected by a plane. No information as to their differentiation would be lost by indicating the Objects' scale values on the two dimen- sions. The same holds for four points in a cube (three dimensions) and in points in a hypersphere of n-l di- mensions. It should be noted, however, that if any three or more points lie along a continuum, fewer dimensions would be needed to precisely describe the system. ll'Recent research by Joseph Woelfel suggeststhat the loss of information by using a space Of reduced di- mensionality is too great to warrant its use. He feels that all n—l dimensions have meaning. The present author supports the liberal position and feels the reduced space is better to work with, because the ideosyncratic re- sponses and the unreliability are removed. 15A Fortran IV computer program which accom- plishes the calculations described in this paper is available at Michigan State University. It is know as Galileo Version 3 (Serota, 197”). CHAPTER THREE BILINGUAL INFORMATION PROCESSING -- THE EFFECTS OF COMMUNICATION ON SEMANTIC STRUCTURE Overview This chapter will describe a single substantive problem--the organization of semantic information by hi- lingual individuals. Bilinguals have been the object of intensive investigation by the social sciences. They have been studied from the psycholinguistic perspective which poses the general question, "How do bilinguals organize their language system?" It is hoped that by examining this phenomenon insights might be gained which would help psycho- logists and linguists describe the processing of language by people in general. Researchers in this area have not been overly successful in agreeing upon a description of the way in which the bilingual organizes his/her meaning system. One reason for this failure as explained in the last chapter has been the lack of a consistent theory of meaning and a measurement system derivable from such a conceptualization. The preceding chapter suggests that people organize their language system according to the meaning of the individual lexical items. Meaning may be taken as a series of fundamental measures based upon four principles, the most important of which is similarity. Bilinguals have also been looked at by socio- linguists, who are concerned principally with how 8” 85 contextual variables in a bilingual's environment affect the language that is used in any given situation. The previous chapter made explicit the point that the indivi- dual language-user is embedded within a social environment which affects the way in which he/she processes semantic information. Also, the individual's system of organizing meaning has been viewed as contingent upon his/her past experiences. These two notions have been ignored by psychologists interested in the bilingual, but have been investigated by sociolinguists. Based upon these two separate lines of research and the paradigm for the measurement of meaning described above, this chapter will conclude with a series of theoretical hypotheses which will apply this conceptualization to bilingual indivi- duals in order to attribute validity to the general para- digm, describe the bilingual's linguistic system and how the semantic structure is affected by environmental in- fluences such as the mass media and interpersonal rela- tions. These hypotheses will be Operationalized and a study designed to test them in Chapter ”. Social Significance of the Bilingual Today, throughout the developing world, there exists a communication problem of some magnitude. Typi- cally, people speak only one language in a nation-state where many languages may be spoken by several indigenous cultural groups. In India, for example, there are over 86 17 different languages apoken. They are as diverse as the IndO-EurOpean languages, Hindi and Bengali, and the Dravidians, Kannadi and Tamil. Because the official na- tional language is Hindi, people from other regions of the subcontinent are obligated to learn that language if they are to take an active part in the national political pro— cess. They are forced to become bilingual if they are to be integrated into the nation-state. Not having a common language often results in conflict between the cultural groups and may ultimately set the stage for the disintegration Of a country. The independence Of Bengla- desh resulted in part from the differences over linguistic policy between the Bengalis and the Urdu speaking Pakistanis. The problem does not end here. Scientific and technical material and the education of these subjects are mainly in western languages, usually French, English or German. If a society is going to industrialize and therefore achieve a degree of equity with the developed nations, this knowledge must be disseminated. However, before the information can be diffused must come the veh- icle whichcarries the knowledge. This vehicle is in- herently a second language. Thus, bilingualism must become a necessary state of affairs in the world of the future. To a certain extent, this problem can also be found in the United States. According to the 1970 census, there are ”2.5 million people in this country who speak a 87 language other than English. These people may be bilingual (English and some other language) or monolinguals in some language other than English. This amounts to a staggering 20.9 percent of our p0pu1ation. The largest of these groups, Spanish speakers, amount to about eight million or 18 percent of the non—English speakers. They are con- centrated in southern Florida, the Southwest and urban barrios. This communication problem also applies to a lesser degree with Black Americans. This difference is not one of language, but one Of the difference between dia- lects. By and large, these people are poor and uneduca- ted and trapped in a cycle of poverty. If they are to solve their social problems, certain information must be Obtained. But where do people go for information, when they do not know the language in which the possible solu- tions are written--English? Education, formal or informal, is harder to obtain when the information necessary to alter one's living conditions are not available in a language one can understand. Most Often the non-English speaker is a recent immigrant to the United States. In the case Of Spanish speakers, they have come from Mexico, Puerto Rico or Cuba. They have left their homes in order to better their quality Of life, similar in their potential to the Europeans that arrived earlier in this century. While wanting to improve their standard of living, they want to 88 save certain aspects of their indigenous culture. It is this melting pot notion--the input Of ideas, practices and artifacts from many different societies that has made this country the technical innovation center of the world. There are two potential solutions to the problem. of distributing the necessary information to non-English speakers. One, translate the knowledge into the other language, and two, teach these peOple English. Both solu- tions have been tried. The key idea to notice is that both processes require a bilingual to translate a message from one language to another. Exact ideas must be translated from one meaning system to another. The second language may have a profoundly different semantic structure, such that the wrong symbol may produce adverse behavioral associations in a member of the seconed culture. Thus, the process by which information is trans- ferred between cultures becomes critically important. An accurate psychological description of how bilinguals pro- cess information may facilitate the exchange of ideas across all cultural boundaries. This would have the advantage over some universal "Esperanto" because it also preserves the native 1anguage--one of the defining attributes Of culture. By understanding the processing of information across language barriers, we can avoid po- tential misunderstandings which Often develop into con- flict. Accurate transfer of ideas interculturally may 89 provide certain necessary mechanisms for the improved quality of life for mankind. The bilingual plays an important role in the process of acculturation. As the language of an immigrant group shifts from the one used in their native country to another used in their new home there is usually a per- iod of time when the immigrant group may become a bi- lingual speech community. This usually takes as long as a generation to develop and lasts only an equal period of time, provided that the immigrants settle in close prox- imity. Lieberson (1970) describes this phenomenon in Canada with the change from French to English as the spoken language in certain regions and the integration into the English language society with later immigrants. The bilingual individual may be considered as a link between two different cultural groups. These groups can be defined by their respective code systems, which are made up of both nonverbal codes, such as the artifacts of dress, or the meaning attributed to time and spatial relations, and the linguistic codes. Language has historically been one of the ways in which cultural entities have been identified. Thus, by definition, the bilingual is integrated into two different cultures. What are the social psychological effects of being integrated into two different cultures simultaneous- ly? Durkheim (1951) might suggest that these people would have a higher degree Of anomie than monolinguals 90 because they might have a problem as to which culture's set Of communication or linguistic rules to use. Children bilingual in French and English have been observed to use the English lexicon combined with the French syntax to produce a language unintelligible to anyone but the children themselves (Lambert, 1972). Indeed, this problem may extend to 311 the normative patterns of the bilingual's social environment. Thus, the bilingual is not completely integrated into either group; nor may he/she be capable because of his/her distinctive linguistic system. Alternatively, Simmel (1950) might take the position that bilinguals because of their abilities to communicate with two different code systems are in some sense strangers to the social system in which they reside. Due to their psychological separation from the normative structure of society, they may be more Objective than monolinguals and thus able to recognize the linguistic rules of both sectors of the social system. As a result, rather than having a greater degree of anomie, they are better able to incorporate both sets of norms. Past Research This section of the chapter will review the past research on bilinguals in psycholinguistics and sociolinguistics. It will show that both lines of research must be taken into account simultaneously in order to pre- dict and explain the bilingual's communication behavior. 91 Psycholinguistics.—-One theoretical problem which psycholinguistics has been concerned with is how language is stored. Since an infinite number of combinations of lexical items may be produced, each with a different mean- ing, theorists have proposed notions of economy of stor- age. They have asked the question, what is the most parsimonious way to store linguistic information (Chomsky, 1957, 1965; Watt, 1970)? The last chapter discussed similarity of meaning as the basic organizational principles of language. Be- cause the bilingual processes two different code systems, one for each language, psychologists have focused upon these individuals in an attempt to gain insight into this problem. Are the bilingual's language systems stored separately or is semantic information organized as one system with the linguistic utterance attached as a label or tag at some later point in the communication process? A related question has been how bilinguals organize their lexical information. DO they use a word's semantic con- tent, its language, or both methods of organization? Kolers (1963) performed a thematic analysis of free association data elicited with each of a bilin- gual's two languages. The subjects were German-English, Spanish-English or Thai-English bilinguals. His assumption was that ". . . if verbally defined past experiences were tagged and stored in a form Specific to the language the individual used to define 92 the experience to himself, a bilingual would have a different story of experiences to refer to for each of his two languages, or would be required to tag a given experi- ence multiply, once in each language." (1963: 291) Only about one-third of the responses in one language translated those in the other, and the proportion did not differ whether the associations were intralingual or interlingual. Of this proportion, two-thirds were lexi- cally similar or translations in the interlingual test. The degree of similar associations changed sharply with the semantic category. Concrete referents produced more similar responses than abstract states or emotions. Kolers interpreted the data to mean that experiences were not stored in common in some superlinguistic form but are tagged and stored separately in the language the subject used to define the experience. In a later study, Kolers (l965)presented French- English bilingual subjects with lists of words in a free recall situation. On some lists the words appeared in red or black; on other lists in French and English. On mixed lists, words appeared in two colors or were trans- lated. The main finding was that only about half as many words were recalled from the list with respect to color as were recalled from the linguistically mixed list. These results were taken to suggest a single processing system rather than two separate ones for bilinguals. Kolers (1966), based upon the finding that the 93 probability of recalling a word increases monotonically with its frequency of occurrence, had bilingual subjects recall linguistically mixed lists. He found that the probability of recalling a word when it and its trans- lation are presented n/2 times in each of the bilingual's language (French and English) is approximately equal to its unilingual presentation n times. Since the words in the two languages are usually phonetically and visually distinct, these results may be taken to suggest a single conceptual system that permits this facilitation. McNamara (1967) found that bilingual subjects read word lists in which the languages alternate systema- tically faster than lists that alternate randomly. This may be taken as evidence for the position of two separate storage systems. The mere knowledge of which language a word will be in helps the subjects. This implies a cer- tain degree of functional separation between the languages. Riegel, Ramsey and Riegel (1967) also found support for the notion of two separate systems in