Florida State University Libraries

Electronic Theses, Treatises and Dissertations The Graduate School

2005 Continuity and Change in Islamic Ethnopharmacological Practice: New Methods for Cognitive Dialectometry Kevin D. Pittle

Follow this and additional works at the FSU Digital Library. For more information, please contact [email protected] THE FLORIDA STATE UNIVERSITY

COLLEGE OF ARTS AND SCIENCES

CONTINUITY AND CHANGE IN ISLAMIC ETHNOPHARMACOLOGICAL PRACTICE: NEW METHODS FOR COGNITIVE DIALECTOMETRY

By

KEVIN D. PITTLE

A Dissertation submitted to the Department of Anthropology in partial fulfillment of the requirements for the degree of Doctor of Philosophy

Degree Awarded: Fall Semester, 2005

Copyright © 2005 Kevin D. Pittle All Rights reserved The members of the committee approve the dissertation of Kevin D. Pittle Defended on October 21, 2005.

______Judy K. Josserand Professor Directing Dissertation

______Peter P. Garretson Outside Committee Member

______Glen H. Doran Committee Member

______Bruce T. Grindal Committee Member

Approved:

______Dean R. Falk, Chair, Department of Anthropology

The Office of Graduate Studies has verified and approved the above named committee members

ii ACKNOWLEDGEMENTS

I would like to express my deep gratitude to the many people who have helped me to complete this dissertation. First, I would like to thank my dissertation supervisor, J. Kathryn Josserand for her commitment to excellence and her unwavering support throughout the course of my studies. In addition, I would also like to thank her husband, Nicholas A. Hopkins, a mentor to whom I owe much of the inspiration behind this research. I am also grateful to my doctoral committee, Dr. Glen Doran, Dr. Bruce Grindal, and Dr. Peter Garretson, who have all contributed significantly to my formation as a scholar. I would like to thank the Garden Club of America, whose generous funding trough the Anne S. Chatham Fellowship in Medicinal Botany made this research possible. In addition, I would like to thank the Florida State University for awarding me a Dissertation Research Grant to complete this project. I am grateful to Dr. James Miller, Mr. Riadh Saadaoui, and the staff at Le Centre d'études maghrébines à Tunis for their assistance during the field component of this project. I would also especially like to thank Khaled for his friendship and hard work, as well as Jamal, Tawfik, Sabr, and the other herbalists who shared their knowledge of healing plants with me. I would like to acknowledge my debt to Bonnie Brown, who edited an early draft of this manuscript, and to David Russell who helped with the figures. Any deficiencies that remain are my own responsibility and do not reflect your excellent work, but rather, my own recalcitrance. Finally, I wish to thank my beloved family. Without the help and inspiration of my parents, Marshall and Eileen Pittle, and my grandmother, Esther Danto, I would not have embarked on this journey. My in-laws, Marge and Bill Glauch also provided a great deal of support and encouragement, for which I am thankful. I owe my deepest gratitude to my wife, Natasha, and to our dear children, Emeth, Mekerah, and Jonathan, who endured much, encouraged much, and without whom I would not have been able to complete this work.

iii TABLE OF CONTENTS

List of Tables …………………………………………………………………………………………………………………………………………………………… vi List of Figures ……………………………………………………………………………………………………………………………………………………… vii Abstract …………………………………………………………………………………………………………………………………………………………………………… xi

INTRODUCTION: SECOND GENERATION COGNITIVE ANTHROPOLOGY AND THE PROBLEM OF HISTORY ………………………………………………………………………………………………………………………………………………………………………………… 1

Toward a Solution for the Problem of History ……………………………………………………………… 3 Necessary Assumptions …………………………………………………………………………………………………………………………… 4 General Research Hypotheses …………………………………………………………………………………………………………… 6 Significance of the Study ………………………………………………………………………………………………………………… 6 Delimitations ………………………………………………………………………………………………………………………………………………… 8 Preliminary Definitions and Key Terms ……………………………………………………………………………… 10

1. BACKGROUND: ISLAMIC MEDICINE IN CONTEXT ……………………………………………………………………………… 14

The “System” of Islamic Medicine …………………………………………………………………………………………… 14 Philosophical Bases of the System ………………………………………………………………………… 17 Prevention and Treatment in the System …………………………………………………………… 19 Underreporting of an Important Aspect of the System: Herbs ……… 20 The Sources of Middle Eastern Medicine …………………………………………………………………………… 21 Ancient Near Eastern Medicine …………………………………………………………………………………… 21 Greek Humoral Medicine ……………………………………………………………………………………………………… 25 Prophetic Medicine and Popular Movements ……………………………………………………… 28 Problems with Identifying the “Sources” of Islamic Medicine …… 29 Relevant Historical Interactions of Middle Eastern Societies ………………… 30 The Pharmaceutical Trade ………………………………………………………………………………………………… 32 Traveling Physicians, Traveling Texts ……………………………………………………………… 34 Trade and Pilgrimage Routes ………………………………………………………………………………………… 36 Socio-Political Factors …………………………………………………………………………………………………… 43 Linguistic Factors ………………………………………………………………………………………………………………… 44 Shifting Intellectual Centers …………………………………………………………………………………… 51 The Significance of Historical Interactions for the Development of Islamic Medicine ……………………………………………………………… 52

2. LITERATURE REVIEW: THEORETICAL AND METHODOLOGICAL SOURCES FOR COGNITIVE DIALECTOMETRY AND DIACHRONIC PHENETICS ………………………………………… 54

Cognitive Categories: G1 and G2 Perspectives and Instruments ………………… 54 First-Generation Cognitive Studies ……………………………………………………………………… 55 Second-Generation Cognitive Studies …………………………………………………………………… 67 Taxonomy, Phylogeny, and Change ……………………………………………………………………………………………… 74 Biological Models …………………………………………………………………………………………………………………… 75 Linguistic Models …………………………………………………………………………………………………………………… 84 Cultural Models ……………………………………………………………………………………………………………………… 104

iv Implications of Prior Theory and Method for the Present Research …… 123

3. METHODS: MEASURING RELATIONSHIPS BETWEEN COGNITIVE DIALECTS …………………………………………………………………………………………………………………………………………………………… 124

Design Parameters of the Research Project ………………………………………………………………… 124 Derivation of Specific Hypotheses and Associated Variables …………………… 131 Sources and Sampling Procedures …………………………………………………………………………………………… 133 Data Collection and Analysis: Techniques and Measures ………………………………… 136 Limitations of the Study ……………………………………………………………………………………………………………… 143

4. RESULTS: A NUMERICAL TAXONOMY OF ISLAMIC MEDICAL SYSTEMS AND THEIR ANTECEDENTS ………………………………………………………………………………………………………………………… 145

Description of the Sample …………………………………………………………………………………………………………… 145 Group I, Modern Sources Only …………………………………………………………………………………… 145 Group II, Pre-Modern and Modern Core Sources ………………………………………… 148 Results …………………………………………………………………………………………………………………………………………………………… 152 The Cognitive Structures of Islamic Medical Systems and Their Antecedents ………………………………………………………………………………………………… 152 Relationships and Influences among Sources ……………………………………………… 161 The Efficacy of Various Instruments and Measures ……………………………… 185

CONCLUSIONS: THE EFFICACY OF A PHENETIC APPROACH TO COGNITION …………………………… 188

Overview of the Study ……………………………………………………………………………………………………………………… 188 Toward the Measurement of Cognitive Dialects of Islamic Ethnomedical Practice ……………………………………………………………………………………… 188 Procedures for Measuring Cognitive Dialects …………………………………………… 190 Research Hypotheses Considered ……………………………………………………………………………… 193 Results and Conclusions ………………………………………………………………………………………………………………… 193 Major Findings ………………………………………………………………………………………………………………………… 194 Conclusions and Interpretation ……………………………………………………………………………… 194 Implications ……………………………………………………………………………………………………………………………………………… 199 Theory ……………………………………………………………………………………………………………………………………………… 199 Methodology ………………………………………………………………………………………………………………………………… 200 Recommendations for Future Research ………………………………………………………………………………… 200

BIBLIOGRAPHY ……………………………………………………………………………………………………………………………………………………………… 199

BIOGRAPHICAL SKETCH …………………………………………………………………………………………………………………………………………… 221

v LIST OF TABLES

1. Similarity Rankings of Sources in Group I, in Arbitrary Units …………………… 172

2. K-Means Clustering Results for Sources in Group I (A) ……………………………………… 175

3. Similarity Rankings of Sources in Group II, in Arbitrary Units ………………… 178

4. K-Means Clustering Results for Sources in Group II (UA/G) ………………………………… 182

vi

LIST OF FIGURES

1.1 The Middle East and Central Asia: Political Boundaries, 1990 ……………………… 31

1.2 The home of myrrh and frankincense …………………………………………………………………………………………… 33

1.3 Ibn Battuta’s Return itinerary from China to North Africa, 1346-1349 … 38

1.4 External Trade Routes of India ……………………………………………………………………………………………………… 39

1.5 Trade Routes Connect the Middle East and Asia ……………………………………………………………… 40

1.6 Some Principal Lines of Trade in Africa and the Middle East ………………………… 41

1.7 African Pilgrimage Routes to Mecca, ca. 1300-1900 …………………………………………………… 42

1.8 Muslim schools of law and Sufi brotherhoods: c. 1500 …………………………………………… 45

1.9 The Eastern Mediterranean and the Middle East, C.E. 600 …………………………………… 46

1.10 The Spread of Islam ……………………………………………………………………………………………………………………………… 47

1.11 Major Middle Eastern and Central Asian Languages ………………………………………………… 49

2.1 Levels of Contrast in “Skin Disease” Terminology ……………………………………………………… 57

2.2 Levels of Branching Diagram ……………………………………………………………………………………………………………… 58

2.3 Comparison of diagrammatic representation of the semantic structure of a fragment of Hanunoo classification of peppers using a tree diagram (top part of figure) and box diagram (bottom part of figure) …………………………………… 58