bi- linguals. They compared conceptual-semantic structures of Spanish-English bilinguals by observing the overlap of responses produced by a series of restricted associations and found that distinctions made in the second language were less clear than those of the native speakers. They attributed the differences to the separate learning contexts of the languages. Dalrymple-Alford (1967) used cued and uncued 9” verbal stimuli and speed of recall to gain evidence to suggest a single storage system. He assumed that the speed with which a bilingual identifies a word when it is pre- sented should be faster if he/she is cued with its language, given that two separate systems exist. A single system model would predict no difference in reaction time between cued and uncued stimuli because no switching is involved. The speed with which bilinguals identified Arabic and English words was not affected by their being cued for the apprOpriate language. Young and Saegert (1966) and Young and Webber (1967) investigated transfer from English to Spanish with bilingual individuals. For half of the subjects, a positive transfer paradigm A-B, A'-B' was employed. A' is the second language equivalent of A, and B', the second language equivalent Of B. For the second half of the sub- jects an A-B, A'-B'r paradigm of negative transfer was used. That is, associations were learned in one language and then a second list composed of translations of the first list words was learned. In this case, the pairings of the translated words were not consistent with their pairings in the first list. Rather, the translated responses were paired with some stimulus other than the one translated from the first list. The results from both studies indicate that associations formed between the items of one list influence the learning of a sub- sequent list even though the second list is in another 95 language. Twenty-eight percent positive transfer occurred in the second study. These results may be taken to indi- cate a single semantic system. Young and Navar (1968) also employed a positive transfer paradigm with Spanish- English bilinguals and ". . . demonstrated that learning of associations in one language is related to the for- getting Of associations in a second language and on this basis alone it can be concluded that the languages of a bilingual are interdependent and not independent" (1968: 115). Lopéz, Hicks and Young (197”) studied retroactive inhibition with Spanish-English bilinguals using a posi- tive transfer paradigm. The results support an interde- pendence hypothesis of bilingual organization of memory. Lambert and associates have performed a number of studies with French-English bilinguals in order to determine the way in which they organize linguistic infor- mation. They mark a significant turning point in research with bilinguals because they imply that bilinguals may use both language and semantic content to structure their memory. Lambert, Ignatow and Krauthamer (1968) used clustering in free recall of mixed language lists and found that ". . . for bilinguals, language is an ancillary means Of organizing information in memory. The semantic categories appear as the powerful organizational schema" (1968:213). Results from another study (Nott and Lambert, 1968), indicate that bilinguals recalled fewer words from 96 mixed language lists than unilingual lists. The reason they suggest is that besides the semantic information, the bilingual must store the word's language tag, i.e., one semantic storage system with separate language tags. Segalowitz and Lambert (1969) used reaction time on mixed- language lists containing French and English synonyms of the concepts. Reaction latencies were used as indicators of within- and between-language semantic generalization. Subjects generalized their responses to both within- language and other—language synonyms, and used the semantic properties of words rather than language to determine cate— gory membership. Since bilinguals do generalize across language through meaning, these results may be taken in support of a single semantic system. Tulving and Colotla (1970) report that subjects could not recall as many words from multilingual lists as from unilingual lists. They attribute this finding to the greater difficulty of forming higher-order memory units with multilingual lists. This may be taken as evidence for two separate semantic systems. Dalton (1973) presents contradictory results. She found that Spanish-English bilinguals recalled mixed language lists as well as uni- lingual lists. Saegert, Obermeyer and Kazarian (1973) used an adaption of whole-part paradigm of negative transfer to investigate bilingual free recall. No differences were observed in whole-list learning between unilingual and 97 bilingual mixed lists. Part-list negative transfer was found in two unilingual conditions. Since the bilingual adaption of the whole-part paradigm provides a situation where discriminability between whole and part lists is high, the observation of negative transfer in one bilin- gual condition was taken as support for the position which considers the languages of a bilingual to be interdepen- dent. Dalrymple-Alford and Aamiry (1969) and Kintsch (1970) provide evidence that rather than either language or semantic content being the organizing principle of the mind, both mechanisms operate to structure the bi— lingual's memory. Dalrymple-Alford and Aamiry report two experiments using free recall of blocked bilingual lists. Analysis of clustering suggested that the main principle of organization was in terms Of interlingual/intra- category groupings, rather than either intralingual or intercategory associations. Kintsch had German-English bilinguals perform a continuous recognition memory test.‘ All items were repeated twice, either in the same or the other language. Results show that depending on task de- mands subjects could respond either on the basis of language-specific or general semantic cues. Thus, both mechanisms appear to be operating. Another way which has been used to investigate the ways in which bilinguals organize their code system is through studies of interlingual interference or 98 semantic satiation. In this research the assumption is that if the languages are stored separately, then they would show different degrees of interference or satiation. Jakobovitz and Lambert (1961) studied verbal satiation with French-English bilinguals. Subjects repeated words or their translated equivalents for 15 seconds. Then semantic differential scales were administered to measure change in meaning. The results indicated that bilinguals were not susceptible to variable satiation in their two languages. Preston and Lambert (1969) performed a series of experiments to examine the functional relations be- tween the bilingual's two languages. The question of interest centered around the following problem: Does the activation of one language system make the other language system in0perative? The results demonstrated that balanced bilinguals suffer interlingual interference in the Stroop color-word task. Kintsch and Kintsch (1969) had bilinguals learn eight item paired-associate lists with four English and four German words as stimuli and the digits 1-8 as responses. Four translated word pairs were used as stim- ulus items for experimental lists and unrelated words for the control list. Interlingual interference was observed which was interpreted as a task specific phenomenon. What has been presented is a rather lengthy re- view of the attempts to describe a bilingual's semantic structure. Of the 21 studies reported, six provide 99 evidence for separate storage systems for each language, twelve for one semantic system and three suggest the position that both language and semantic content explain the organization of a bilingual's linguistic system. There are a number of reasons for these contradictory find- ings. First, the subjects varied in their degree of com- petence in each of the languages. Kolers (1965, 1966) and Young and Navar (1968) used students of the second language, while Lambert 23.21; (1968), Nott and Lambert (1968) and Segalowitz and Lambert (1969) used bilinguals who may be considered native speakers of both languages. This is known as the compound-coordinate bilingual dis- tinction. These varieties of bilingualism were first de— scribed by Ervin and Osgood (195”). Couched in terms of Osgood's mediational theory, the distinction between com- pound and coordinate linguistic systems results from the associations formed while learning the languages. If only one set of associations is formed during the pro- cess of learning the two systems for coding meaning, then the person may be labelled a compound bilingual. This type is characteristic of bilingualism acquired by a child growing up in a home where two languages are spoken more or less interchangeably by the same people in the same situation. Neither language becomes dominant. In the case of coordinate bilingualism, a set of linguistic signs and responses appropriate to the other language 100 becomes associated with a somewhat different set of repre- sentations. This type is typical of an individual who learns a "second" language sometime after the first. The underlying distinction theorized according to Ervin and Osgood is summarized in figure two. COMPOUND Language S A +rm ______ +Sm R A Language S B////f \\\\\\* R B B COORDINATE S A —-—-—+ I'ml ————+ Sml —--—> R A RB S B ——-——-+rm2 ———-+ sz —————> Figure,2.--The Compound-Coordinate Distinction Using age and context of learning the second language as the criteria (less than age six, compound; greater than six, coordinate; school, coordinate and pre- school, compound) Jakobovitz and Lambert (1969) found that compound bilinguals were less able to switch languages than coordinates in their study Of semantic satiation. Earle (1967) studying Chinese-English bilinguals with semantic differential scales found differences in conno- tative meaning between compounds and coordinates. Lambert and Rawlings (1969) and Segalowitz and Lambert (1969) also 101 found distinctive differences between these two types. In the former study, compound and coordinate bilinguals, equally skilled in French and English, were compared for their ability to search out "core concepts," such as table, when given mixed language clues, such as chaise, food, desk, bois, and manger. Coordinates made more errors and were generally less successful in searching out the core concepts. The second study found that the seman- tic prOperties of test words played a more important role in organization of recall for coordinates than for com- pound bilinguals. The second reason why inconclusive results are reported deals with the methodologies employed. At least part of the confusion may be due to the number of different techniques used and the variety Of ways in which these methods were applied. Also, the methods themselves may be inadequate to answer the question which the re- searchers posed. Free recall was used in a number of studies, free or constrained association in others, and the semantic differential in two studies. These measures may provide a distorted picture of semantic structure and may lead to erroneous conclusions. Additionally, the methods employed were all unidimensional. The phen- omenon under investigation may in fact be multidimen- sional. At least two factors, semantic content and language, have been suggested. Indeed, semantic content has been shOwn to be a multidimensional construct (G. A. 102 Miller, 1969; Henley, 1969; Szalzy and Bryson, 197”). If both Of these mechanisms Operate as Lambert, Kintsch and Dalrymple-Alford suggest, then the methods employed are inadequate to answer the posed question. Third, an important variable, acoustic related- ness, was not controlled for in these studies. In free recall situations, words which sound similar and are not related in free association have been found to cluster (Fagan, 1969; Bousfield and Wickland, 1969). This may be of significance when critiquing the work of Young and associates with Spanish-English bilinguals and Lambert and his researchers in French and English. While these languages make considerable use Of rhyme, English does not. In the romance languages, the gender of the noun determines its final phoneme and the ending of any rela- ted adjectives, such that the words will rhyme. For example, in Spanish pretty girl is translated as chica bonita and pretty boy, chico bonito. Thus, a possible intervening variable may be the gender of the words used in the recall studies. Finally, the particular problem posed may, in fact, be unanswerable. Cofer (1965) noted that the fact of whether individuals organize their recall either by the principle of superordination or similarity of seman- tic content cannot be determined. In the case of bilin- guals, language may be taken to be the superordinate mechanism. 103 It is the author's belief that both schemas operate and that both may be viewed as particular cases of a more general variable, similarity. Using cognitive similarity as the principle Of organization, one can consider two words in the same language as more similar on that particular attribute. Likewise, semantic content may be described in terms of similarity on the attributes used by the bilinguals to differentiate the terms. Sociolinguistics.--Commenting on the work of the psycholinguists, Fishman writes (1968:22), . . . The fact is that psychological studies of bilingualism have neither yielded an explicit model of bilingual functioning nor have they revealed the same componential sophistication as has shown in connection with psychological explanation of other behaviors. The reason for this that Fishman suggests is that the psychological perspective is context free. It does not take into account that the bilingual is a member of a speech community. Fishman writes (1969a:152): Among the major messages of sociolin- guistics is that which states that the individual should be viewed as a member of a speech community. A speech community is characterized by definite norms of language and behavior. These norms not only encompass the varieties or languages that exist within the speech community for its own internal communication needs but also relate them to the types of other than speech behaviors (e.g., the interactions, the mutual rights and Obligations, the roles and statutes, the purposes and identifica- tion), in which various networks within the community are engaged. 10” As has been demonstrated, language is regulated by an individual's long-term storage system. Cicourel (1973) takes the position that the structure Of the long- term stores are equivalent to socially distributed know- ledge. Thus, when an individual interpretes the meaning of a lexical item, he is making use of society's general normative rules. Their use is however, dependent upon the social setting or communication environment in which the language user finds himself/herself. The meaning of a message can be considered context bound, but it is con- text that enables the individual to draw upon certain long-term stores. Thus, the particulars of an inter- action setting can be subsumed under the more general set of normative rules. Cicourel (1973:”6) writes, Interpretive procedures are always Opera- tive within, or in reference to social settings, and their necessary use in making norms recognizable and relevant in particular and general cases means that semantic issues are not indepen- dent of syntatic, phonological and eco- logical features, Or of situated body movements and gestures. Further, the properties making up interpretive pro- cedures are not hypothetical, but can be derived from behavioral manipulations Of socially organized settings. By placing the bilingual in a communication en- vironment, sociolinguists have been able to describe the interaction within a community. Their primary question has been, "what are the different varieties of language and who uses them and when?" A research team headed by 105 Fishman studied language behavior in a Puerto Rican neigh- borhood in Jersey City, New Jersey. Four hundred thirty- one people of Puerto Rican birth or extraction constituted the target or core population. The researchers lived there for four months. The results of this comprehen- sive investigation can be found in Bilingualism in the Barrio, edited by Fishman, Cooper and Ma (1971). Among the conclusions of this research was: that the particular language used (English or Spanish) was determined by con- text. Edelman (1969) found that children used more Spanish when talking to other bilinguals in the contexts of family and neighborhood than they did in educational or religious situations. Findling (1969) found that Spanish was used less in situations which expressed future orientation and need affiliation. Greenfield and Fishman (1971) found that bilinguals used Spanish when inter- acting in informal situations with family and friends (e.g. when at home or the beach), and English in formal conversations (e.g. with priests, teachers or employers, in situations of church, school, or work). Fertig and Fishman (1969) report that a significant overall differ- ence in frequency of use between English and Spanish as well as a significant language by domain interaction between home domain words and school domain words. Fish- man (1969b) and COOper (1969a) report that the use of a particular language is context specific. English is used in formal situations and Spanish in informal. Finally, 106 COOper and Greenfield (1969) report that young people speaking among themselves use English more Often than Spanish in all domains, including the family, suggesting that the language in the community under study was char- acterized by language shift. That is, acculturation into the American society is taking place, which can be identi- fied by the change in the spoken language. Lieberson (1965) who found no intermixing among linguistic groups in Montreal and that a stable propor- tion of the population has been bilingual since 1920. Rubin (1968), studying Native American-Spanish bilinguals in Paraguay, found that choice of language was deter- mined by age, sex, kinship, relations, occupation, wealth, education, religion, family background, the social setting, the content Of the conversation, the history of the social interaction and the presence of a third speaker. In order of importance of determining language use the dimensions isolated were: location, formality, intimacy, seriousness of the situation and sex. Lieberson (1970) studied the social class differences between mono- linguals and bilinguals in Canada. He found that bi- linguals tended to have higher incomes than francophones and that one Of the forces which induced French mono- linguals to learn English was employment pressure. A greater variety of occupations with higher salaries were available for the bilingual. While the sociolinguists have included many 107 variables present in a bilingual's environment to account for their use of two different languages, one factor, the mass media, has been almost ignored. Mackey (1968: 562)writes: Radio, television, the cinema, recordings, newspapers, books, and magazines are power- ful media in the maintenance of bilingual- ism. Access to these media may be the main factor in maintaining one of the languages of a bilingual, especially if his other language is spoken in the area. Regular attendance at foreign film programmes and the daily reading of foreign books and magazines may be the only factor in maintaining a person's comprehension of a foreign language which he once knew. Reading is often the contact that a person may have with the second language. In addition, the mass media may be an important mechanism for the learning of a second language. This is particularly true where the media is in a different language than an individual speaks. Yet despite the recognition of the importance of the channels of mass communication, there has been little research in this area. The exception is Woelfel, Woelfel, Gillham and McPhail (197”) who have found differential patterns Of media use and interpersonal interaction between monolinguals and. bilinguals in Montreal. This resulted in differential attitudes toward separation of Wuebec from the Confed- eration of Canada. While the Woelfel gt a1. piece is a beginning at describing the relationship between mass media exposure and psychological processes, it failed to explicitly relate mass media usage to the manner in 108 which bilinguals process information. Theoretical Hypotheses.--In the review of the psycholinguistic literature, it was concluded that bilin- gual subjects organize their semantic structure according to the semantic similarity of the symbols used. Two of the mechanisms (attributes) which the psychologists suggest are semantic content and language. Based on the paradigm for the measurement of meaning, the following assumption can be made. Assumption: Symbols will cluster in a semantic structure according to the semantic content for any one given language. Based on the psychological literature and the above assump- tion the following theoretical hypothesis can be justi- fied: 1: In a mixed semantic structure (symbols from two different languages are pre- sent), the symbols will cluster according to the semantic content, but with the introduction of an additional language dimension, sep- arating the symbols of one language from the other. On the basis of his investigation of Native Amer- ican languages (Eskimo, Aztec, Navaho, and HOpi) Whorf (1956) concluded that speakers of these languages organize their experiences differently than speakers of English. This has become known as the notion of linguistic rela- tivity. It has been discussed by Chapman and Kowieski (1975). Because of these cultural differences a word in one language may have a different set of associations 109 than its translated equivalent. Thus, a lexical item in one language may cover a different domain than its trans- lated counterpart and therefore both symbols will have a different set of relationships with the other words in both languages. The semantic structures will be different. Rosenzweig (1957) compared the associational re- sponses of groups of American and French students. He noted that the French group gave more diversified re- sponses than the American group and that the two groups had equivalent responses in only half of the cases. In a later study, Rosenzweig (1959) used free association tests to compare word associations between English and French speaking populations. He found that the degree of overlap (the relative frequency of associative equiv- alence of French and English words with a given stimulus term) to be high. He concluded ". . . that associative habits tend to be held in common among different lan- guage communities." (1959:3”7) These results were sub- stantiated in a later study (Rosenzweig, 1961) with French, Italian, German and American subjects. Similar conclusions are reported by Osgood (197”) based on data from 27 countries with the semantic differential. He reports that three dimensions, evalua- tive, activity and potency can be found in varying degrees in the affective semantic spaces of all languages. The loadings on these dimensions were in the .8 to .9 range 110 on the evaluative dimension, .” to .7 for the potency fac- tor and .3 to .7 for the activity dimensiOn. From these results, Osgood concluded, "This is rather convincing evidence for the universality of the affective meaning system (197”: 33-3”)." Thus, while the individual responses may vary by degree between different cultures, the process by which language is organized is identical despite cultural differences. This would be as expected because we are all people with the same physiological makeup. As a result, the process of communication is not likely to work in a totally discrepant manner despite cultural variation. Additionally, in this discussion the medium of communication has been limited to language which re- quires the same sensory-motor mechanisms to Operate in order for the exchange of ideas to be completed. What may be considered unique to each language is the cultural context in which associations between the label and re- ferent are formed, although similarity may be taken to be the primary principle of organization of semantic information in all societies. Thus, while the linguistic process by which information is communicated and meaning attributed to Objects and relations can be considered identical across language boundaries, culture produces individual variation in the semantic structure of a language to render it significantly different from any other. 111 This discussion suggests the following theore— tical hypotheses. H2: The semantic structure generated by symbols from one language will be significantly different from the structure generated from its trans— lated equivalents, for monolingual subjects. H3: The semantic structure generated by symbols from one language will be described by the same number of underlying dimensions as the semantic structure generated by its trans- lated equivalents. There is some support for these two hypotheses. Barnett and Wigand (1975) used M.D.S. and found little difference between two English language societies (United States and South Africa) and greater discrepancy between the English and Spanish (Mexico) speaking cultures. How- ever, in all cases, four dimensions provided the best description of the data. In the case Of the United States samples, this explained 77.” percent of the "real" variance; for Mexico 88.8 percent; and for South Africa 76.8 percent. These results were determined by the use of a scree test (Tatsuoka, 1971: l”7).l The bilingual individual has the ability to re- ceive information from two different language groups. This suggests that he/she forms associations in a manner that takes into account both languages, resulting in a semantic structure significantly different from both but not nearly as different as the two are from each other. 112 Lambert and Moore (1966), using free association, found that bilinguals living in the same environment have inter- mediate degrees of response similarity between mono- linguals in French and English. In communication, the two monolingual groups could easily miss the full significance of one another's messages because such associational discordance color the meaning and shunt the line of associations Off on quite different routes. In this example, the bilinguals would likely transmit the discrepancy with fidelity from one monolingual group to another, switching from one associational network to another as they change languages (Lambert and Moore, 1966:319). The notion that the bilinguals' semantic struc- ture will be significantly different from both the groups from which they obtain their language is perhaps a lin- guistic manifestation of Durkheim's (1951) notion of anomie. Bilinguals may in fact be incapable of internal- izing the semantic rules of either language due to their use of both. McLuhan (1962) has shown how the phenomenon Durkheim Observed could have resulted from the change in communication technology in the later portion of the 19th century. Given that anomie may result from a change in the overall communication media, it seems reasonable to suggest that semantic normlessness may result from the change in language or the simultaneous use of two significantly different semantic systems. It should be pointed out that bilinguals can communicate successfully in either of their two languages. 