2.4 Taxonomic structure derived from a pile sorting of home garden produce 59

2.5 Critical Contrasts Differentiating the ‘Sores’ …………………………………………………………… 61

2.6 Mean semantic differential factors …………………………………………………………………………………………… 63

2.7 Diagrammatic representation of Romney feature analysis of English male kin terms ……………………………………………………………………………………………………………………………………………………………… 64

2.8 Example of Sort Table ……………………………………………………………………………………………………………………………… 65

2.9 Equivalent Response Sets ……………………………………………………………………………………………………………………… 65

2.10 American-English Belief-Frames ………………………………………………………………………………………………… 71

vii 2.11 Taxonomy for American-English Disease Terms ……………………………………………………………… 73

2.12 ‘Two-Dimensional Scaling and Clustering Model of 33 A’ara Verbs of Interpersonal Conflict’ …………………………………………………………………………………………………………………… 74

2.13 Phenogram of 81 species of gulls and terns based on UPGMA cluster analysis of correlation coefficients of 51 skeletal measurements ……… 80

2.14 Determination of the most parsimonious cladogram using Hennegian argumentation ……………………………………………………………………………………………………………………………………………… 83

2.15 The Indo-European family tree …………………………………………………………………………………………………… 88

2.16 A wave diagram of the Germanic family ……………………………………………………………………………… 91

2.17 Some overlapping features of special resemblance within the Indo- European languages, conflicting with the family-tree diagram ………………… 92

2.18 Pidgin English in the Pacific around 1880 ………………………………………………………………… 103

2.19 Linguistic Influences on Tok Pisin around 1900 …………………………………………………… 103

2.20 Listing, classifying, analyzing, and system: data map ………………………………… 104

2.21 The morphology of a Culture Region (after D. Meinig) …………………………………… 110

2.22 Principal Sun Dance influences ……………………………………………………………………………………………… 112

2.23 A phylogeny of nine Kenyan pastoralist cultures ………………………………………………… 115

3.1 The interactive model of research design used in this study ……………………… 125

3.2 Basic relationships among groups, techniques, and measures ………………………… 129

3.3 Path Diagram of Hypotheses ……………………………………………………………………………………………………………… 133

3.4 Diagram of procedures for data collection and analysis …………………………………… 137

4.1 Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions from Aleppo, Syria ………………………………… 154

4.2 Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions from Cairo, Egypt …………………………………… 155

4.3 Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions from Gaziantep, Turkey ……………………… 156

4.4 Hierarchical clustering tree representing the distributional similarity of drug plants from India, as classified by their Ayurvedic properties ………………………………………………………………………………………………………………………………………………………… 157

4.5 Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions from Karachi, Pakistan ……………………… 158

4.6 Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions from Marrakech, Morocco …………………… 159

viii 4.7 Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions from Sanaa, Yemen …………………………………… 160

4.8 Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions from Modern Syria …………………………………… 162

4.9 Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions from Modern Egypt …………………………………… 163

4.10 Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions by al-Kindi ……………………………………………… 164

4.11 Hierarchical clustering tree representing the distributional similarity of drug plants noted by ibn Wafid, as classified by their humoral properties …………………………………………………………………………………………………………………………………………………… 165

4.12 Hierarchical clustering tree representing the distributional similarity of drug plants noted by Chisti (Pseudo-Avicenna), as classified by their humoral properties ……………………………………………………………………………………………………………… 166

4.13 Hierarchical clustering tree representing the distributional similarity of drug plants used in c. 400-500 A.D. Greco-Syriac prescriptions (Syriac α ……………………………………………………………………………………………………………………………………………………… 167

4.14 Hierarchical clustering tree representing the distributional similarity of drug plants used in c. 200-300 A.D. “native” Syriac prescriptions (Syriac β ………………………………………………………………………………………………………………………………………………………… 168

4.15 Hierarchical clustering tree representing the distributional similarity of drug plants used in c. 800-900 A.D. Coptic prescriptions (P. Chassinat …………………………………………………………………………………………………………………………………………… 169

4.16 Hierarchical clustering tree representing the distributional similarity of drug plants used in c. 1500 B.C. Egyptian prescriptions (P. Ebers) …………………………………………………………………………………………………………………………………………………… 170

4.17 Complete Linkage Cluster Analysis of Sources in Group I Pearson Correlation Coefficient of A …………………………………………………………………………………………………… 173

4.18 Complete Linkage Cluster Analysis of Sources in Group I Pearson Correlation Coefficient of UA …………………………………………………………………………………………………… 173

4.19 Two-Dimensional MDS Solution for Sources in Group I, UA ……………………………… 174

4.20 Plot of Principal Components of Similarity Data, A ………………………………………… 174

4.21 Complete Linkage Cluster Analysis of Sources in Group I Pearson Correlation Coefficient of UA …………………………………………………………………………………………………… 179

4.22 Complete Linkage Cluster Analysis of Sources in Group I Pearson Correlation Coefficient of UG …………………………………………………………………………………………………… 179

4.23 Two-Dimensional MDS Solution for Sources in Group II, UA …………………………… 180

4.24 Two-Dimensional MDS Solution for Sources in Group II, UG …………………………… 180

ix 4.25 Plot of Principal Components of Similarity Data, UA ………………………………………… 181

5.1 Principal ethnopharmacological influences as reconstructed from values in Table 1 ……………………………………………………………………………………………………………………………………………………………… 197

x ABSTRACT

This research project entailed an investigation of whether the degree of similarity between various Islamic and pre-Islamic Middle Eastern societies’ overall patterns of drug plant prescription, as calculated using principles of numerical taxonomy, would correlate with the known facts of culture area morphology and the succession of intellectual traditions in the region. The attempt to quantify similarity in overall pattern is tantamount to a “cognitive dialectometry” of Islamic ethnopharmacology and its precursors, and is a first step in the development of a comparative historical approach to cognition analogous to that used in linguistics. The study considers fourteen sets of prescriptions, or “native” descriptions of medicinal attributes of drug plants, composed between 1534 B.C. and the present. For each source, patterns of grouping were identified by applying a hierarchical clustering program to a data matrix reflecting its drug plant prescription/attribute correlations. The resulting cluster trees were treated as pile sort results. Traditionally, pile sorting is a technique where cultural consultants are asked to sort items into groups based on their similarity. Shared groupings across sources were tallied, various means and functions of similarity were calculated based on sharing of groupings among sources, and degrees of overall similarity among sources were modeled using a battery of four key statistical techniques: hierarchical clustering, multidimensional scaling, k-means clustering and factor analysis. A numerical taxonomy approach to Islamic medicine shows a clear relationship between the proximity and shared history of contemporaneous localities and their overall degree of similarity in practice. It also shows that the degree of similarity between sources from different time periods correlates with the relative strength of their presumed relationships of descent and influence. The results substantiate the existence and measurability of “cognitive dialects” analogous to linguistic lects and allow for the possibility of future analyses of “cognitive creolization.”

xi INTRODUCTION: SECOND-GENERATION COGNITIVE ANTHROPOLOGY AND THE PROBLEM OF HISTORY

This study was conducted to address a methodological and theoretical gap in cognitive anthropology. It represents one possible way to apply a historical approach to cognition, illustrated through a comparative case study of Middle Eastern herbalists’ folk medical knowledge, beliefs, and practices. Since the late 1950s, cognitive anthropologists have investigated numerous cultural organizations of mental and material phenomena, centering their work on questions of meaning, knowledge, and concept formation (Tyler 1969:3 and D’Andrade 1995:12). These cognitive studies can be divided into two distinct types of inquiry which have been variously labeled by adherents of the field. For the purposes of the present discussion the two main versions of cognitive science may most conveniently be labeled “first- generation” and “second-generation” studies (see Lakoff and Johnson 1999:75- 78). First-generation cognitive studies, noting the “language-like” nature of cognition (Tyler 1969), borrow their basic ideas by analogy from Saussurian synchronic linguistics. Cultural categories (or “taxa”) are thus seen as arbitrary signs best defined by the presence or absence of selected contrastive features. Such taxa are ordered into hierarchical “folk taxonomies” by relations of inclusion and contrast, which are contingent on the presence or absence of these necessary and sufficient features (D’Andrade 1995:22, 34, 35). In the 1970s, cognitive anthropologists discovered that not all cultures clearly delineate categories based on the presence or absence of only a few minimum features. Nor do they necessarily order their categories into hierarchical “kinds of” relationships. Rather, second-generation studies reveal that categories frequently have “fuzzy” boundaries (Shaul and Furbee 1998:147) and are primarily based on the Gestalt apprehension of

1 perceptual attributes. Category members “belong” because they are perceived as similar to prototypical “clearest” or “best-case” exemplars which possess the highest average of properties in the pool of attributes present in the larger category (D’Andrade 1995:118, 119). Taxa are gradient categories, and their members share Wittgensteinian “family resemblances” expressed at varying levels of intensity (Shaul and Furbee 1998:149). Until the present study, only first-generation studies have entertained the possibility of using the tools of historical linguistics to analyze cognitive systems from a diachronic perspective (e.g., Friedrich 1970, Frisch and Shutz 1967, Hockett 1964, and Merrifield 1981). Second-generation studies have been more concerned with questions of intracultural variation and consensus than of history. However, their focus on variation parallels a shift in historical linguistics to an approach that looks at variable “developmental” patterns and mechanisms of language in diachronic motion. These developmental studies define “lects” (i.e., languages, dialects, or sociolects) as evolving, complexive, constellation-like webs of implicationally-ordered features (Bailey 1996 and Müllhäuser 1985). Other linguists, while not focusing so much on variation as the developmentalists, have undertaken language and dialect classifications based on global investigations of similarities through “mass comparison” of their features (Greenberg 1974 and Ruhlen 1987, 1994). Some have also attempted to recognize patterns of relationship through numerical classifications based on the presence or absence of lexical, morphological, and syntactic items (cf. Embleton 1993 and Dyen 1975). Linguists are not the only researchers who have developed methods for classifying their objects of study based on “family resemblances.” Practitioners in several other fields of scientific inquiry have also been working on the same kinds of classificatory problems. Biologists, linguists, and archaeologists have all used variations on phenotypic (or, more properly, “phenetic”) and cladistic approaches to categorize cases and clarify historical relationships in their respective domains of study. The “fuzzy” nature of folk categories suggests that a profitable way to engage the problem of history in second-generation cognitive studies might be through an analogical approach. Such an approach classifies cognitive variants (i.e., cognitive dialects) based on “family resemblances” in how similar their similarity judgments are to one another.