113 This suggests that they have the ability to switch codes or, at the least, use that language's semantic rules when speaking in that language. Thus, the bilingual's seman- tic structure while using one language would be more similar to that language's monolinguals than to people who use the other language. Based upon the above discussion, the following hypotheses seem justified. H”: The semantic structure generated by symbols from one of a bilingual's languages will be significantly different from the semantic structure generated by symbols from the other language. H5: The degree of discrepancy between the semantic structure produced by a group of bilinguals (in either of their two languages) and a group of monolinguals in one of the bilingual's languages will be less than the dis- crepancy between the two monolingual groups. . H6: The degree Of discrepancy between semantic structures of equivalent terms will be ordered in the following manner: monolingual language A, bi- lingual in A, bilingual in B, and monolingual in B. Research on bilingualism has ignored how mass media exposure has influenced the way in which these peo- ple process information. McLuhan (196”) has written that the form Of the media affects ". . . the way in which individuals organize experience and fix perceptions" (Carey, 1967: 17). There has been little research dealing with how the form of the medium upon which a message is 11” transferred affects the receiver of that message. Wilkus, Woelfel, Barnett and Fontes (1973) failed to find any effect on pattern recognition ability that might be attributed to exposure to a variety of different media within a given medium, Watt and Krull (197”) found that two dimensions Of the structural or form characteristics of television programs, dynamics and unfamiliarity, could be used to predict viewing patterns as well or better than measures of content. One aspect of form (as opposed to content) is the language in which the message is presented. Might language not also be a predictor of media usage patterns? Woelfel 33 21; (197”) found differential patterns of media use and interpersonal interaction between monolinguals and bilinguals in Montreal. Thus, given the potential for unequal distribution of media content in each of a bilingual's language the following hypothesis seems justified. H7: The patterns of media usage will be Significantly different for mono- linguals in language A, monolinguals in B and bilinguals living in the same geographical location. Given that the above hypothesis concerning the mass media exposure can be confirmed, the following ques- tion can be raised. How does exposure to the mass media and interpersonal relations in each Of a bilingual's language affect the way in which he/She processes informa- tion? More Specifically, how iS an individual's semantic 115 structure affected by the variable exposure to both lan- guages? It was pointed out in chapter two that the seman- tic structure Of an individual was contingent upon his/her past experiences. Clearly, when dealing with linguistic information, the past experiences with each of the two languages should affect the structuring of the semantic components of the bilingual's language. H8: The greater the proportion of mass media and interpersonal exposure in language A (the less the proportion of exposure in language B), the greater the discre- pancy of the semantic structure of the individual from the "ideal type" of monolingual in language B (average structure of language B) and the less the discrepancy from the average structure produced from the mean space of language A. This is for a semantic structure produced by translated equivalents. Summary In summary, this chapter has described a single substantive problem, the organization Of the semantic structure of bilingual individuals. After a description of the extent of the problem and the social significance of the bilingual, the chapter reviewed the literature in two separate but complementary disciplines, psycholin- guistics and sociolinguistics. It concluded that both lines of research are necessary in order to describe how bilinguals process semantic information. The chapter then concluded with eight theoretical hypotheses which derive from past research with bilinguals, the paradigm 116 for the measurement of meaning presented in chapter two and the discussions relevant to each specific hypothesis. 117 FOOTNOTES FOR CHAPTER THREE lThe screei test was performed on the positive roots only. Because metric scaling results in a non- positive semi-definite matrix (it has negative eigen- roots), this test may lead to erroneous findings. One possible improvement might be to perform the scree test on the absolute values of all the roots. This may ex- plain more reliable variance and might not distort the findings. CHAPTER FOUR METHODS Overview This chapter will propose a study to test the theoretical hypotheses presented in the preceding chapter, through the use of the methods outlined in chapter 2, a paradigm for the measurement of meaning. It will restate the hypotheses in operational form, describe the setting for the research, the sample, the instrumentation, the problems of instrument translation and the design to test the hypotheses. Qperationalization The cognitive-semantic structure (meaning system) of an individual may be measured as a fundamental theore- tical variable through the use of metric multidimensional scaling (M.D.S.). An in-depth discussion of the theore- tical significance and the advantages of this method were presented in chapter 2. Barnett (1972) reported relia- bility coefficients for the method of .90 with groups of 75 cases and discussed ways of increasing the over-time reliability by the number and selection of concepts. The predictive validity of time series M.D.S. has been demon- strated by Marlier (197a) in a test of social judgment theory, and Taylor, Barnett and Serota (1975) who demon- strated that the outcome of a political campaign could 118 119 be accurately predicted with the method. Multidimensional scaling has been used success- fully with subjects from non-English speaking cultures. The languages in which the research has been conducted were: Japanese (Kuno and Suga, 1966), Dutch (Van Der Kamp and Pols, 1971), Swedish (Ekman, 1955; Hanson, 1963), Finnish (Nordenstreng, 1968) and Spanish (D'Andrade, 23.21;, 1972; Barnett and Wigand, 1975). Heider and Olivier (1972) used M.D.S. for cross—cultural comparisons to test the Whorfian hypothesis concerning the relation between cognitive and linguistic structure. Subjects from the United States and the Dani culture of New Guinea were asked to perform two tasks. One involved scaling color names and the other Munsell color chips. M.D.S. on the four data sets yielded structures that were more similar under the cognitive conditions than the naming condition. In neither culture were distinct colors confused in mem- ory more than across name boundaries. Thus, retention of color images appears to be unaffected by cultural differences in the semantic reference of color words. Restating the first theoretical hypotheses into operational terms would look like this: H1: In a mixed language space (symbols from two different languages are present) the symbols will cluster according to the semantic content. Additionally, there will be the introduction of an additional lan- guage dimension separating the sym- bols of one language from another. 120 In order to test this hypothesis, one must first generate a multidimensional space from a group of symbols from both of the bilingual's languages. The specific language of each concept would be randomly determined from the pool of lexical items which will be used to test the rest of the hypotheses. From this space it would be determined if on one or more dimensions the concepts of one language were polar opposites from the concepts in the other lan- guage. Since it is not likely that the language attri- bute would be orthogonal from the others used to differ- entiate the concepts, one would next attempt to regress a language dimension through the Space. This is done by treating the dimensions of the space as a series of inde- pendent variables with the loadings of the concepts on each vector as the scores of the case on that variable. A vector of zeros and ones would comprise the dependent variable. Zero would signify that the term was in one language and one, the other. Concepts, not subjects, would be the unit of analysis. Thus, an attempt to predict the loadings on the language vector from the dimensions of the space is occurring. If there is a language attri- bute present and it can be used to differentiate the concepts in the space then it will be reflected in the multiple correlation. Theoretical hypothesis two becomes: H2: A space generated by symbols from one language will be significantly differ- ent from a space generated from its 121 translated equivalents, for monolingual subjects. It can be tested in the following manner. Generate spaces in two different languages from a series of translated equivalents. Next, through a series of translations and rotations place the two spaces to a least-square best fit.1 Then, through the use of t-tests using concepts as the unit of analysis, see if the differences between the spaces differ significantly from zero. The null hypo- thesis would be that the two spaces are not significantly different.2 H3: The space generated by symbols from one language Will be described by the same number of underlying dimensions as a space generated by its trans- lated equivalents. Hypothesis three can be tested through the use of the scree test (Tatsuoka, 1971). It operates as follows. Plot the absolute values of the eigenroots for each di- 3 Then connect these values. The mension of each space. number of underlying dimensions is determined where there is a drastic change in the slope of the line, an "elbow" in the graph. This quantity is the number of dimensions which lie off the line connecting the smallest root to this point (Tatsuoka, 1971). Theoretical hypothesis four then becomes: H”: The space generated by symbols from one of a bilingual's languages will be significantly different from the space generated by symbols from the other language. 122 Hypothesis four can be tested in the same manner as hypo- thesis two. Theoretical hypothesis five becomes: H5: The degree of discrepancy between a space produced by a group of bi- linguals (in either of their two languages) and a group of monolin- guals in one of the bilingual's languages will be less than the dis- crepancy between the two monolingual groups. H5 can be tested as follows. Rotate all three spaces to a least-squared best fit, with the bilingual space placed between the monolingual spaces. Then perform a t-test to see if the degree of discrepancy between the bilingual space and one of the monolingual spaces is significantly less than the discrepancy between the two monolingual spaces. Again the unit of analysis is concepts. H5: The order of discrepancy between spaces of equivalent terms will in- crease in the following manner: monolingual language A, bilingual in language A, bilingual in B, and monolingual in language B. H6 can be tested through the use of trend analysis. The hypothesis as stated only indicates a monotonic relation- ship between the discrepancy and their rank orders. As such, a linear function may provide the best estimate of the relation. Thus, linear trend analysis will be the method of choice. It is discussed in depth by Hays (1973: 691-69”). Linear trend analysis operates in the same manner as any comparison among means. However, since analysis of variance in the linear case agrees exactly 123 with linear regression, the later method may be used. The significance test will be performed on the correlation between the predicted rank order and the discrepancy scores among the spaces. H7: The pattern of media usage will be significantly different for mono- linguals in language A, monolin- guals in B and bilinguals living in the same geographic location. H7 can be tested by gathering media exposure data through a series of items on a questionnaire and then comparing the groups on these items. The items may be worded as follows: On the average, how many hours per week total do you spend with newspapers? Of these hours, how many are in language A? Of those hours, how many are in language B? A question of this general form may be asked for seven different media (newspapers, television, film, radio, books, magazines and records or tapes). This question's form may also be used to gather data on the frequency of inter- personal communication in each language. Theoretical hypothesis eight becomes: H8: The greater the proportion of mass media and interpersonal exposure in language A (the less the proportion of exposure in language B), the great- er the discrepancy of the space of the individual bilingual from the mean space of language B monolinguals and the less the discrepancy from the average space of language A monolin- guals. This is for a space produced by translated equivalents. This hypothesis may be tested in the following 12H manner. First determine the mean space for the two mono- lingual groups. Then compare each individual bilingual space to the monolingual spaces. This is performed by a series of rotations and translations on the underlying coordinates.l+ The difference or separation between the space of the individual bilingual and the average mono- lingual space is then determined by simple subtraction of the coordinate values. This value is then entered in a multiple regression as the dependent variable with the hours of media exposure as the independent variables. The correlation between the variables will be the statistic used to infer the test of the hypothesis. The setting.--The setting for the test of the above hypotheses will be Canada's capital, Ottawa, and its sister city, Hull. This metropolis rests on the border between Quebec, and its francophone culture and the English language culture of Ontario. Here both cultures meet and their languages mix. According to the Royal Commission on Bilingualism and Biculturalism (1969: 35), 30.8 percent consider English to be their mother tongue and 37.7 percent, French. 55.8 percent are anglophones and 13.2 percent are francophones, while the remaining 30.8 percent are bilingual (RCBB, 1969: 35). Of the bilinguals, less than 15 percent of the English ethnics speak both languages, while almost 85 percent of the French are bilingual. Thus, French and English will be the languages used to test the hypotheses. 125 (Mass Media.--The mass media also reflects this unequal distribution. There is more English language con- tent available than French. Certain content may only be available in English. This is probably the case with television. Programs from the United States are only available in English. The same is true of movies, be- cause of the importance of Hollywood productions. At the time of the study, there were 1n different television stations broadcasting on cable in Ottawa. Of this total, three were in French, and 11 were in English. Of these 11, four were emnating from the United States, and seven from Canada. There are a total of 21 movie theatres in Ottawa-Hull, 18 featuring English language pictures and three French language. Radio is balanced. There are a total of 12 stations which can be received, 6 AM and 6 PM. On the AM band, there are two French and four English stations and on the FM frequencies, four French and two English. Three daily newspapers are published in Ottawa. The English are the Ottawa Journal and The Citizen. £2.22213 is published in French. In addition,‘ newspapers from both Toronto and Montreal are available.5 Thus, although there is an unequal distribution of the media in each language, both French and English material are readily available. The media available in each language is summarized in Table One. 126 Table l. The Available Mass Media in Ottawa-Hull, February, 1975. Total English French Television 1a 11 (u US) 3 Theatres (film) 21 18 3 Radio 12 6 6 u AM 2 2 FM H Newspapers 3 2 l 127 Instrumentation The English language version of the instrument for? the research is located in Appendix A. It is composed of two sections. The first contains a series of unidimensional items including a series of demographic questions, and language use questions. Also included are preference of the media and a series of media usage questions. These attempt to describe the individual's pattern of media exposure in both French and English. Additionally, there are questions concerning the language of interaction with friends and family. Bilinguals are asked when and where the second language was learned. Finally, this section is concluded with two questions attempting to get at the variable enjoyment and effort for each of the bilingual's languages. These questions can all be justified from the socio- linguistic literature with the exception of the two questions dealing with the time and place of the learning of the second language. Those can be justified from the research on the compound-coordinate distinction. The second section of the instrument was com- posed of #5 direct pair comparisons based on ten differ- ent concepts, using the criterion standard of red and white as 100 "galileos" apart. These will be used to generate a multidimensional space to examine the per- ception of the mass media. The concepts were: 1. Books 6. Television 2. Magazines 7. Sports 3. Newspapers 8. Movies u. Music 9. Information . 