2 Toward a Solution for the Problem of History

This project is an attempt to develop quantitative mechanisms for making accurate inferences about historical relationships of descent and influence among cognitive systems. More specifically, the example of drug- plant prescription in Islamic societies will be used to confirm or refute the accuracy of phenetic techniques (also known as “numerical taxonomy”) when applied to similarity judgment data. Numerical taxonomy is, at its simplest, a method for quantifying similarity of overall form. I will test the usefulness of phenetic techniques for examining cognitive history through the application of a five-point process to recipes for compound drug-plant prescriptions and statements regarding drug-plants’ medicinal properties that I have selected from fourteen different sources. Six of the sources derive from late twentieth-century Islamic societies (e.g., Egypt, Syria and Morocco) and the others from historical antecedents in the region (e.g., Ancient Egyptian), as well as from one neighboring non- Islamic society chosen as an “out-group” (namely, India). The process involves the following: 1. The use of a hierarchical clustering computer program to model use-based similarity judgments for the drug-plants present in each source’s prescriptions or descriptions of drug-plant properties. 2. Comparison of the individual sources’ results using a “sort table” technique developed by Agar (1996), modified to convert the similarity judgments into “features.” 3. Calculation of the degree of similarity among the sources, based on the features derived in this manner. 4. Use of the resulting similarity ratings to compare the sources and to create classifications for them based on overall family resemblances in the way the sources group their plants. 5. Examination of the classifications and historical inferences generated from the above method, in light of the geographic and historical facts of the case, as they are known from other non-cognitive approaches to the history of Islamic medicine. In an effort to take a first step toward filling the “historical void” in second-generation cognitive anthropology, this study will (a) provide an overview of the geographic and historical facts of the case study (i.e., Islamic medicine in its geographic, historical, and cultural context), (b) 3 review the development of theory in cognitive studies and of classificatory and evolutionary theory in other domains, (c) discuss techniques that have been used to look at similarity and variation in cognition and describe and evaluate methods that have been used to look at the evolution and historical relationships of systems in other scientific fields, and (d) apply modified versions of some of these techniques to the second-generation problem presented by the case study.

Necessary Assumptions

Several assumptions underlie this study. These will be presented, with their justifications, in the paragraphs below. First and foremost: It is assumed that one can speak of “Islamic societies” or “the Middle East” (and therefore, of “Islamic” or “Middle Eastern medicine,” as well) as real, researchable entities.

According to Eickelman, the term “Middle East,” “encompasses the region stretching from Rabat to Tehran . . . [and] is often extended to include Afghanistan and Pakistan” (2002:2). While there is significant cultural and linguistic diversity across the region, there are strong general trends of commonality in many domains of culture (including medicine, cf. Ullman 1978 and Ushmanghani et al. 1986). Some historians and social scientists have also referred to societies in this region by the adjective “Islamic” (Lapidus 2002, Ullman 1978). While not all members of these societies are Muslims, non-Muslims constitute a socio-politically dominated minority in the region, and have been profoundly influenced in their worldview and lifestyle by the habitus (see Bourdieu 1990) of the people they live among. As Esposito (1999:x) observes, “Islam . . . is the dominant symbolic and ideological force” in the part of the world under investigation. In this study, the terms “Middle Eastern” and “Islamic,” will be used interchangeably, with the understanding that non-Muslims participate, to varying degrees, in what may be termed “Islamic” societies.

Second: It is assumed that the sources selected are representative of other contemporary herbalists and other drug-plant-prescribing medical practitioners in the locality where they ply their trade.

4 While it is clear from prior studies that specialized cognitive knowledge is variable and distributed throughout a society, it has also been shown to be shared enough that professionals can be said to take part in a common knowledge and belief system with each other and their clientele (Garro 1986). From a purely economic perspective, herbalists could not last long in the profession without meeting expectations for the provision of relevant prescriptions to their clientele. Third: It is assumed that named entities identified as the same plant (at the generic level of classification) by professional herbalists, botanists, or lexicographers may be treated as the same entity.

While ethnobotanists have shown that different cognitive systems may partition the world in sometimes radically different ways, they have also found that “generic taxa often approximate in their content the genera and species of Western biology” (Berlin 1992:25-26). Therefore, given a plant named x in society A and a plant named y in society B, both identifiable as representative of the same genera in Western biology, x and y may be conceived of as “the same plant.” In addition, even though plant names occasionally vary across the culture area under study, natives of the region have long recognized many terms as synonyms for the same entity. If a term used by one source has been historically identified as synonymous with a different term used by another source, then these two terms can be conceived of as equivalent. Fourth: It is assumed that the drug-plants most often prescribed together for the same illnesses will be thought of as more similar to one another than drug-plants prescribed together less frequently (or not at all).

Cognitive anthropologists have demonstrated that named members of highly useful domains (such as medicines) may be organized into taxonomies by their “distributional similarity” (the degree to which the items occur in the same environments or have the same practical uses), as much as by a purely intellectual classificatory “search for order” in the world (D’Andrade, et al. 1972 and Morris 2000:70). Finally: It is assumed that cognitive systems evolve in ways that are analogous with other evolving systems and may therefore be fruitfully studied in analogous ways.

5 Cultural knowledge systems (such as Islamic medicine) exhibit the basic features necessary for evolutionary change: variation, selection, and transmission. They consist of co-adapted idea units or “memes” that vary, are selected, and are transmitted in ways that are analogous, though not identical, to other replicators such as genes (Aunger 2002:3,50 and Dawkins 1989:189-201) and “linguemes” (Croft 2000:28). They are therefore most appropriately examined with analogous classificatory and historical methodologies, such as numerical taxonomy. The five primary assumptions constrain and direct the kinds of hypotheses that can be formed regarding the history and interrelationships of cognitive structures in the culture area under study.

General Research Hypotheses

Two main hypotheses were formulated in regard to the application of numerical taxonomy to the history of Middle Eastern ethnopharmacology. They are tentatively advanced as a guide for the research methods to be developed in this study. The following research hypotheses were generated:

1. A numerical taxonomy approach to Islamic medicine will show that there is a relationship between (a) the proximity and shared history of the localities where two or more sources (i.e., collections of prescriptions for drug-plants or descriptions of drug-plant properties) were produced and (b) the degree of similarity between the overall use-based similarity judgments of drug-plants by those sources.

2. A numerical taxonomy approach to Islamic and pre-Islamic Near Eastern medicine will show that there is a relationship between (a) the relative strength of antecedent-successor relationships of sources and (b) the degree of similarity between the overall use- based similarity judgments of drug-plants by those sources.

Significance of the Study

First-generation cognitive studies of historical questions continue to become more scarce (to my knowledge, none have been published since 1981). Researchers undertaking second-generation studies have yet to ask historical questions of their data. It seems to be an opportune time to draw on models from other disciplines that use variation to explore issues of diachrony and

6 evolution as inspiration for the development of a fresh approach to cognitive classification systems. If a phenetic approach proves to address the problem of diachrony in second-generation cognitive studies in a way that allows for the classification of “cognitive dialects” based on their overall gross similarities, then this will open the door for a wide range of future projects. The development of methods for cognitive dialectometry will dramatically broaden the scope of second-generation cognitive science. Such an approach would constitute the basis for a powerful new discovery procedure and serve as a useful tool for case history-specific hypothesis generation. It would also provide a foundation for the future elaboration of second- generation reconstructive techniques analogous to those used in the comparative method of linguistics. Unlike first-generation techniques, a phenetic approach would be able to make accommodation for the fuzzy boundaries of categories and their prototype-centered nature. A valid quantitative approach to the classification of cognitive dialects would provide ancillary means for supporting and illustrating previously posited historical “hunches” regarding inter- and intracultural intellectual influences and relationships. The scholarly study of Islamic medicine has been characterized by a rich tradition of translation, interpretation, and comparison. However, where primary historical records lack explicit statements of descent or borrowing, claims for inter- relationships have often been put forward in the form of unsystematically examined, anecdotally supported conjectures. The present study may quantitatively support some of these claims for historical influence, help lay others to rest, and, more generally, help to clarify the relative strengths of antecedent-successor relationships. In the event that the study’s results do not prove to reflect the basics of the known historical and geographic facts of Islamic medical practice (a negative result, i.e., a Type I error), then, at the very least, the project will have served to narrow the possible approaches for future solutions to the problem of history in second-generation cognitive studies by excluding some or all features of this particular approach. In either case, it will provide a more elaborate portrayal of the diversity of the system(s) of Islamic medicine and its/their antecedents in the region than has been heretofore presented.