5. Radio 10. Entertainment 128 These concepts were chosen for a number of rea- sons. First, since the eighth hypothesis attempted to ‘ relate media exposure to semantic structure, among the logical domain of concepts which would be affected by the variable use of the mass media would be the percep- tion of that institution itself. Thus, six different media were chosen, two functions of the media, entertain- ment and information, and two terms which dealt with the content of the media, sports and music. Terms which dealt with the perception of the media were used by Barnett and Wigand (1975), with success in three different countries. Additional data concerned with the perception of the media are being gathered in five additional countries, including Micronesia, Israel, Great Britain, Australia and Nigeria. This has been planned so that the Canadian data will be comparable to this multinational study. The second reason for the choice of these terms dealt with the gender of the concepts in French and their frequency of use in English. The need to control the gender of the terms in French has already been discussed. Additionally, it was felt that frequency of use may affect a symbol's degree of association with related terms. The Specific concepts were generated in the following manner. Using Thorndike and Lorge (19%”) and Carroll, Davies and Richman (1971), a pool of potential terms were divided into high frequency of use (more than 100 times per million), moderate use (50-100 times per 129 million) and low frequency of use (less than 50 times per million). Next, the gender of each word in French was determined in order to control for the potential acoustic relatedness of the items. Finally, the cells of the matrix presented below were filled such that concepts from both domains (media names and media functions or content) were present, and at least onemale or female term at each level of use. Finally, the concepts were placed in random order in the form presented on the questionnaire so as to minimize the effects of order on the scale values. Domain 2 Media Functions 8 Content Domain 1 Media Names Male Female Male Female‘ HiilOO/lo6 Newspapers Music Books Med Movies - . - 50'100/1011Ma azines Telev151on Information Low _<50 / 10 5 Radio Sports Entertainment Translation The questionnaire was translated into French following the back-translation procedure. follows: This is done as More than one bilingual individual translates the instrument from English into the language into which the research will be performed, in this case French. The reason for the use of multiple translations is to provide checks on the reliability and validity of the translation 130 (Ervin and Bower, 1952). Then, a second set of bilinguals translate the tentative instrument back into English. Next, the results are compared. Did the back-transla- tions produce the original terms exactly in all cases? If the answer is yes, the terms are retained. Where there are discrepancies, both sets of translators should meet to discuss the differences to see if a consensus can be reached. If consensus is reached, then that item should be translated back into English by an independent trans- lator to insure validity. If it becomes impossible to agree upon one translation, the item should be deleted. Variations of this procedure have been used by Kumata and Schramm (1956), Trandis and Osgood (1958), Tanaka, Oyama and Osgood (1963), and Sarbaugh (1967). For this study, an undergraduate French major and a Ph.D. candidate in communication familiar in survey research translated the questionnaire from English into French and then two undergraduates translated it back into English.6 Only slight disagreements were present. They were resolved upon discussion. This high degree of agreement should be exPected with French and English, owing to their similarity and popularity among students. The instruments were then sent to Canada to allow the opportunity for modifications in the French questionnaire to conform to the French-Canadian vernacular. No changes were made. The French language version of the instrument is also located in Appendix A. 131 Design.-—The design for the study outlined above would include five groups. 1. English monolinguals 2. French monolinguals 3. Bilinguals with an English instrument u. Bilinguals with a French instrument 5. Bilinguals with a mixed-language instrument The bilinguals (groups three and four) will emerge out of the presentation of the questionnaire in either language to large groups of subjects. This is made possible because of the large numbers of bilinguals and their high degree of integration into Ottawa area. The English language instrument was administered to a larger group of English speaking subjects, some of whom are monolingual and others bilingual in French. Likewise, the French language instrument was given to a sample of French speakers, some of whom are franc0phones and others bilingual in English. On the top of page u of the ques- tionnaire is a set of instructions which reads, "If you speak both French and English, please complete this page. If not, go on to the next page. Thank you." In this way bilinguals can be sorted out from the anglophones and franc0phones. While this may seem a poor method to separate bilinguals from monolinguals because it does not take into account their competence in the two lan- guages, it should be noted that in Canada, generally and specifically in Ottawa, there are strong social norms connected with bilingualism. Thus, it is not likely that subjects would take lightly a series of questions 132 concerning their use of a second language. It should be noted that the Canadian census only asks three questions relating to language. They concern the respondents ethnicity, his/her parents language, and his/her knowledge of the two official languages, French and English. The last question is the only direct measure of bilingualism in the census (Macrae, 1969). These were questions 15 and 16 and the sorting procedure on page u of the questionnaire. Additionally, the questions on effort and enjoyment could serve as a validity check on the sorting process. An additional design manipulation check was per- formed. The author examined all the bilinguals' completed questionnaires and then noted if the respondent made any use of the second language. If the language of mass media usage and interpersonal communication were all in a single language, then that potential bilingual was re- classified as a monolingual. This was a very minimal criterion for a bilingual because as will be explained in chapter 5 most monolinguals make at least some use of the second language. In this study, bilingualism has been treated as a dicothomous variable. Subjects were labelled as either bilingual or monolingual. Perhaps, it would have been more accurate to conceptualize the variable contin- uously, as the degree of bilingualism. However, to gather an adequate measure would have required a great deal of time and effort and have been of questionable‘ 133 utility concerning the hypotheses in question. These have been a number of measures of bilin- gualism which taken into account an individual's variable ability to communicate in each of his/her languages. They have been reviewed by Macnamara (1969) and Cooper (1969b). Because of the complexity of linguistic ability there are no direct measures of bilingual competence. Rather, there are a number of tests which indirectly tap some component of an individual's communication skills in the languages. Among them are self-rating scales, tests of verbal fluency (Speed of responding to verbal stimuli, speed of verbal production in two languages, word completion tests, Speed of reading and word naming), measures of flexibili- ty in each language or dominance of one language over another. Theoretically, these have been designed to take into account one or more aspect of encoding (Speak- ing or writing) and/or decoding (listing or reading) Skills in both languages. Because these measures may tap a number of di- mensions other than the individual's ability to commun- icate with both languages, one would next have to con- struct some sort of factor weighted index as the measure of the degree of bilingualism. Mackey (1969) questions whether these measures accurately reflect the bilingual's ability in each language. Cooper (1969b: 194) notes, "There have been few attempts to relate the verbal traits which have been studied to social correlates." They are 13” therefore of questionable validity. Lieberson (1969) questions whether the benefits gained from this sort of data is worth the potential hazards in respondent fatigue in a survey situation. Perhaps in an experimental setting these measures would have more utility. But due to the time constraints and the limits of the number of ques- tions that could be asked in a questionnaire, the benefit that might have been gained by this conceptualization is limited. One of the problems in evaluating the past re- search was that there was no attempt to control for the differential ability in both of the bilingual's languages. No standard operationalization of bilingualism was applied. This same criticism may also be levelled at this study. Also, no attempt was made to control for the compound-coordinate distinction. However, all the bilingual subjects were coordinates. The subjects for group five, bilingual subjects with a mixed-language instrument, would have to be gathered separately. Before receiving the questionnaire, subjects were asked if they were bilingual. A positive response was necessary to obtain this version of the questionnaire. For this group, the concepts in the direct-pair comparison section were in the format out- lined in the methods chapter, page 119. This portion of the questionnaire is also located in Appendix A. 135 On February 10, 11 and 12, 1975, questionnaires were administered to a series of large university classes in Ottawa. The administration took 30 minutes. The institutions were Carleton University, where the language of instruction is English, the University of Ottawa and St. Paul's University, where French is used in the class- room.7 The present author administered the question- naires at Carleton and a bilingual professor performed this function at Ottawa and St. Paul's after being briefed at great length by this author.8 . Subjects.--The subjects for this study were all students at the above mentioned Canadian universities in Ottawa. This, of course, has its obvious limits to generalizability, but has been the method of choice due to the subjects' availability and restrictions due to cost. The total sample consisted of 32H subjects, 232 students at Carleton and 92 at Ottawa and St. Paul's. Of the 232 students at Carleton who participated in the study 150 (68 percent) were monolingual in English and 82 (36 percent) were bilingual. Of the 92 students at Ottawa and St. Paul's who served as subjects, 15 (16 percent) were francophones, and 77 (8“ percent) were bilinguals. This is in line with what Lieberson (1970) describes as the language usage patterns for Ottawa. More of the English-background subjects are bilingual than one might expect from the general population. This is probably-due 136 to the fact that the subjects are college students, who typically are required to learn French as a second lan- guage. The overall sample was disproportionately male (57.7 percent male to 02.3 percent female), and almost entirely white, (95.3 percent). The average subject had completed 10.27 years of school. On these variables, the anglophones, franc0phones and bilinguals did not differ significantly. Comparing the three groups, the French language subjects tended to be younger than the other groups. Also, they differed Significantly by religion and socioeconomic status. While the total sample had a mean age of 20.8 years, the English average was 21.6, the bilinguals, 20.2, and the French 19.8 years. This differ- ence is significant at the .05 level. However, it is not clear what effect, if any, this difference Should have on the results of this study. As expected, the francophones were all Catholic, while the bilinguals were 59.2 percent Catholic, 17.2 percent Protestant, 5.1 percent Jewish and 17.2 percent Atheists or other. The ang10phones were 25.3 percent Catholic, u7.3 percent Protestant, 2.0 per- cent Jewish and 25.3 percent Atheists or other. The fathers of the English language sample had significantly more years of education (p5,05) than the bilinguals and the francophones. The anglophones' fathers had signifi- cantly higher occupational status (p:.05) and income p:.05) than the French group but did not differ from the 137 bilinguals on these variables. However, the bilinguals had significantly higher social status than the francophones on these two variables (p§.05 on both). Following Lieberson (1970), these relationships were expected. Summary To summarize, this chapter has proposed a study to test the theoretical hypotheses presented in chapter 3, through the use of the methods outlined in chapter 2. It restated the hypotheses in operational form, described the setting where the hypotheses will be tested, the construc- tion of the instrumentation for the test, the procedures for the translation of the questionnaire into French, the design to gain the necessary data, and finally, it de- scribed the sample of university students which were used as the subjects for this study. 138 FOOTNOTES FOR CHAPTER FOUR 1A computer program (Galileo, version 3) which accomplishes the calculations necessary to rotate two or more spaces to a least-square congruence is available at Michigan State University, Department of Communication. For an in-depth discussion of the procedures see Serota (197”). A number of rotational algorithms exist which provide variable quality of solution. At present, the least-squares best-fit seems to provide an optimum result compared to other rotations. Inherent in this procedure is the problem of overestimating some changes while under- estimating others. The author is currently involved in the testing of a new procedure in which a theoretically defined set of concepts is held constant (this subset is rotated to a least-squares best-fit) and the remaining concepts are positioned accordingly. It is identical to the heliocentric notion for calculating the motions of celestial bodies. For an indepth discussion of this t0pic see Woelfel, et al. (1975). While the least—squares rotation is appropriate—for the comparisons of static samples like these, the overtime analysis of changes re- quires the alternative procedures suggested by Woelfel ' 33 a1; (1975). h 2The use of significance tests and inferential statistics of any sort runs against the spirit of the method. Chapter 2 pointed out that metric M.D.S. is a series of continuous ratio scaled distance estimates. It assumes the height of absurdity to reduce these estimates to a dichotomous decision of an acceptance or rejection of the null hypotheses. These data can and should be used as a description of the structure of the semantic space of individuals or groups. Thus, one could say they de- scribe a certain relationship without attempting to infer beyond the sample of subjects or concepts. Additionally, these data are based on a large number of independent observations of the relationship between a particular pair of concepts. This notion is not taken into account by this significance test, where the unit of analysis is the number of concepts or spaces and thus the degrees of freedom are some small numbers rather than the number of independent observations. At the present time there does not exist an adequate significance test for the differences between multidimensional Spaces or for that matter factor struc- tures. Although there have been a number of notable attempts (among them Lawley and Maxwell, 1963), these have been inadequate. The Lawley and Maxwell solution takes the natural log of the determinant of one matrix (factor structure) and subtracts it from the log of the other. The resultant value may be then tested for significance by 139 the use of the chi square distribution. This solution is inappropriate because the structure of the M.D.S. produces a singular matrix. It is of rank N-l. It has one column of zeros making the determinant zero by definition and the test inapplicable. The test presented in the text will serve in this dissertation but the author's reserva- tions concerning the use of any test of statistical inference should be noted. 3The reason why the absolute values of the eigen- roots are used is because metric scaling inherently re- sults in a non-positive semi-definite matrix. The multidimensional Space is non—Euclidean. When this matrix is orthogonally decomposed, negative roots result. These imaginary roots are reliable and meaning can be attributed to them (Danes and Woelfel, 1975). As a result their absolute value rather than Signed-value should be used in the scree test. l“These calculations can also be performed by Galileo, version 3, see footnote 1 above. 5Data on the availability of the media were gathered by the author in a number of ways: Through newspaper advertisements and listings, the listings in the Ottawa-Hull telephone book and direct participation with the media. 6The author would like to thank Timothy Mabee and Elizabeth Ekdahl for their help in the translation of the instrument into French. 7The author would like to thank Dr. Thomas McPhail, School of Journalism, Carleton University, Ottawa, Canada, for making all the necessary arrangements at Carleton, the University of Ottawa and St. Paul's University. Without Dr. McPhail this dissertation would only be a dream. The author also would like to thank Dr. Roger Byrd, and Dr. Peter Johansen, Carleton University, Dr. Ross Hastings, University of Ottawa and Dr. Andrew Ruszkowski, St. Paul's University for the use of their classes. 8Dr. Hastings administered the questionnaire at the University of Ottawa and Dr. Ruszkowski at St. Paul's University. CHAPTER FIVE RESULTS Overview The objective of this chapter will be to report the results of the February, 1975, data collection. It was designed to test the hypotheses presented in the previous chapter. This chapter will proceed hypothesis by hypothesis. From these results inferences can be drawn which may help demonstrate the usefulness of the paradigm presented in chapter two and its applicability to the substantive problem of how bilinguals organize their semantic system. Hypothesis One H1: In a mixed space (symbols from English and French are present) the symbols will be organized according to the semantic content. Additionally, there will be the introduction of an additional language dimensions, separating the symbols of one language from the other. Fifteen students at Carleton University filled out the mixed-language questionnaire during the February data collection. The results of this set of data pro- duced the mean distance matrix presented in table two, and the spatial coordinate matrix presented in table three. Through the use of a scree test, it was determined that only the two largest positive dimensions should be re- tained in the multidimensional space. They accounted 190 191 for 75.1 percent of the variance in this space. The graphic representation of this spatial manifold is presented in figure one. These two dimensions may be labelled as evaluative (books to sports) and entertainment-information.l Further, the first dimension separated the electronic media from printed media. ----------------------------------- .1------------ Tables Two and Three and Figure Three About Here Neither of these two dimensions nor any of the remaining seven can be easily labelled a language dimension. The reason is that on none of the factors are the English concepts at one end and the French, at the other. This can readily be verified by comparing the signs of the factor loadings. On no dimension do the English concepts (books, magazines, radio, sports and entertainment) share the same Sign, while the French terms (des journaux, la musique, la television, cinema and l'information) have the opposite. The results of the attempt to predict the language vector from the spatial manifold also met with failure. First, each of the nine vectors were correlated with the language vector, producing the coefficients in table four.2 Because the multiple correlation of all nine factors with the language vector with 1.0, only the 2 of .15 or better were entered into a factors with an r multiple regression.3 These were dimensions four, five and six. It should be noted that these factors would not have been retained after the scree test, which would make 102 mH n z 0.0 00.00 00.00 00.00 00.00 00.00 00.00 00.00 00.0: 00.00 unmsc0munmpam .00 0.0 00.00 00.00 00.00 00.00 00.00 00.00 00.00 00.00 :o0pmap0000.0 .0 0.0 00.00 00.00 00.00 00.00 00.00 00.000 00.00 mswa0o ma .0 0.0 00.00 00.00 00.000 00.00 00.000 00.000 000000 .0 0.0 00.00 00.00 00.00 00.00 00.000 co000>W0w0 ma .0 0.0 00.00 00.00 00.00 00.000 o0cmm .0 0.0 00.00 00.00 00.00 00000:: mg .0 0.0 00.00 00.00 xamcpao0 man .0 0.0 00.00 00:000002 .0. 0.0 mxoom .0 00 0 0 0 0 0 0 0 0 0 cowvwocoo ownswcmq omxflz m :0 mamamcflawm mom xwhpmz monopmwn :moz .m manna 1H3 00.00300 mocha oo.ooH 00.000 00.000 00.300 00.300 00.00H 00.000 00.0HH 00.ooH 00.00 ::900 omucsooom mocmpmwv nzpmcfimmew 0cm 00090 Hmpou mo mmmmucwouoa m>0000=sdo 00.30 00.00 00.00 00.000 00.000 03.00 03.00 00.00 00.00 00.03 ::nou ompcsooom wocmpmwo 0009 no mommucmopma m>wumasfizu 00.0H: 00.0: 03.0: 00. 00. 00.0 00.0 0H.NH 00.00 00.03 ::aouom> Hmsvw>va0 0n pom oovcaooom mocmpwwo Mo ammuCoopmm 00.0330: 00.0000: 33.000: Ho. 30.00H 03.0000 03.0000 00.0003 00.0000 30.03000 ::xfipume nepom>cowmo no 0090000 uwsam>cowwm o3.03 00.0H: Ho.3 00. 00.0: 00.0 00.00 00.0: 00.0H HH.0 ucoscwMHnopcm .00 00.00 03.00 00.3 00.: 00.0: 03.0 00.00: 00.30 30.00: 00.0: cowamfipomca.q .0 00.0H: 00.0: 00.3H No. 0H.3 00.0: 00.3: 00.00 00.03 03.00 mecwo 0g .0 00.00: 00.00 00. 00. 03.0: 00.0 00.00 00.: 00.00: 00.00 muuonm .0 00.3 00.00: 00.30: 00. 00.3: 00.00: 00.00: 00.0 00. 03.00 coamw>w0wa mg .0 03.3: 30.00: 03. 0o. 30.: 00.00 00.00: 00.00: 03.0 00.00 owomm .0 00.0 00.00 00.3: Ho. 03.0 00.3: H0.0: 00.00: 03.00 00.0 osvwmzz mg .3 00.0 00.0: 00.0: 00.: 03.00 30.0: 00.0 00.0: 00.03: 00.00: xnmcpaou 0mm .0 00.00: 00.0: 03.30 00.: 00.0: 30.00: 00.0: 00.00: 00.30: 00.00: mafinmmm: .0 00.00: 00.0 00.00: 00.: 00.0: 00.0 00.00 03.00 00.00 00.00: mxoom .H 00 0 0 0 0 0 3 0 0 H .cowuwccoo mmmsmcmq vmxwz 0 c0 mamsmc«dam pom mopmcwvnooo 0000000 .0 manna III. IliIIIIIII 1w e BOOKS OLA MUSIQUE 0 LE CINEMA O ENTERTAINMENT oRADlO G OMAGAZINES L'INFORMATIONO DES JOURNAUX O LA :Eu’vmow SPORTS (D Figure 3. Bilingual Mixed-Language Space. N=15. 1H5 their contribution questionable from the start.u Together, they produced a multiple correlation of .83, which when squared created an R2 equal to .69. It is significant at the .056 level. This probability is not low enough by traditional standards to reject the null hypothesis. However, the multiple correlation squared corrected for shrinkage was .HH7.5 Thus, in the population of potential concepts, one would estimate that nu.7 percent of the variance on these three dimensions would be explained by language. It should be noted that the inclusion of the language variable did not alter the dimensionality of the space. According to the scree test two underlying dimensions were present. As Will be pointed out later, this result also occurred in the unilingual conditions. Thus, in the mixed-language condition the language attrib- ute did not become manifest as a separate underlying dimension. In summary, hypothesis one cannot be supported. Neither of the two dimensions which would be retained after the scree test could be considered a language dimension. From the regression analysis on all nine dimensions, it was determined that nu.7 percent of the variance of the language dimension is accounted for by these three dimensions. However, this was not significant at the .05 146 Table u. Correlations of Individual Dimensions from Bilinguals Mixed Language Space with Language Vector. Dimension r r2 l .1u7 ‘ .027 2 .1u1 .020 3 .318 .101 u -.535 .287 5 -.uu1 .19” 6 .H60 .212 8 -.155 .02u 9 .255 .065 10 .27u .075 * Dimension 7 was eliminated from this analysis because its variance equalled zero. 147 level. Clearly, the language attribute is present in the space. However, it only accounts for a small part of the total variance of the space. Hypothesis Two H2: A space generated by symbols from one language (English)-W1ll be Significantly different from a space generated from its translated equivalents (in French), for monolingual subjects. During the February data collection 150 anglo- phones at Carleton and 15 francoPhones at Ottawa and St. Paul's completed the questionnaires in their respective languages. The results of these sets of data are presented as the mean distance matrices (tables five and six) and the spatial coordinate matrices (tables seven and eight). A scree test on both sets of data determined that a two- dimensional solution made up of the two largest positive roots was appropriate for the English and French language spaces. In the case of the English space, these two dimensions explained 71.20 percent of the variance. The first two dimensions accounted for 63.69 percent of the variance for the francophones. The graphic representation of the English spatial manifold is presented in figure four and the French space in figure five. As in the case of the mixed-language space, the first dimension may be labelled evaluative and the second, entertainment- information. It also separated print media from electronic media. The French space was then rotated to a least- square congruence upon the English space using all ten factors. The distances between the two samples for each concept are given in table nine. The mean difference was 35.59 units. Under the assumption that if these two spaces were the same (the null hypothesis) the distances between the concepts would be zero, a t-test was performed to determine if these distances differed significantly from zero; t equalled 10.n1. This is significant beyond the .001 level with df equal to 18. Thus, the null hypothesis of no difference between the two spaces can be rejected. Hypothesis Three H3: The space generated by symbols from one language (English) will be described by the same number of underlying dimensions as a space generated by its translated equivalents (French). Based on the scree tests from the English language sample, the French speaking subjects and all sets of bilinguals in either of their two languages or when both languages were used to construct the Spatial manifold, it was determined that there were two underlying dimensions 1&9 omH u z 0.0 ma.ss ms.mfi OH.am mm.mm ~m.sm :m.ma mm.mm mm.m: ca.mm paws:«m»pmycm .oH 0.0 N:.ms am.moa s~.Hm mm.mm mm.HHH Nm.Hm :N.m: mm.am monumegomcH .m o.o :m.m~H ::.mm m:.soa mm.am mm.HmH Hm.:NH o:.ma mmfi>oz .m 0.0 ms.sm mo.oHH m~.:mH mm.sm mm.mmH mm.mma munoam .5 0.9 mm.mm :m.~m ms.mm mH.soH mm.:ma :oflmw>mama .m 0.0 mm.mm mm.mm sm.~oa ~H.-H owvmm .m 0.0 am.HmH Hm.o~fi ~m.m~H ohms: .: o.o sm.mm mo.mm mnoamamsmz .m 0.0 NO.»@ mmszmmmz .N o.o mxoom .H OH m m s m m s m N H .nmwamcm HmsmCflHoaoZ pom xwaumz mocw#mwn ado: .m mHAMB 150 ma u z 0.9 mm.mm sm.mm m:.m~ ms.:m oo.mm m~.m~ :H.mm m~.mm m:.mm unmaommwphm>wo m4 .0H 0.0 am.:m so.Hm :m.mm Hu.~m HN.moH oo.m~ mm.m: ms.mm :oHpmeuomaH.H .m o.o ao.om so.:m sm.mm H~.Nm oo.mm Hm.mm :H.oa mawcHo mH .m 0.0 mm.mm :H.mm so.mm mm.om Hm.mm mm.:HH upomm mg .a 0.0 oo.m: H~.:s am.mm mm.ms om.:m conH>WHwa mg .m 0.0 m~.Hm m~.sm so.ms :H.mm ovam MH .m 0.0 m:.am m~.Ha om.am mavasz ma .: o.o mm.mz no.0m xsmcnsoe moo .m 0.0 mu.m: mm=>mm mmn .N o.o mmp>HH mom .H 0H m m a m m a m N H .Aocohm Hmswaflfiocoz pom anpmz monopmflm cum: .m GHQMB 15]. om.smomm oomph oo.ooa mu.bHH Hu.omd on.:md mw.mma mm.mmH um.mmH ~>.o~H mm.moa oo.om napom omucsooom oocmumwo Axumcwmmefi cam Hmouv HMHOP mo wommpcmopma o>wumHneso ma.mp om.mm mm.mm nm.mm oo.ooH oo.ooH mm.hm :m.mm om.ms mm.:: nunom oopcsooom mocmpmwo Hmmn mo wmmmvcmoumo o>wpmH5550 ~m.~Hn mm.mu ~m.mn m:.Hu oo. ::.~ -.m :m.ma m~.Hm mm.:: nupovom> Hmsow>wocfi an now ompcsooom mocmpwwo no ommucoopmm hm.mmbm| mn.