7 Delimitations

Attempting to understand cognitive systems in historical terms is a broad problem. It was deemed necessary to set some boundaries for the scope of this study in order to achieve a usable result (in a realistic time frame) with the resources available. The first delimitation made was the narrowing of the problem to a second-generation cognitive study. First-generation studies attempting to reconstruct proto-cognitive systems based on componential feature analyses have previously been successfully undertaken (insofar as semantic feature- based analyses and taxonomic hierarchies may be said to represent actual cognitive structures). With first-generation approaches already established, an additional first-generation-style study seemed less interesting than moving the discipline’s historical perspective forward into a second- generation approach. While it would have been theoretically possible to test the hypotheses for this project with data from any domain of knowledge in any well-defined culture area (or even beyond and across culture areas), it did not seem expedient to attempt the initial trial with any randomly selected domain or culture area. Instead, Middle Eastern medicine was selected as the focus for the case study on the grounds that it has a better documented history of greater time depth than is found elsewhere or in other domains of “traditional” or “folk” science. Taking into account its ancient Near Eastern antecedents in Egypt and Mesopotamia, the records of Middle Eastern medicine go back over three millennia. Another necessary delimitation was the restriction of the theory and methodology to evolutionary approaches that do not necessarily distinguish between formal, phenotypic similarities in outward appearance and behavior and those that result from purely phylogenetic relationships of descent. Cladistic approaches and other reconstructive techniques that base their classifications on subjectively selected characters “known” to represent genetic descent were excluded as too problematic at this early stage of inquiry. From a second-generation perspective, there is no way (yet) to distinguish between similar cognitive characters or features deriving from descent with modification, evolutionary convergence, or diffusion through areal influence. However, since phenetic classificatory results might reveal

8 something about phylogenetic relationship, the identification of similar tendencies in classification by different sources or subjects is a first step in the direction of discovering how researchers might be able to identify genetically inherited characters in the future. Finally, in terms of the sources and data selected for analysis, the modeling of distributional similarity judgments was limited to the utility of drug-plants as presented in previously published sources. This delimitation can be understood as consisting of three related restrictions: first, the restriction to plants only; second, to utility only, and third, to previously-published sources only. Islamic and pre-Islamic medical systems in the Near East exhibit a broad range of pharmaceuticals, including animal and mineral products as well as plants and plant-derived materials. However, the names of animals, minerals, and the medicinal drugs derived from them have proven particularly difficult to identify and translate, especially in earlier sources. They were therefore excluded. In addition, some plant-derived products (such as olive oil, crystallized sugar, and cotton wadding) were also excluded on the grounds that the degree of processing to which they are subjected before they ever reach the herbalists is so extreme that they may no longer be conceived of purely as “plants.” Only those plants and plant products that remain in their original morphological shape until the herbalists (or other traditional medical practitioners) themselves modify them were included. While it is feasible to model similarity judgments based on the gestalts generated by attention to an extremely wide range of possible attributes (the author previously conducted such a study on ancient Egyptian “trees,” Pittle 2002), this exhaustive approach seemed inexpedient for the problem at hand. The present study was limited to utility only (or the named properties that ultimately determine utility, where the utility derived from them is not presented by a source). It was not feasible to collect all known attributes for all of the drug-plants used in all of the locales under study. Published sources were selected as the only appropriate basis for this initial trial. While a fine-grained, region-wide, fieldwork-based collection of relevant data might ultimately yield more nuanced results, the monetary expense and cost in personnel and time resources were prohibitive for the initial application of this previously untested model. However, once the approach has been successfully demonstrated, there would be sufficient reason to seek the funding and resources to undertake a similar study with a large

9 number of “live” subjects, over a broad area, in an effort to further develop and validate methods for elaborating the emergent field of “cognitive dialectology.”

Preliminary Definitions and Operational Terms

The following definitions are presented to provide an early clarification of some of the key variables, constructs, theories, and models used in this study. They may also be used as a reference guide to these concepts.

Variables and Constructs

Drug-plant. Drug-plant (or plant drug) is operationally defined as a botanical entity used with the intent to cure or improve the health of an ill person. More specifically, it is defined as entities that are most familiar or useful to healers (herbalists, etc.) in the least modified form possible. For the purposes of this study, drug-plants given alternative names in different locales are measured as the same drug-plant if they have been identified as representative of the same botanical genera or have been historically recognized as synonymous terms for the same entity. Drug-plant Property. Drug-plant properties are features of a drug-plant describing its healing qualities in terms of a limited set of dimensions belonging to a traditional system that dictate how the plant is to be prescribed (such as Galenic humoral qualities of “hot,” “cold,” “moist,” and “dry,” or Ayurvedic qualities and actions). Prescription. Prescription is operationally defined as a list of two or more drug-plants to be compounded or used together as a remedy for the same illness or negative condition (such as spirit possession). Proximity. For the purposes of the present research, proximity entails a consideration of both the geographic nearness of two or more locales to one another and the geological and constructed structures that impede or facilitate interaction between those locales. While locales A and B may be geographically proximate in the normal sense of their relative adjacency “as the crow flies,” mountains or large bodies of water may significantly impede the interaction of their populations. Additionally, two locales may not be proximate in the traditional sense, but may have heavily-trafficked trade or

10 pilgrimage routes connecting them. For the purposes of this study, localities so linked will be treated as having greater proximity. Shared History. In this study, shared history refers to historically documented periods of interaction and influence or to specific indicators of direct or indirect borrowing of cultural ideas or features (especially explicit primary source statements). Antecedent-Successor Relationship. Antecedent-successor relationships are inferred from the presence of intervening links between the systems of medicine practiced by two or more temporally removed sources. More specifically, such relationships are characterized by the influence of one extinct cognitive system upon a later system, either directly or by means of one or more intermediate systems to which both are related. For example, system A and system B are contemporaneous and mutually influence one another. System A becomes extinct, while system B persists. System B later interacts with and influences System C. System C is therefore the successor in antecedent-successor relationships with both systems A and B. Similarity. According to D’Andrade, “Terms that share many features will generally be judged to be more similar than terms which share few features” (1995:64). One important kind of similarity judgment used in this study, “distributional similarity,” represents “the degree to which sets of items are found to co-occur in the same environments” (D’Andrade et al. 1972:10). This study will examine the distributional similarity of a set of items, namely, “drug-plants.” Following D’Andrade et al. (1972), the distributional similarity of drug-plants is conceived of as being determined by their environments, i.e. the “prescriptions” that they occur in or the “properties” that they co-occur with. The global set of distributional similarity judgments for a source will ultimately be transformed into “features” (items representing tendencies to distributionally group a particular set of plants into the same cluster) using a statistical computer program’s hierarchical clustering tree-generation capabilities. The overall similarity of the sources will then be measured in terms of the number of these derived features that they share in common. The “degree of similarity” between sources will be quantified by means of one or more of several measures, including the arithmetic mean, the geometric mean, and the usability index (to be described in chapter 2).

11 Theories and Models

Cognitive System. First- and second-generation cognitive studies operationalize the idea of “cognitive system” differently. From a first- generation (G1) perspective, a cognitive system is a structure made up of units that can only be defined in relationships of contrastive or complementary distribution with one another (typically modeled as a hierarchically arranged taxonomic structure or paradigm). From a second- generation (G2) perspective, a cognitive system consists of a set of related gradient categories centering on prototypic objects and including the extended set of more or less prototypical objects that share the highest number of attributes, on average, with them. Memetics. Memetics is a newly emerging scientific approach founded on an “analogy between cultural and biological contagion” (Lynch 1996:2-3). “Memes” are “units of cultural transmission” (Dawkins 1999:192) or “idea units” that are carried by “vehicles” (humans, their behavior, and the artifacts they produce) and “evolve via differential replication” (Croft 2000:12). Groups of memes that are replicated together (such as a cognitive system, like Islamic medicine) are called “coadapted meme complexes,” or “memeplexes,” for short (Blackmore 1999:19). Like genes, memes are formally expressed in phenotypes that are sometimes referred to as “memotypes” (Aunger 2002:160). Phenetics. Phenetics is “the mathematical assessment of similarity” (Heywood et al. 1964:iii) through “an extension of genetic approaches and principles to . . . forms that are difficult or impossible to study [otherwise]” (Yablokov 1986:45). Phenetic classification (also called “numerical taxonomy”) is based on similarity of form in multiple traits. Since the numerical categorizations of pheneticists are based only on similarities in phenotype, they yield “taxa in which members of [a higher level] taxon are in some sense more similar to one another than any of them is to members of any other taxon” (O’Brien and Lyman 2003:33). Hierarchical Clustering. Hierarchical methods involve the sequential joining of objects, by similarity, into clusters, ultimately forming a tree- like structure where families of clusters “higher up” on the “tree” incorporate smaller families lower down (Wilkinson et al. 1996:607).

12 Summary

This chapter served to introduce the study, delineate the problem under consideration, provide an exposition of the assumptions underlying how it will be approached, and present the general research hypotheses to be tested. In addition, it included an argument for the significance of the study, a description of its delimitations, and a collection of brief definitions of key concepts and operational terms relevant to the research at hand. The next two chapters, the background of the case study and the overview of relevant theoretical and methodological literature, involve an overview of selected literature related to the problem of applying a numerical taxonomy-based second-generation approach to the elucidation of the cognitive history of Islamic medicine. More specifically, they entail a review of literature pertinent to the general background of the study, the theory necessitating and underpinning it, and the instruments previously used to study related phenomena in other domains and disciplines that have been modified to undertake the present research.