>m~ml mH.mmmHl no.2msl ha. nm.womd o:.H~m: mm.Hmmm mm.ommmH mm.mhumm uuxwuvma povoo>comwo mo AquOpv mmsam>cumwu 5H.mm oo.mH mm.: mm.oHn mo. m:.:a mm.m~u «H.mau «H.2m mm.H~ Homeoprpupcm .oa H:.o mm.H: m~.oH| mo.mu mH.| ma.wu m:.mm Hm.ma Hm.o:u no.~:n covaEpomcH .m :m.:d m:.mal ~5.HHI mm.:d NH. hm.m|. m~.mu No.02: Hm.mm mm.: mow>oz .m Hp.mmu m:.HH mm.hu mm.| mo.u om.ma mm.mH| hm.m| mm.m~u Ho.HHH wupomm .5 m:.Hn :m.m~| ::.md mu.mn co. mm.:fi m:.um om.~ mm. sm.mm cowmw>mHme .m sm.oa oo.mmu so.mn 3m. Ho. mo.mHn mm.ma mm.mm mo.oH H:.~ owomm .m om.mun mm.m~ mm.oH mm.» hm. mm.z mm.mn Hu.m~ no.3» ~m.oal owmsz .: H~.m «3.- ~:.- Hm.mH ~N.- as.~- H:.HH- mm.~ um.mm- mm.mHu mamamamzmz .m s:.H m:.:- hm.mH- aH.H mo.- 5H.m~ =5.m~- mm. :~.m~- mm.~mu mmcHNmmmz .N m:.sma Hm.uau mm.m ~:.mu Ho. sm.oan uo.oan mm.omn mH.m mm.mha mxoom .H oH m a s m m a m u H .smwamcm Hmamcwdocoz now wopmcflonooo Hmapmmm .5 manna 152 «H.2mmmu manna oo.ooH mm.HHH H:.mHH os.-a o>.~NH 1:.oua mu.mad sm.ooa ma.mh mu.H: Inaom oopcsaoom mocmumfio “apocammsw can downy Hmuou mo wmwmucmopom o>wumansdo om.Ho :N.Hm mm.um oo.ooa oo.ooa ma.mm mm.mm om.mm mm.mo ow.mm nunom oovcaooom mo:MVmfio Hoop mo mommVCmonoq o>wumasEgo :s.ms mo.mn mm.~| oo. :m.H o~.: mu.aa Hm.ma mo.om oo.mm InOpom> Hmaow>wocw an pom oopcsooom cacopmflo no mmmvcmopmm mm.~muu| mm.nmnan on.mmnl mo. mm.m~m mH.HomH om.smmm :m.mm~m Hm.mmmm m:.oomm unxwppme napom>cmwwo mo vaOQV mosam>comwm mm.mm hm.b mm.: No. mm.ml mm.oal m~.mu mm.od m:.m| mm.mml Have Immwwvnm>wa «4 .od No.5H H:.mHI wH.oH| Ho.u ca.mu :m.m mu.ma ~:.ma mm.HH: mA.m: cowpusuomcH.A .m w~.mn mH.:H HH.HH| mo. hm.H mm.m mm.:m Hm.:H mc.mH :~.mmn MEwcwo on .m on.san :o.m| 2:.ml 0H. mo.H ms. sm.ch mm.:: om.::| :m.mt upoam ma .5 om.~HI 3:.mu mm.ma Ho.u ~m.H mm.m| um.am Hm.m| mm.hmu mm.H cowww>wawa an .m ~m.mHn ma.- :a.mu mo.| mm.mu mm.mu mm.HHa mu.o:: hn.s~s mn.HH owomx mg .m m~.H :m.ma| mH.HI 50.: H:.u hm.oH wN.NHI mo.:ml mm.H mm.m:l uswwmdz MA .2 ma.ma mm.m mo.m Ho. :o.mH mw.H mm.m| Hm.m Hm.~H mm.m: xsmcnsOb mma .m mm.su oo.w mH.~H No. mm.ma mm.mH mm.uH| mu.m Ho.>~ mm.HH mms>mm «on .w m:.mau m:.m| 55.“: Ho. Hm.aa Hw.mnc nm.s| mm.m mm.mm mo.m mop>fiq moo .H 0H m m h m m 3 m N H .nocmnm Hmsmcaaoco: pom moumcwonooo Hmwpmmm .m magma MUSICO OBOOKS OMOVIES oENTERTAINMENT ORADIO TELEVISION 6 MAGAZINESO INFORMATION 0 NEWSPAPERSO SPORTS o Figure H. Graphic Representation of English Monolingual Space. N=150. 15H DES LIVRES o 0 LA MUSIQ UE LE CINEMAo c LA RADIO 0 LE DIVERTISSEME NT DES REVUES o L'INFORMATION O oss JOURNAUX o' °LA TELEVISION 0 LE SPORT Figure 5. Graphic Representation of French Monolingual Space. N=15. 155 Table 9. Differences Between the French and English Monolinguals' Spaces after Least-Square Rotation on all Ten Dimensions Concept Differences in Spatial Position Books 39.08 Units Magazines ”0.26 Units Newspapers 37.n3 Units Music H1.67 Units Radio 25.n8 Units Television 27.83 Units Sports 59.28 Units Movies 27.13 Units Information 35.59 Units Entertainment 22.12 Units Mean Difference 35.59 Units 156 which were shared by all groups of subjects. In all cases, the dimensions selected were the two largest positive dimensions. They may be labelled evaluative (print media- electronic media) and entertainment-information. For the English monolinguals, these two dimensions accounted for 71.20 percent of the variance in the space. For the French monolinguals, it attributed 63.69 percent, for bilinguals in English, 75.80 percent, for the bilinguals in French, 72.50 percent and for the bilinguals with concepts in both languages, 75.07 percent. Thus, this hypothesis seems to be supported by the data. Although no significance test was applied to these values, confidence may be placed in them due to the consistency of these scores. The range in the proportion of explained variance is only 12.11 percent and the greatest deviation from the mean percentage (71.66) of variance explained by these two factors is only 7.97 percent. Additionally, both dimensions can be easily identified and the same label can be applied equally well to the corresponding dimensions in each space. In summary, hypothesis three can be confirmed with a high degree of confidence. Hypothesis Four H”: The space generated by symbols from one of a bilingual's languages (English) will be significantly different from the space generated by symbols from the other language (French). 157 During the February data collection, 82 bilinguals completed the English language version of the questionnaire at Carleton and 77 bilinguals completed the French version at Ottawa and St. Paul's. The results of these sets of data are presented as the mean distance matrices (tables ten and eleven) and the spatial coordinate matrices (tables twelve and thirteen). A scree test on both sets of data determined that a two dimensional solution made up of the two largest vectors was appropriate for both spaces. In the case of the English bilinguals, these two factors explained 75.80 percent of the variance. The first two dimensions accounted for 72.50 percent of the variance for the French bilinguals. The graphic representations of the English bilingual space is presented in figure six and the French bilingual in figure seven. As in all the other cases, the first dimension may be labelled evaluative (print media—electronic media) and the second, entertainment- information. The French bilingual Space was then rotated to a least-square best-fit upon the English bilingual space. The mean discrepancy between the individual concepts was 21.80 units. The distance between the concepts is given 158 «muz 0.0 :m.am mm.mH s:.m: H~.~m mo.mm 35.:H sm.:m ms.~m mo.mm “coechpamwcm .oH o.o 5H.mm HH.:mH mm.mm mH.mm Hm.omH Hm.mH mH.~m mm.H: :oHumapomaH .m o.o :a.o~H mm.mm Hm.mm Hm.mm om.mHH mm.sm Ha.as moH>oz .m 0.0 :o.ms ms.mHH mo.smH mm.mHH Hm.N:H os.omH museum .5 c.o ms.:s os.~m mm.:m N:.om ma.m~H :onH>mHma .m o.o om.m~ ms.om :m.ms Hm.moH oHomm .m 0.0 mo.m:H Hm.:OH :m.moH onsz .: o.o ms.sm mm.Hm mammmamzoz .m o.o mm.mm mmcHummmz .N 0.0 mxoom .H 0H m m a m m s m N H .AmHHmcm cH mHmsmcHHHm mom anpwz mocmumHo cam: .OH mHnma 159 NN u z o.o No.NHH mm.mm mm.mm mm.m: m:.mm om.mN mo.mm mH.:N mo.mm Pawammmfivnm>wo ma .oH 0.0 mm.moH mm.NHH mN.Nm mm.H: oo.NmH m:.H: mN.Nm mm.om aoHpmanoqu.q .m 0.0 mm.mNH N:.mm oo.Nm Hm.mo mo.NNH Ho.mHH mm.moH MawcHo 8H .m 0.0 mm.mm mo.Nm Hm.mNH mo.NHH N:.NHH mm.mmH “scam 8H .N o.o mm.mm mm.:N om.mm om.moH mN.o:H aonH>WHwB mg .m 0.0 mm.o: Hm.mm N=.mm om.NmH oHumm mg .m o.o o:.m:H m:.NHH oo.m:H msonsz we .3 0.0 HN.om mm.om xnmcpaos mun .m o.o HN.mN mos>mm mmn .N 0.0 mmn>HH mma .H OH m m N m m z m N H .socmhm Ca mandmcwawm 90m xwfiumz wonmumwn ammz .HH manna 2160 mm.mmmom moans oo.ooH H:.mHH om.mNH mo.mma ma.mma mH.me ¢:.Hma Ho.maa oo.5m mm.Hm unpom voucsooow mocmpmwo AanmcMmMEa can Amway Hoqu mo mowmucoopma o>«uuH:E:o 2:.M5 m5.:m :N.~m :5.5m oo.ooH oo.ooH 5m.mm H:.5m :N.H5 ma.mm unhom voucoooom vacuumwo Hoop mo mowmucoopoa o>wumaseso um.HHn m:.5u m:.mu mN.Nn oo.n ma.» ma.m 5H.oH mo.mm ma.om nupouoo> Hmsofi>fiocw 5n pom voucoooom mocmpmflo mo mmmucmopom 05.mmmmn mo.:~mm| 5o.~mmmu m5.5mHHn mo.n mm.HomH 5o.mom: mm.mm:o :N.5mm5H oa.o~oou unanums popoc>cmwwo mo Amuooav mo=Hm>cwmwm mm.m: mo.mNn HH.mH :m.m :o. 55.mu mm.5mu mm.mau ~:.mu mo.H~ unoECwmupmpcu .oH mm.:m om.m 5o.ma m5.oun mo.a ma. om.mm No.HH No.5au mm.om| :owumspomcH .m 5m.m| 55.mH| mN.5Nu m5.od mo. om. 5m.m~ 5:.m31 mH.o~: mo.m: mofi>oz .o Hm.mau om.:a 3H.0H| :H.oHu No.1 NN.mn :m.o5| mm.om| «n.5o mo.HHn munoam .5 ma.mmn am.mu Hm.om mm.H mo. No.5N mm.H~ ma.wu m5.u: mo.mu co«m«>oHoH .m mm.u~u m5.m5u no.5: 3:.u :o. mo.mHu Hm.: 5m.:m :m.ma ma.ou owomm .m H3.” 5m.mm m5.uu Hm.H 5H. m5.H mm.HH| ma.wN mu.m~| 05.Hm cams: .: Q5.ma mm.5H :m. mm.~5 mH.a Na.mn mH.oH mo.:H :m.mH mm.:5| mumamamzmz .m c:.man mm.oHu. :m.5Hn mm.~: oa.n 25.35 Ha.muu 0:.ma m5.Nm| mm.5:| mocwnmmmz .N ~m.m~| mm.m ma.m5 mm.:n mo.n 5m.mH| mm.mu om.mw| :m.o5| m:.mNn mxoom .H ad a m 5 m m 2 m 5 H .anHmcu cH mHmsmaHHHm pom wmumcHnaooo HMHpmam .NH oHnme 161 mm.mmmd: oomph 00.005 :o.maa m:.m5H :5.m5H mm.m5a mm.o5a ma.55a mo.HHH mm.Hm mm.:m napom omvczooom mocmumfio Amamc«mmafi.ocm Hmopv Hmvou mo mmwmpcmopoa o>wumazfido :m.m5 m5.Hm m:.5m m5.mm oo.ooa oo.ooH 5:.mm mm.5m om.55 mo.m: nanom omuczooom monopmwo Hook «0 mommvcmopmm o>wumasssu mm.5an 05.mu m5.H| 35.: oo. 5m.m 35.5 ma.ma H:.m5 mo.m: unpouomm amsow>«UC5 an new ompcsooom mocmvmflo mo ommvcmosom 5m.:mmm| mm.:m5m| m:.::mu mm.mmm| m5. mm.ooma H5.w:m: mm.o5om mm.mmmma :5.mom55 auxflppmfi pouom>cowwo mo nmpoonv mmoam>cmwwm mH.m: 55.5HI mm.o5| 35.5: 50.: no.5: m5.mu mm.o: om.55u m:.m5n pamEcMmppoucm .oH ma.mm :m.5H mm.H mm.ma :H. mm.HH 5m.md 55.mmn 5m.:m ma.:: coaumE&owcH.4 .m Hm.a om.mn oa.mH 50.5 50.: 35.5: mm.mm mm.5| 5a.:ml 5m.H5s newcwo mg .m m5.m5n 55.35 om.: om.m| mH.I m5. am.5 5m.m: mH.5m 55.mmn mpuomw .5 55.55- :m.5o 55.mH| om.w HH.I :m.m om.om mm.m5| 53.05 oo.:m| :owwfi>waw9 on .m mm.mHu mm.omn Hm.m H5.ma 50.: o:.5Hn mm.55n :o.mma 5m.ma mm.55a owomm .m mm.H 3:.mm m5.n mo.5 mH.n 35.: o:.om| Hm.man 05.53: oo.5m| «scams: mg .2 mm.oa 53.55 5m.5 mm.oH ma. ma.:5l m:.mn 53.5: mo.5: m5.m: xswcpsoo won .m 55.51 55.mHI mm.m mm.m ma. om.m5 50.55: om.H5 mm.m mm.H: monanmwmz .5 H:.:mu mm.m m5.5| 55.mn 55. m5.m| m5.5 mm.mH mo.om| m:.mm mxoom .5 ca m m 5 m m a m 5 H .nocohm cw mamsmcwawm now wmpmcwonoou Hmwpmom .ma manna 162 MUSIC 0 O MOVIES oENTERTAINMENT MAGAZINES 0 O TELEVISION spomso INFORMATION 0 NEWSPAPERS 0 Figure 6. Graphic Representation of English Bilingual Space. N=82. Ii: . .III‘i'fl‘II- .u.J-TQ..1|I.1I‘IIISI'II. . I I111! III I1I..IIIII,III..IIII t‘lv‘.u \ ..I IIII‘IIi (II I I . ,1 ‘lIIII 3|.r‘.‘ . I 163 LE CINEMA o DES LIVRES 0 0 LA MUSIQUE OLE DIVERTISSFMENT DES REVUES ° OLA TELEVISION 0 LA RADIO DES JOURNAUX O O L'INFORMATION 0 LIE SI’CRT A Figure 7. Graphic Representation of French Bilingual Space. N=77. 164 in table fourteen. A t-test was performed to see if these two spaces were significantly different. t equalled 7.96, which was significant beyond the .001 level with df equal to 18. Thus, the null hypothesis of no difference between the spaces can be rejected. Hypothesis Five H5: The degree of discrepancy between a space produced by a group of bilinguals (in either French or English) and a group of monolinguals in one of the bilingual's languages will be less than the discrepancy between the two monolingual groups. The test of hypothesis five was performed in the following manner. The bilingual spaces both in English and French (tables twelve and thirteen) were rotated to a least-square best-fit upon the anglophone space (table seven) and the francophones' coordinate system (table eight). These operations produced a mean discrepancy of 19.28 units between the monolinguals and the bilinguals in English, a mean difference of 21.92 between the English monolingual and the bilinguals' French space, one of 29.10 units between the bilinguals' French space and its monolingual counterpart and a mean distance of 28.95 between the francophones' spatial manifold and the bi- 1inguals in English. The vectors of discrepancy are given in tables fifteen to eighteen. These degrees of discrepancy are all smaller than the 35.59 units between the two 165 Table 1a. Differences Between the Bilinguals' English and French Spaces After Least-Square Rotation on all Ten Dimensions. Concept Differences in Spatial Position Books 31.99 Units Magazines 18.07 Units Newspapers 22.97 Units Music 17.95 Units Radio 8.96 Units Television 19.97 Units Sports 35.23 Units Movies 19.69 Units Information 32.51 Units Entertainment 16.10 Units Mean Difference 21.8“ Units 166 monolingual spaces. Thus, all the relations were in the predicted direction. Next, a series of t-tests (differences of means) were performed to see if the differences between the bi- lingual spaces and ones produced by monolinguals were significantly less than the degree of discrepancy between the two monolingual spaces. The t-test found that the degree of difference between the English monolingual and its bilingual space was significant at the .025 level. (t=2.u8, df=18). The other differences were not significant. The English and the bilinguals in French difference was not significantly smaller than the discrepancy between the two monolingual spaces (t=l.62). This is significant between the .10 and .05 levels. While it is in the predicted direction it does not meet the probability level traditionally required to reject the null hypothesis. The differences between the French and bilingual in English and French and French bilingual were also not significant. t=.87 for the former groups and t=.91 for the later. These results are summarized in table nineteen. In summary, while all four comparisons are in the predicted direction, only one of the four is significant beyond the .05 level. One reached a probability level of 167 Table 15. Differences Between the English Monolingual and the Bilinguals' English Spaces after Least-Square Rotation on all Ten Dimensions. Concept Differences in Spatial Position Books 19.nu Units Magazines 20.02 Units Newspapers 22.72 Units Music 12.53 Units Radio 10.3” Units Television 16.07 Units Sports 29.22 Units Movies 20.92 Units Information 23.38 Units Entertainment 18.15 Units Mean Difference 19.28 Units 168 ITable 16. Differences Between the English Monolingual and the Bilinguals' French Spaces, after Least-Square Rotation on all Ten Dimensions. Concept Differences in Spatial Position Books 30.13 Units Magazines 18.33 Units Newspapers 23.05 Units Music 25.57 Units Radio 13.75 Units Television 21.68 Units Sports 15.99 Units Movies . 22.20 Units Information 30.85 Units Entertainment 17.66 Units Mean Difference 21.92 Units 169 Table 17. Differences Between the French Monolingual and the Bilinguals' English Space after Least-Square Rotation on all Ten Dimensions. Mean Difference 28.95 Concept - Differences in Spatial Position Books 3u.15 Units I'Magazines_ 20.96 Units ‘Newspapers I 23.35 Units _Music 36.05 Units Radio 16.81 Units Television 23.57 Units Sports 5H.29 Units Movies 17.20 Units Information 39.06 Units Entertainment 2n.02 Units Units 170 Table 18. Rotation on all Ten Dimensions. Differences Between the French Monolingual and the Bilinguals' French Space after Least-Square Concept Differences in Spatial Position des livres des revues des journaux la musique la radio I I . . la telev151on le sport . J le Cinema l'information .le divertissement Mean Difference 96.71 29.83 21.65 35.35 21.04 17.38 35.55 26.90 32.52 2u.09 29.10 Units Units Units Units Units Units Units Units Units Units Units 171 Table 19. t-tests Performed to Test Hypothesis Five. Group t Significance (p) English Monolingual- Bilingual in English 2.H8 .:.025 English Monolingual- Bilingual in French 1.62 .