13 CHAPTER 1

BACKGROUND: ISLAMIC MEDICINE IN CONTEXT

This project is primarily concerned with drug-plant prescription in present-day Islamic societies, but it is essential to recognize at the outset that patterns of pharmaceutical use do not exist in a vacuum. Ideas regarding which medicinal plants are “good for” a given illness are thoroughly enmeshed in a complex ethnomedical framework of praxis and precedent, with deep roots in history. This chapter will provide the broad context of ethnographic, historical, and geographic setting necessary for evaluating the applicability of numerical taxonomy to patterns of drug-plant prescription in the medical system of the Islamic Middle East. It includes a description of the system as a whole, a discussion of the sources from which it is commonly recognized to derive, and a brief overview of the various historical factors and interactions through which it has been shaped.

The “System” of Islamic Medicine

Several medical anthropologists working in the Middle East have observed that the local expressions of health-related beliefs and praxis they have observed “in the field” incorporate elements synthesized from multiple sources (Good 1977, Greenwood 1981, Inhorn 1994, Millar and Lane 1988, Myntti 1985). The main components they note include a humoral system inherited from the Greeks, hadith-based “prophetic medicine” derived from the folk cures practiced in the Arabian Peninsula during Muhammad’s time, survivals of

14 indigenous pre-Islamic theory and practice (e.g., pharaonic or Coptic Egyptian, Assyrian, or Syriac), and independently innovated adaptations to the immediate local ecology. Taken together, these various inputs have amalgamated into a bricolage (see Levi Strauss 1966:16-36) of useful ad hoc tools for confronting illnesses “cook-book style” (Good 1977:30). The historical factors relevant to the development of this “tool kit” are of great import for the present research and will be discussed at greater length later in this chapter. The next pages focus on Islamic medicine in the “ethnographic present.” Byron Good (1977) has described the state of Islamic medicine as he encountered it in the Iranian town of Maragheh. His observations mirror those of other anthropologists working elsewhere in the Middle East. The components of Islamic medical systems are typically drawn from diverse sources, some of which were initially worked out in locales greatly removed from their present environments. Islamic illness and curing constructs integrate and preserve the values of local communities, while integrating them into frameworks stemming from diverse historical periods and high theoretical traditions (Good 1977:33). Although bricolage-type configurations of illness (and concomitantly, of curing) aligning multiple historical streams of theory and practice are community specific, ecology and history have presented peoples throughout the Middle East with many of the same “finite” and “heterogeneous” instruments and means to work from (see Levi-Strauss 1966:17). Consequently, a high degree of commonality is apparent, despite variations in “local color” (Shiloh 1953 and 1968). No description of Islamic medicine can justly claim to portray the system in invariant, overarching terms, but it is possible to paint a generalized portrait of a gestalt Middle Eastern medicine abstracted from the convergence of features widely found throughout the culture area. The system of Islamic medicine described in this section is therefore both “nowhere” and “everywhere” present in the Islamic world. It is a description shaped from the “tools and materials” at hand (see Levi-Strauss 1966:18). Most of the medical anthropology published on the Middle East has to do with the symbolic implications of the philosophy and application of Islamic medicine. These works generally focus on how gender, social class, and other power differentials are manifestly expressed through illness, prevention, and treatment. “Us/them” distinctions are pervasive and commonly dominate the

15 attribution of meaning to symbols of health, illness, and curing. This presents a unique difficulty for the diachronic study of cognition as it relates to Islamic medicine. While “the same” symbols (e.g., the heart, tattoos, spirit beings, etc.) are fairly consistent across the Middle East, the meanings ascribed to them differ from social group to social group and village to village. Some of the most important symbols are artifacts of material culture (such as eye- shaped beads or saints’ shrines), endowed with a wide range of potential meanings, and “natural symbols” intimately associated with near universals of human “being,” such as blood, the heart, or feces (see Douglas 1973). The pervasiveness of these signifiers throughout the region coupled with their extreme heterogeneity of signification make an attempt to examine the finer vicissitudes of historical relationship extremely challenging, if not impossible, using such natural and artifactual symbols. In the 1980s, medical anthropologists specializing in the Middle East became increasingly aware of deficiencies in the existing ethnomedical studies of the region. One major critique was in regard to a pervasive lack of “long-term historical perspective” and inattention to a broader framework in a region where written records of medical systems go back literally thousands of years (Millar and Lane 1988:651 and Morsy 1981). Most researchers have since paid closer attention to the broader framework and the importance of the grand sweep of history. But Middle East anthropologists still seem to be caught in a “Catch 22” in which they vacillate between (a) descriptions of area-wide constructs (such as the humoral system) originating in the common inheritance of the “great traditions” of the past or (b) of locally bound ethnomedical symbolisms. All this is done without ever giving more than a nod to mid-level relationships between contemporaneous neighbors of varying proximity and levels of interaction. Their quandary results partly from the highly symbolic and variable nature of the elements of the system they have chosen to focus on. With the constant metamorphosis of an outward signifier’s signified meaning constantly varying from one sub-population to another, there is little to quantify relationships apart from the presence or absence of broadly defined cultural features (such as eye-shaped beads or possession by spirits of foreign ethnoreligious or geographic origin). The description of the philosophy and practice of Islamic medicine that follows is a synthesis derived from works by several anthropologists and

16 folklorists working in the Middle East. Soheir Morsy has written extensively about social roles, gender, and power as it relates to health and illness in Egypt but also includes many informative details of local practice (1980a, 1980b, 1993). Morsy’s writings include more details about curing and prevention than most other students of Middle Eastern ethnomedicine. Others address these issues, but only a few, such as Pamela Constanides (1977), Hani Fakhouri (1968), and L. Lewis Wall (1988), do so adequately. Wall produced a useful ethnography of conceptions of illness and wellbeing among Muslims in West Africa, which includes an important overview of Islamic anatomical ideas (1988). Evelyn Early (1982, 1988, 1993), Bernard Greenwood (1981), Beth Kangas (1994) and Cynthia Myntti (1985) describe the physiological dynamics of Islamic medicine. Frode Jacobsen (1998) and Mary-Jo Good (1980) focus their observations on the role of emotion in physiology and wellbeing. Several authors have addressed issues relating to illness causality, including Ailon Shiloh (1953), Bernard Greenwood (1981), and Hania Sholkamy (1998). All of these researchers neglect to describe intracultural variations or inter-societal interrelationships in any meaningful level of detail. Their portraits tend to revolve around the uniquely local or, more often, the pan-Arabic or pan-Islamic.

Philosophical Bases of the System

Islamic medical practice is founded on a specific, culturally contextualized set of philosophical concepts relating to anatomy and physiology, etiology, and the diagnosis of causation. Islamic anatomy differentiates the external parts of the body better than the internal parts, although both are thought of as intimately interconnected. Every organ, internal or external, has its divinely appointed purpose as part of the essential structure upon which the body’s survival depends. The heart is conceived of as both the seat of emotion and the prime mover in regard to physiology. Blood derived from the heart (and the pulse, which is considered to be its breathing) gives strength to the other organs through the life force the blood carries. The brain makes the other parts work. Bones exist solely to support the flesh, while muscles allow work. Blood vessels, ligaments, and tendons are all seen as important but otherwise inscrutable string-like things, so similar that they are frequently all lumped together under the same lexical label. The liver, kidney, gall

17 bladder, womb, spleen, and lungs are all typically differentiated, but nearly everything else internal is systematically linked, collectively conceived of as the belly, or stomach, all associated with the function of digestion as the organ which contains and processes the body’s fuel. Internal body parts are thought of as a complex interlocking system expressing a kind of fluidity among components. The body is a dynamic organism with overlapping parts. These components are all integrally related in ways through which foreign objects, e.g., an IUD (Early 1988:73 and Kangas 1994:2), heat or cold (Greenwood 1981:224), or psychic dispositions (Early 1988:75) may freely travel. Every individual has their own natural physiological temperament, which is affected by climate, geographic locale, foods, and emotional and environmental influences. Temperament varies by sex, place in the life cycle, and current life situation. The body is thought of as trying to maintain a systemic internal balance or equilibrium, based on good digestion, which is the product of a balanced diet and right relationship with ecological, social, and supernatural elements in the environment. Good nutrition (contributing toward good digestion and therefore healthy equilibrium) is described in terms of quantity of food consumed and appropriateness of the food’s qualities for the physiology and circumstances of the individual concerned, by age, weight, gender, and current health state. Health is defined as physical “well-being, intactness, peace, safety, and security” and social “faultlessness, rightness, correctness, genuineness, truth, and credibility” (Antes 1989:177). Health is not so much the absence of illness as the presence of wholeness. Illnesses and injuries (although under supernatural regulation and ultimately attributed to God) are thought to be caused by outside forces acting with negative purpose. Etiology is not thought of in terms of pathology or symptomology so much as of causation. The cause of any given illness is universally some kind of imbalance, whether it be humoral (heat or cold), social (excessive jealousy or dissatisfaction), or spiritual (contamination or possession). More often than not, illnesses are symbolic expressions of strained social relationships and may be seen as rooted in social relations set off balance by deviation from role expectations through interpersonal conflict or by emotional distress. Illnesses may be classified as either stemming from natural, physical causes (such as worms, insect or animal bites, bad food or nutrition,

18 unhygienic water, inclement weather, overindulgence in food or sex, excessive exposure to the sun, humidity, darkness, strong winds, heat, or cold) or from supernatural sources. Supernatural causes may include the envious “evil eye” of another human being or the displeasure of a spirit-being (i.e., a Jinn, or ‘genie’) or the ire of a supernatural relative (Morsy 1993:102). Whatever their etiology or the proximate cause of their symptoms, as Shiloh (1953:278) has noted,

All illnesses or injuries are subjective affairs arising out of personal actions conducted or not conducted, or caused by someone or something possessed with power. Illnesses or injuries do not just occur--they befall a certain victim, at a given time, and in a definite manner because of specific causal actions.

Short-term, self-limiting afflictions are not thought of as sicknesses; only a protracted state of being unwell is considered to indicate true illness. The diagnosis of illnesses is generally a post hoc, retrospective matter, and has more to do with which treatment ultimately proved successful in alleviating the problem than with what the symptoms of the sick person were.

Prevention and Treatment in the System

Sick persons are referred to particular practitioners based on the best guess as to the cause of their malady. Bone setters and health barbers are consulted for obvious physical injuries to the external body, and herbalists provide botanical and other materials through which home remedies for illnesses may be attempted. If the herbalists’ ministrations are successful, then the illness is usually deemed to be one resulting from natural causes. If home remedies are ineffective, this indicates the illness is likely caused by a non-natural agency, although herbal cures may also be effective in some spiritual matters. If herbs fail and Western biomedicine is unavailable for financial or other reasons, a holy person (a fikih, a priest, or a rabbi) may attempt an exorcism. Should exorcism prove ineffective, it becomes clear that the possessing being is a nonremovable zar spirit (a kind of jinn), who has taken up permanent residence in the individual and can only be negotiated with by means of the life-long repetition of ceremonies conducted to placate it. Generally, the primary emphasis for illnesses is on prevention, while for injuries, it is on treatment. Illness prevention may involve

19 ornamentation or other treatments on the outer body as a kind of external medicine. Such practices-as leaving a child a bit dirty to avert envy, tattooing to relieve pressure that may result in clouding of vision, kohl use to prevent eye problems, ear piercing to allow fevers to escape, the wearing of blue beads to avert the evil eye, and cutting a notch in children’s ears to disfigure them and make them less subject to envious stares are all common preventative measures. Also, invoking the name of the deity or remaining in physical proximity or contact with a book of Holy Scripture or written charm is considered a wise prophylactic against supernatural insult to the body. Remaining in the presence of strong persons or the tombs of saintly individuals (thereby partaking of their power through contagion) yields a similar effect.

Underreporting of an Important Aspect of the System: Herbs

The most promising “symbols” for quantifying mid-level relationships within the culture area are also the most underreported. Greenwood has observed that the lines between classification of illness by perceived cause and by treatment may not always be as clear-cut as normally expressed in the literature, especially in regard to diet and herbal treatments. He reports two cases where one “conceptual structure” of the diet and herbs considered appropriate for a particular set of symptoms served both the natural and supernatural interpretations of causation (1981:228). Since it is possible for the same drug-plants to be applied to identical symptoms thought to originate from more than one causal circumstance, botanical medicines represent a set of symbols less dramatically in flux than the other treatments and preventions outlined above. Many researchers have noted the existence of herbal remedies as part of the larger complex of Islamic medicine (see Auerbach 1982:1500, Bakker 1992:822, Constantinides 1985:688, Early 1982:1493, Greenwood 1981:220, Lane and Millar 1987:162-163, Millar and Lane 1988:165, and Myntti 1985:165). However, few (with the exception of Greenwood 1981, Lane and Millar 1987:162- 163, and Millar and Lane 1988:165) provide any actual data whatsoever on the use of drug-plants in curing illnesses (and even this is scant). The only thorough reporting of herbal remedies in Islamic societies was undertaken by a team of Japanese pharmacologists affiliated with the Institute for the Study of Languages and Cultures of Africa, who worked in association with native consultants (see Ahmed et al. 1979, Başer et al. 1986, Bellakhdar et 20 al. 1982, and Ushmanghani et al. 1986). Despite the boon their data provides, the lack of clear ethnographic context for their work (the complete absence of even the most basic descriptions of locales where herbalists operate, simple demographics of the communities, etc.) would frustrate any anthropologist.

The Sources of Middle Eastern Medicine

As already noted above, Islamic medicine, in whatever local expression it takes, is an amalgamated bricolage of features drawn from diverse times and places. It is a pluralistic system of fragmented or separate but integrated elements brought together through historical circumstances of commerce, proselytization, and empire (Greenwood 1981:219 and Millar and Lane 1988:656). This subsection of the chapter presents (in broad outline) some of the sources of cultural categories upon which Islamic medicine is based. More specifically, it consists largely of a survey of a deeper level of historical substrata than has typically been examined in prior studies of Islamic medicine (following the recommendation of Millar and Lane 1988). The discussion will cover a wide array of possible theoretical and literary influences, as well as proposed etymological inter-associations between diverse periods and cultures. The review will be given in a semi-chronological order and includes some details of the ways in which antecedent systems may have passed on some of their elements to their successors. It is not primarily intended as a chronology of the development of Islamic medicine. Rather, it is offered as a collection of portraits of the major inputs that have most likely contributed to the development of the Islamic medical system described in the previous section. Once these primary influences have been described in “broad strokes,” it will be appropriate to move on to a brief but more intentionally historical sketch of the external and internal interactions that brought these elements into contact of varying degrees of intensity.

Ancient Near Eastern Medicine

The earliest known recorded medical traditions in the world are those of ancient Egypt and Mesopotamia (present-day Iraq). Mesopotamia, a territory that saw the successive rise and fall of the city-states of Sumer, Akkad, Babylonia, and Assyria (Bates and Rassam 2001:23), boasts the oldest 21 surviving texts, dating from around 3,000 B.C.E. (Majno 1975:36). However, its medical records (primarily magico-medical treatises, glossaries of drug names, and exceptionally difficult to translate prescriptions) are highly fragmentary in nature, are distributed over a roughly 3,000-year span (with huge time gaps in between sources), and the majority of the cuneiform tablets come from the final days of Assyria’s influence on the Near Eastern scene, around 600 B.C.E. (Majno 1975:36). The oldest-known Egyptian medical papyri date from around 1900 B.C.E. A scribe who produced a medical text in the 1550s B.C.E., however, claimed that his prescriptions were originally authored in the time of a pharaoh who ruled over 1,500 years earlier (see Nunn 1996:31). The Egyptian medical system is well attested in relatively lengthy and complete texts (the lengthiest set of prescriptions boasts nearly 900 recipes). These were mostly produced at times separated by only a few hundred years at a stretch (see Manniche 1989:62). It remained relatively consistent, with “no evidence of major changes in . . . format or content” from the earliest known sources, through to 525 B.C.E. (Nunn 1996:206). Since these two ancient civilizations were in long-term commerce with one another (at one point, Egypt was in vassalage to Assyria), it is likely that there was some degree of intercourse in their intellectual traditions. Much less is known about Mesopotamian medical theory than its practice. Unless otherwise noted, the following description is primarily based on the work of medical historian Guido Majno. In terms of anatomy and physiology, ancient Mesopotamians saw the heart as the site of intelligence, the liver as the place where anger rested, strength was found in the kidneys, and the brain was largely thought to be unimportant. Etiologically, they believed that illness was either the fault of the victim (who must have committed some sin to merit such calamity) or an attack from malicious outside agents (such as an evil spirit or god, cold, dust, or a bad smell). Treatment for injuries and illnesses was proffered by the áshipu ‘sorcerer’ and the asu ‘physician’ (a member of the temple clergy), both of whom collaborated together in their practices. They may have also been accompanied by the bāru ‘omen-reader’ who told the others what he saw in regard to (a) all of the persons, creatures, and objects met on the way to the consultation, (b) whether the patient’s head was hot, cold, moist, or dry, and (c) what these observations portended (Budge 1996:51). In the earliest cuneiform set of prescriptions (dated to circa 2158-2008 B.C.E.), which is not significantly different in its pharmacy from that presented in

22 the very latest texts, 80 percent of the remedies are for external use (Majno 1975:46). The use of healing amulets was not unknown and a number of drug- plants were used for particular illnesses simply because their names punned with those of the maladies suffered. According to at least one author, the “native” Syriac medicine of C.E. 300-800 presently known to us shows that “many of . . . [the prescriptions] were taken from the native medical works of the [ancient] Babylonians and Assyrians” (Budge 1996:74). Levey (1967:29) has argued that, after the decline of the Assyrian Empire, Akkadian medical knowledge was likely passed on orally into Aramaic, Hebrew, and Syriac traditions. From there, it ultimately had enough influence on Islamic medicine that in one Arabic manuscript (produced in C.E. thirteenth century Persia) about 20 percent of the plant names can be traced back to the Babylonian dialect of Mesopotamia (Levey 1967:27). It is less clear whether Mesopotamian medicine had any direct influence on the Greeks. An emphasis on prognosis, the idea of critical days for treatment, and the adoption of over 250 specific drugs have been proposed as plausible evidences for its influence. Diagnostic attention paid to imbalances between “four bodily fluids”—yellow, black, white, or red— may represent a precursor to the four Hippocratic humors of later Greek medicine: yellow bile, black bile, phlegm, and blood (King 2001:7). The medical papyri of ancient Egypt provide a much clearer picture of the Egyptians’ ideas regarding anatomy, physiology, illness causation, and etiology than the Mesopotamian tablets afford us. J. Estes (1989) and John Nunn (1996) have written insightfully on Egyptian medicine and the description of it that follows draws largely on their observations. Anatomical hieroglyphics reveal that, as in the Islamic medicine that would later prevail in Egypt, a greater emphasis was placed on observation of the externals of the body than on its internal organs. Of the sixty-three human parts listed in the standard student grammar of Middle Egyptian (Gardiner 1957), all are external features. Although the brain’s directing function seems to have been unknown, other internal organs were recognized, including the heart (as the seat of the emotions), the stomach or belly (called “the mouth of the heart”), the liver, lungs, spleen, and bladder. Arteries, veins, nerves, tendons, and muscles were all lumped together into a category labeled metu, which was perceived as a system of conduits for circulating air, blood, water, and pathogenic or medicinal substances.