<.10 French Monolingual- Bilingual in English .91 >.10 French Monolingual- Bilingual in French .87 >.10 172 .l and the other two failed to reach even this degree of confidence. Thus, while the null hypothesis for hypothesis five cannot unequivocally be rejected, there seems to be a certain level of support for the notion that the degree of discrepancy between the bilingual and one monolingual will be less than the difference between the two mono- lingual spaces. Hypothesis Six H6: The degree of discrepancy between spaces of equivalent terms will increase in the following manner: monolingual language A (English), bilingual in language A, bilingual in B (French) and monolingual in language B. In order to test hypothesis six all spaces were rotated upon all the others' coordinate systems. A graphic representation of all four spaces together is presented in figure eight for explanatory purposes. The discrepancies between them produced the matrix of mean differences presented in table twenty. With the exception of the difference between the francophones' space and the one produced by the French-bilinguals the results are in the predicted order. The means were then entered into a linear trend analysis. That is, the means were correlated with their predicted rank order of discrepancy from the monolingual spaces. In other words, the English monolingual system was assigned a position of zero. The discrepancy from the English monolingual space was one for the English bilingual space, two for the French 173 bilingual manifold, and three for the francophones. Like- wise, the discrepancy between the English bilingual space and the franc0phone system was assigned a value of two and the French bilingual francophone discrepancy a one. These values produced a correlation of .65LI8.6 A significance test was then performed on this coefficient which failed to reject the null hypothesis at the .05 level (F=6.00, df=l,5). Thus, although the means are in the predicted order, their placements are not significant and therefore, the null hypothesis of hypothesis six cannot be rejected. Hypothesis Seven H7: The pattern of media usage will be significantly different for mono- linguals in language A (English), monolinguals in B (French) and bi- linguals living in the same geographical location. Based on the data from questions 21 through 25 and 27 to 29, the null hypothesis of no difference between the pattern of media use for the two monolingual groups and bilinguals can be rejected. The English language sample on the average used the media a total of 51.H hours per week, as compared to the French subjects who used the media an average of 39.1 hours per week. This difference 17H mH 55 Nm omH z o.o oH.NN mm.NN mm.mm gunman 0.0 :N.HN Nm.HN nocmnmuHmsmcHHHm 0.0 NN.NH amHHmnmuHmsmnHHHm o.o pmHHmam Hmswcfiaocoz nonopm Swwdmcm HOSMCNHOCOZ nocmgm HmswcHHHm HammaHHHm :mHHmcm .mmommm Ham cmngmm mowocmaonomfln mo Nahum: .O5 magma I75 MUSIC 0’0 J J 7 XMOVI ES LENTERTAINMENT RADIOa//8 a o v cko KS P TELEVISION MAGAZINESZ/ SPORTS [,0 INFORMATION NEWSPAPERS Figure 8. of All Groups. Graphic Representation of Combined Spaces The English monolingual space is closest to the label, followed by the English bilingual, French bilingual and finally, the French monolingual at the end of the connecting line. 176 is significant at the .05 level. This direction is maintained with all specific media except television. Television's use was significantly greater for the franco- phones (p:.02). They used TV on the average 10.9 hours per week as compared to 6.1 for the anglophones. This relationship was in reverse of the expected direction based upon the availability of the media. The other media (film, newspapers, magazines, radio, records/tapes and books) were all used to a greater degree by the English language group than the French language group. The differences were all significant at the .05 level, except records, which did not achieve significance at that level, and books, where a probability of .001 existed in the data. While their total number of media hours per week (5H.5) exceeded both the French (39.1) and the English (5.1) monolinguals, the rest of the figures place the bi- linguals between the other two groups. Of the 50.5 hours, “0.7 (76 percent) are in English and 13.8 (2” percent) are in French. This compares to #7.? hours (93 percent) in English and 3.6 (7 percent) in French for the anglophones, and 7.9 hours (19 percent) in English and 31.7 (81 percent) in French for the francoPhones. These relationships are significant beyond the .001 level in the predicted direction. With the exception of magazines, radio and records/tapes, these hours are moderate between the other two groups. For each individual medium the usage patterns in each language for bilinguals was moderate between the 177 franc0phones and anglophones. These results are summarized in table twenty-one. It should be noted that the overall pattern of media use in the three groups is quite similar, indicating that the structure of media use is very homogeneous for the monolinguals and bilinguals. The rank order correlation of the hours used of the seven media between the anglophones and bilinguals is .96. It is .71 between the anglophones and francophones and .75 between the bilinguals and francophones. Again, although bilinguals' patterns are more similar to the anglophones, this finding indicates that the structure of media use for bilinguals is moderate between the two monolingual groups. This analysis was extended to interpersonal communication (question 25), under the assumption that due to the bilinguals' embeddedness in the communication structure of society, they would tend to use interpersonal channels to a greater degree than monolinguals in order to transfer information across the language boundary. This notion was supported. Bilinguals used interpersonal communication more than monolinguals according to the data. On the average, they interacted a total of 30.7 hours per week with friends and family. Of these hours, 13.9 (93.7 percent) are in French and 17.3 (56.3 percent) are in English. This compared to 25.5 (2u.u or 95.7 percent 178 Table 21. Media Use Patterns of Bilinguals, English and French Monolinguals. English Bilingual French Hrs. % Hrs. % Hrs. % Television English 5.8 (.89) ”.5 (.66) 1.9 (.17) French .7 (.11) 2.3 (.30) 9.0 (.83) TOTAL 6.1 6.8 10.9 Radio English 9.5 (.83) 11.0 (.79) 1.0 (.11) French 1.9 (.17) 3.9 (.26) 8.3 (.89) TOTAL 11.” 19.9 9.3 Film English 2.5 (.93) 2.0 (.83) .1 (.07) French .2 (.07) .H (.17) 1.4 (.93) TOTAL 2.7 2.9 1.5 Newspapers English “.8 (.96) 3.2 (.78) 0.0 (.00) French .2 (.08) .9 (.22) u.1 (1.0) TOTAL 5.0 9.1 “.1 Books English 19.7 (.99) 8.6 (.79) .8 (.17) French .2 (.01) 2.5 (.26) 5.9 (.83) TOTAL 14.9 11.1 6.7 Magazines English 2.7 (.96) 2.3 (.72) 0.0 (.00) French .1 (.08) .9 (.28) 1.0 (1.0) TOTAL 2.8 3.2 1.u Records/Tapes English 8.2 (.97) 8.5 (.77) 3.6 (.69) French .3 (.03) 2.6 (.23) 1.6 (.31) TOTAL 8.5 11.1 5.2 Total English u7.7 (.93) no.7 (.75) 7.u (.19) French 3.6 (.07) 13.8 (.29) 31.7 (.81) Grand Total 51.u (1.0) su.5 (1.0) 39.1 (1.0) N = 150 159 15 179 in English and 1.1 or ”.3 percent in French) for the anglo~ phones and 2”.2 (22.6 or 93.” percent in French and 1.6 or 6.6 percent in English) for the francophones. These differences are significant at the .05 level and are summarized in table twenty-two. In summary, hypothesis seven can be supported. While the overall structure of media usage is quite similar between the three groups, bilinguals use the total media and interpersonal communication more than either mono- lingual groups. Anglophones use the media significantly more than francophones with the exception of television, where the pattern is reversed. The pattern of use of the individual media for bilinguals is moderate between the two monolingual groups based upon hours and the language of use. Hypothesis Eight H8: The greater the proportion of mass media and interpersonal exposure in language A (English) (the less the proportion of exposure in language B - French), the greater the discrepancy of the space of the individual bilingual from the mean space of language B (French) monolinguals and the less the discrepancy from the average space of language A (English) monolinguals. This is for a space produced by translated equivalents. In order to test this hypothesis 61 bilinguals in English and 60 bilinguals in French who completed all 180 .Hoo. H.m o:o>mn cowpomnwo OOHONUONQ on“ a“ Pcmowmwcwwm ea .mo. ”(a cowpom9wo oopowoopg may cw “snowmacmfim e Amm. V m.55 550. v m.H Aoo.Hv 5.35 w madom mHMdmcHHocoz nocwnm A33. v 3.mH Amm. V m.5H moo.Hv 5.0m w whoom HmswcHHHm A39. v H.H «ewnsom HMGOONOQNOHOH nocmnm Amm. V 3.35 «ewpsom HocOONOQNOHOH :mwawcm Aoo.HV m.mw w mpsom «mason HOOOmhomhmch HONOR mamswcwaocoz anHmam .mamsmcwaocoz gocmnm paw mesmcflme .mamsmnwaocoz .Amfiamcm CONHOONGOEEOQ HMGOmpomnmch mo acmwpmmeoo .55 magma 181 ”5 pair-comparisons were individually rotated to a least- square best-fit upon the aggregate space of both mono- lingual groups.7 The degree of discrepancy for each individual was then calculated in the same manner as for the groups in the earlier hypotheses. These values were then entered into regression equations as the dependent variables with the proportion of mass media usage and interpersonal communication in each language as the independent variables. These results failed to reject the null hypothesis. The zero-order correlations of the proportion of mass media and interpersonal communication in English with the degree of discrepancy from the French mono- linguals' space were .20 and .07 respectively. The former coefficient is significant at the .0” level (F=”.35, df= 1,118) and the later did not achieve a .05 level (F=l.16, df=l,118). The zero-order correlations of the proportion of the mass media and interpersonal communication in French with the degree of discrepancy from the English monolingual space were .0” and .18. The later coefficient was significant at the .02 level (F=5.22, df=l,118) and the former did not achieve a .05 level (F=.l7, df=l,118). Additionally, it was in the Opposite of the predicted direction. With these sets of zero-order correlations, multiple correlations were calculated. The attempt to predict the degree of discrepancy from the French space 182 from the two variables produced a multiple correlation of .20 which when squared was .0”. This was not significant at the .05 level (F=2.”9, df=2,ll8, p:.088). Likewise, the attempt to predict the degree of discrepancy from the English monolingual space with the language of mass and interpersonal communication was not significant at the .05 level. The multiple correlation was .21 and its square .0”. (F=2.70, df=2,ll8, p:.07). It is worth noting that in one case the language of mass communication was responsible for the size of the multiple correlation, while in the other the language of interpersonal communication explained most of the variance. In the attempt to predict the English discrepancy, the mass media variable had a beta weight of .13, while the interpersonal variable's beta was .22; In the French case, mass media had a beta of .”5, while the interpersonal communication's beta was .03. These two tests may be taken as an attempt at cross validation of the predictive 8 equation. Because of the unstability of the beta weights this hypothesis may be rejected. Summary In summary, this chapter has reported the results of the test of eight hypotheses dealing with the organization of semantic information by bilingual individuals, how their system of organization relates to monolinguals and the causes for the arrangement of concepts 183 in their semantic system. Of these eight, four were confirmed, one was partially supported and the null hypothesis could not be rejected for three hypotheses. The implications of these results for the theory of bi- lingualism and the paradigm for the measurement of meaning presented in earlier chapters will be discussed in the following chapter. 18” FOOTNOTES FOR CHAPTER FIVE 1The use of M.D.S. requires that the dimensions be taken into account simultaneously (Torgerson, 1958). Thus, the notion of identifying individual dimensions may in fact be running against the spirit of the method. However, for ease of explanation and interpretation, this approach was adopted. In factor analysis because the data matrices are standardized, only the unique con- tributions of each factor can be considered, rather than the interrelations of the concept loadings on these factors. By attaching an attribute label to one dimension of the space, it is implied that all the variance of the dimension could be explained by tHE? attribute. This need not be the case and may only lead to confusion. A dimension should only be used to refer to the orthonormal reference vector and never to an attribute. Two variables may interact in such a way as to produce the variance on one dimension. Additionally, if an attribute is present in a space it may be congruent with the variance on two or more dimensions. Most likely, the attribute's variance can be explained by multiple vectors. Finally, the labelling of dimensions is a post hoc intuitive pro- cess which, although verifiable through regression pro- cedures, should be avoided. 2The seventh or zero vector was removed from the analysis. The reason is that any N concepts may be plotted into a space of N-l dimensions without any loss of information. For example, any two points may be connected by a line, yielding a single dimension upon which to differentiate the points. Three points may be connected by a plane. Likewise, four points may be connected in a cube and any N points in a hyperspace of N-l dimensions. However, if any three or more points lie along a single continuum, fewer dimensions would be needed to precisely describe the system. 3As the number of variables (columns or dimen- sions) approaches the number of cases (concepts), the multiple correlation approaches 1.0. In this case, there were nine factors and only ten concepts. The R2 was equal to 1.0. l'An argument against the use of the scree test and the notion of only using a limited subset of the dimensions in the space may be made from these results. Had only the first two dimensions been used to describe the space, almost no variance could be attributed to the language variable. This obviously would have been an erroneous conclusion. On the dimensions to the right of 185 the elbow language produced considerable variance, which, in the light of the total variance in the space is limited, but nonetheless present. Thus, all dimensions probably should be included in the analysis to avoid fallacious interpretations. 5The formula for the R2 corrected for shrinkage is: R'2 = 1 - (l-R2)N - 1/