23 Illnesses were usually conceived of as caused by enemies (assumedly through witchcraft), god or spirit attacks and invasion (possession), or overindulgence in food and drink. One of the main pathogens with which the ancient Egyptians were concerned was wekhedu, a kind of putrid “pain stuff” initially manifested in the feces as the remains of undigested food. It could pass from one part of the body to another (perhaps through the channels of the metu system in a way similar to that provided by the “fluidity” model of later Islamic medical belief). Because of the focus on wekhedu and the metu-system, treatment was primarily concerned with diet and the encouragement of excretory processes, frequently assisted by the use of herbal remedies. The preventative maintenance of stability (social, political, and corporal) was the first line of defense against illness. Amulets, incantations, and various drugs, both simple and compounded, were used to heal body and soul. The legacy of Egyptian medicine may be seen in a few elements of the Greek system (Breasted 1930:16) which was born during the sixth to seventh centuries B.C.E. Greek medicine was further developed on Egyptian soil (at Alexandria) during the Ptolemaic period (305-330 B.C.E., to be discussed in greater detail, below) and after. It may also survive in folk healing beliefs and practices of the Egyptian fellahin (‘peasants’). The Egyptian concept of decay (wekhedu) may have resurfaced in the perittoma (pathogenic digestive residues) of the Alexandrine Greeks (Estes 1989:122 and von Staden 1989:122). Other features adopted from the Egyptians by the Greeks are thought to include: “several medicinal plants, the structure of drug prescriptions, the practice of prenatal and gynecological fumigations, the concept of ‘defluctions,’ and the healing value of temple sleep (‘incubation’)” (Estes 1989:122). Whether or not Egyptian medicine significantly influenced Greek medicine, the cross-pollination of the two medical cultures is attested in a C.E. third-century papyrus (P. London and Leiden). It was written in Demotic (a late phase of Egyptian that would eventually be superseded by the development of an even later Egyptian dialect, Coptic, written using the Greek alphabet, plus a few extra letters), which Nunn 1996:208 describes as “freely embellished with Greek inserts and glosses” even though its parts “are in the general format of the [old Egyptian] medical papyri” (Nunn 1996:208). According to Nunn, Coptic medicine (which developed primarily in the C.E. third through fifth centuries) “was predominantly Greek, though

24 containing many Egyptian remedies [and] partly written in a format reminiscent of classical Egyptian medicine” (1996:209). Regarding the survival of Coptic (and older Egyptian) medicine into the present, Coptologist Walter Crum (1950:185) asserts: “Of Coptic medicine, whether derived directly from Ancient Egyptian or influenced by Greek practice, a great deal remains nowadays in the homes of certain classes of people”. In support of this, he contends that a number of Coptic and Demotic Egyptian words have passed into Egyptian Arabic and are in common twentieth- century use. These include illness-related terms (such as those for ‘epidemic’, ‘diarrhea’, ‘fever’, ‘chill’, ‘runny nose’, ‘nasal voice’, ‘headache’, and ‘skin rash’) and medicinal plants, minerals and other substances (such as words for ‘clover’, ‘nitre’, ‘gum’, ‘acacia’, ‘sesame’, ‘pomegranate’ and ‘lily’) (1950:185-188). Outside of Egypt, however, the lexical influence on later Arabic terminology seems to be less profound. Levey found that only 2-3 percent of the names for plant drugs in two Arabic medical texts (C.E. ninth and thirteenth century, from Baghdad and Persia, respectively) could be traced to Egyptian origins (1966:20 and 1967:27). Crum compared a number of prescriptions and curative procedures of the fellahin to those found in the C.E. fifth- through tenth-century Coptic papyri, noticing what he considered to be convincing parallels in their specifics (1950:186-188). Others have also reported analogues between earlier pre-Islamic medical practices in Egypt and those characteristic of twentieth-century fellahin, including the use of various kinds of dung for diverse ailments, henna for headache, wormwood for a stomachache, grease and onions for wounds, khol for the eyes, and charms burnt as fumigants (Reisner 1905). Others report the continued use of “many of the same drugs . . . [used in the papyri], such as honey, oil, [specific] plant remedies, animal products, and human milk” (Estes 1989:117), and finely detailed similarities in procedures for processing scarab beetles for use as a drug.

Greek Humoral Medicine

Greek medicine entered the Middle East through two primary conduits: the city of Alexandria in Egypt, and the town of Gundishapur in Persia. Under the Ptolemies (305-330 B.C.E.), Greek physicians practicing Hippocratic medicine (based on the teachings of Hippocrates, a native of Kos in Asia Minor who flourished roughly a century earlier) served members of the immigrant Greek ruling class. With the founding of Alexandria (331 B.C.E.)

25 and its swift development into “the major cultural and scientific centre of the Hellenistic world” (Nunn 1996:206), Greek medicine came to Egypt in earnest. There, it was further refined under the influence of legendary physicians, including Herophilus and Galen. From thence, it was disseminated throughout the Greek (later, Roman, and then Byzantine)-ruled world (i.e., the Mediterranean, parts of northern Africa, and the Middle East, excluding areas east of Syria). Greek medicine came quite some time later in the Persian-dominated East (the realms of the Parthian, and later Sassanian Empires, including present- day Iran, Iraq, and much of Central Asia). In the C.E. sixth century, Khosrow Anushirvan, Sassanian king of Persia, founded the town of Gundishapur in southwest Iran. It was primarily populated by deportees from Syrian Antioch (prisoners taken in campaigns in what is today southwest Turkey), and Greek settlers (Browne 1962:20 and Ullman 1978:17). By C.E. 555, he had founded the famous school of Gundishapur (Esposito 1999:271). When Syriac- speaking Nestorian Christians of Edessa fled Byzantine persecution in the West and sought refuge in Persia, their medical literature (including Syriac translations of major Greek works) came with some of them to the school at Gundishapur. There, Nestorians taught alongside Greek scholars and continued their own efforts toward translating the Greek medical corpus into Syriac and Pahlavi Persian. The Arab-Islamic Empire rose to become the dominant power in the Middle East beginning in the late C.E. 630s and early 640s and proceeded to expand throughout the following century. Translations of Greek medical works were commissioned by the new Arab elites who now controlled the former Persian Empire and large portions of the prior holdings of the Byzantine Empire, including Syria, Egypt, and parts of the southern Mediterranean. The first translations were of “Greek and Egyptian books,” undertaken under the patronage of Khalid ibn Yazid ibn Mu’awiya (probably in the 680s or 690s) at Alexandria. The major effort to translate Yūnānī (‘Greek’) medical works into Arabic, however, was mainly done through the medium of Syriac and was undertaken at Gundishapur, the primary conduit for the transmission of Greek medicine to the Muslim world (Esposito 1999:271). Greek medicine, as it came to the Arabs, is based on the Hippocratic theory of the humors as refined and exposited by Galen (who studied at Alexandria between C.E. 147 and 158) and later writers of the Galenic school that followed him. The survey of Islamic Galenism that follows is based on

26 descriptions by E.G. Browne (1962), J.C. Bürgel (1976), James Esposito (1999), Byron Good (1994), Bernard Greenwood (1981), Marcia Inhorn (1994), Helen King (2001), and Marcia Millar and Sandra Lane (1988). In the Galenic system, each part of the body exists for a unique purpose. Food is believed to be transformed through a cooking, boiling, or fermenting process in the stomach and the liver that converts it into four humors (blood, mucus or phlegm, yellow bile, and black bile), which are ultimately conceived of as both nutrients for and constituents of the body. The humors combine with the four primary qualities of the basic elements of Aristotelian theory in a relationship where “fire corresponds to yellow bile, which is hot and dry; air corresponds to blood which is hot and wet; water corresponds to phlegm which is cold and wet, and earth corresponds to black bile which is cold and dry” (Ullman 1978:58). Due to the influence of such factors as climate, age, gender and habitual behavior, each individual’s temperament is dominated by one of the four humors. If the humors and qualities are in a state of being well-mixed or (‘balanced’) relative to an individual’s temperament, then that person can be said to be in a state of health. If one or more of the humors and concomitantly, the qualities associated with it becomes uncharacteristically dominant and unbalances the system, illness will result. Additionally, incompletely digested food remaining in the stomach, liver, or veins is thought to rot, putrefy, or stagnate, which then causes the humors to become morbid. In the Greek tradition, healing is primarily effected through surgery, dietetics (and other life-style changes), or the use of pharmaceutical drugs. As in the other systems described above, surgery is the main treatment for injuries. Illnesses, on the other hand, are generally treated through other means, which revolve around an approach based on the principal of contraria contrariis ‘opposite cures opposite.’ On the basis of contraria contrariis, to effect healing, either (a) hot substances (foods or drugs) must be used to treat cold illnesses and wet and cold substances to treat dry and hot illnesses, or (b) “In a body which deviates from the right proportion [i.e., a “well-balanced” way of life], a way of life must be followed which deviates from the right proportion in the same degree but in the opposite direction” (Ullman 1978:98). These two approaches are not far removed from one another when we consider that the Greek term diaita ‘diet’ refers broadly to peoples’ “way of life” and that

27 the line between foods and medicines is not very clearly drawn, since all such substances have active properties that can affect health. Drugs can be prescribed as simples (a single remedy) or in compound form (of multiple components). Compound remedies are prescribed where the nature of the illness, the state of the organs, or the taste, strength, or some other characteristic of the drug itself merits the combination of herbs. Most medical ethnographers of the Middle East agree that Greek (Galenic or humoral) medicine and the so-called “Prophetic medicine” (to be described in the next subsection) are the mainstays of the pluralistic system of Islamic medicine. However, they also report that twentieth-century expressions of Greek concepts in Islamic medicine are more concerned with the hot-cold dichotomy than they are with the moist-dry axis, if moisture or dryness is even recognized as medically significant at all.

Prophetic Medicine and Popular Movements

The system of medicine known as at-Tibb an-Nabi ‘the medicine of the Prophet’ or ‘Prophetic medicine’ consists primarily of collections of teachings of Muhammad in the Koran and, more importantly, health-related sayings attributed to him in the Hadiths (Bürgel 1976:34, Greenwood 1981:220, Inhorn 1994:61, and Millar and Lane 1988:652). This system serves as a quasi- religious alternative to the secular, “heathen”-born Yūnānī system. While its theoretical and practical aspects are ascribed to Muhammad, prophetic medicine is actually a codification of syncretic principles and practices (see al-Jawziyya 1998 for a typical example), legitimized through association with his authority. The content of Prophetic medicine derives from Jewish medicine, the Persian medicine of Gondeshapur, nomadic Bedouin medicine, and Galenic medicine (Inhorn 1994:61). The following discussion of Prophetic medicine is based on the work of J.C. Bürgel (1976), Dale Eickelman (2002), Peter Gran (1979), Bernard Greenwood (1981), Marcia Millar and Sandra Lane (1988), and H.M. Sa’id (1998). In the view of adherents to at-Tibb an-Nabi, illness and misfortune are all ultimately in the hands of God. However, the immediate agents through which illnesses are contracted are primarily sorcery, the evil eye, and jnun (sing. jinn) ‘genies.’ In addition, Bedouin ideas of pathology at the time of the Prophet included a focus on symmetry not very different from that of the Greeks, where illnesses originate from improper nutritional intake, stomach disorders, or indigestion. Spirit, soul, and body are seen as

28 intimately interconnected in the medicine of the Prophet, and “naturalistic” causes ultimately have spiritual origins. While sicknesses may be either “of the heart” or “of the body,” both are under divine regulation and can only be addressed through wisdom, either religious or medical that originates ultimately with “the Supreme Curer.” A balanced diet and lifestyle without excess; the use of honey, bleeding, scarification, and cauterization; the pouring of water to treat fevers, and the prescription of a few simple drugs typify at-Tibb an-Nabi in the forms where it is least mixed with Greek scientific medicine. Healing and prevention of illness in prophetic medicine may also involve the use of amulets, the writing and wearing of talismanic charms (hajib), the drinking of water into which the ink from words of the Koran has been rinsed from a washable object, and the seeking out of the intervention of holy descendants of the Prophet (shaikhs) and marabouts (‘pious ones’) who possess the trait of healing baraka (‘blessing’ or ‘blessedness’). Complementary to the literate tradition of at-Tibb an-Nabi, popular mystics developed nonliterate Sufi medicine beginning in the sixteenth century. Sufi fraternities or cults practice a holistic psychological or psychiatric approach, mostly focused on dealing with jnun through ceremonies involving ecstatic dance and self-mutilation. Adherents to Sufi orders also commonly seek out the baraka of marabouts for healing, not limiting their entreaties to living healers alone, but frequently resorting to the tombs of deceased “pious ones” to receive from the residual baraka that adheres to their mortal remains.

Problems with Identifying the “Sources” of Islamic Medicine

While there surely are historical connections between elements of the systems outlined above and the Islamic medicine described in the first sub- section of this chapter, most previously proposed attempts to show relationships are based on purely anecdotal conjectures. The assertion that Greek medicine might have been strongly influenced by Egyptian medical thinking because they both have concepts of pathogenic decay is a highly subjective conclusion. It is based on little more than the notion that a cultural form similar to one found at an earlier period in the same (or a nearby) vicinity (in this case, Alexandria) necessarily implies a direct line of cultural continuity. As anthropologists concerned with the doctrine of

29 “survivals” have long argued, such anecdotal, individual instances of the occurrence of common features do not necessarily confirm a relationship. Etymological studies, like those of Crum and Levey, are on firmer ground. However, the survival of only a handful of lexical items does not prove continuity of a medical system utilizing the objects those terms signify either. The presence of less than a half-dozen similar recipes (Crum 1950:186-188) does not significantly support the contention that “a great deal [of Coptic medicine] remains nowadays [in use among segments of Egyptian society]” (1950:185). Only the most meager elemental pieces of systems supposed to contribute to Islamic medicine have so far been put forward as piecemeal evidence of descent or borrowing. None of these assertions have yet addressed any subsystem (such as drug-plant prescription) holistically, looking at the overall patterned structure of one or more major components of a system in an objective manner. The current study will remedy this problem by treating the subsystem of drug-plant prescription from a holistic perspective, as part of an effort to investigate systematically evidence for relationships between sources from a more objective, numerical affinity-based perspective.

Relevant Historical Interactions of Middle Eastern Societies

Writings on the history of the Middle East, the influences shared by peoples within its bounds, and the interactions of Islamic societies with cultures beyond the borders of today’s Middle Eastern states fill untold numbers of volumes (see Figure 1.1 for a map of the culture area under study). The history of the internal and external interactions of Islamic societies is a vast enough topic that this subsection of the chapter only includes those geographic and historical facts that seem most relevant to the development of medicine in the Middle Eastern culture area. The organization of this discussion is more thematic and topical than it is chronological, although certain specific periods of time will be referenced where deemed appropriate.

30

Figure 1.1. The Middle East and Central Asia: Political Boundaries, 1990. Reprinted from Eickelman 2002:3.

31 The multifarious origins of several streams of medical thought and their ultimate integration into Islamic medicine can be readily related to (a) the pharmaceutical drug trade of ancient and medieval civilizations; (b) the peregrinations of medical scholars (and the texts they produced) in the ancient and Islamic periods; (c) the presence of trade and pilgrimage routes connecting major population centers throughout the region and beyond; (d) sociopolitical and linguistic factors (including movements of slaves, loyalties to legal schools and religious orders, and shifting political and linguistic boundaries); and (e) the history of shifting intellectual centers (and the ways in which they served as entrepôts in the transmission of medical traditions from one language or society to another). Each of these will be examined briefly, in turn, in the following pages.

The Pharmaceutical Trade

Trade in medicinal plants (many of which were also used as aromatics and spices) and other materia medica linked together the empires of the ancient and medieval worlds that shaped the Islamic medicine of the present. Evidence for such connections can be found in written accounts of drug and spice routes (or narratives about the sources of medicinal drugs) and in the form of archaeological remains (like the heaps of south Arabian myrrh found in Egypt dating to as early as 3000 B.C.E., Majno 1975:215). It is also revealed in borrowed nomenclature (like Arabic names for plants clearly derived from Sanskrit terms through Persian intermediaries, Levey 1967:26). As Majno has noted, “When a drug is imported, it is not only the substance but also the idea that travels” (1975:377). A consideration of some of these drugs and the paths they have followed, both geographically and linguistically, proves enlightening. Frankincense (Boswellia carterii) and myrrh (Commiphora myrrh) are two drug-plants which have been important for healing in the Middle East as far back as medical history records. Both come from a very limited geographic area: the southernmost tip of southern Arabia and the Horn of Africa adjacent to it (Majno 1975:211). From there, they were, and continue to be, traded up the Arabian Peninsula and on to points further north and west; across the Arabian Sea to India, Ceylon, and beyond (Majno 1975:211), and in earlier centuries, likely carried across the Sahara as well (see Davidson 1995:74). Two other important drug-plants, cinnamon (Cinnamomum zeylanicum) and cassia (Cinnamomum cassia), were borne along similar routes (see Figure 1.2), but in

32 the opposite direction. Cinnamon from India and Malaysia and cassia from northern Vietnam and southern China came to the Gulf of Aden (where the Red Sea meets the Indian Ocean), from whence they were transported along the same paths north and west as frankincense and myrrh (Majno 1975:219).

Figure 1.2. The home of myrrh and frankincense. Reprinted from Majno 1975:211.

Key species used as Greek (and later, Islamic) materia medica came from Egypt (cumin, Cuminum cyminum), from India (cardamom, Eletteria cardamomum; nard, Nardostachys sp., and sesame, Sesamum indicum), and from Persia (coriander, Coriandrum sativum) (Levey 1967:24, 25). Up until at least the early days of the twentieth century, clove (Eugenia carophylyta) came to the

33 Middle East from as far away as Molucca and Sumatra (Levey 1967:316). The corpus of texts attributed to Hippocrates makes direct references to Greek borrowing of Indian drugs and medical formulae (Shankardass 2002:279). Theophrastus noted that some of his aromatic plants came from there as well (Levey 1967:25). At times, empires waged battles over access to drug-plants (Majno 1975:211). Hearing about the wound-healing properties of aloe (Aloe succotrina) in Egypt, Alexander the Great sent an army to seize both the plant and its native island off the coast of Somalia (Castleman 1995:62). The names of pharmaceuticals also tell tales of trade and transmission of ideas about plants (at least in terms of what they ought to be called) that have traveled through the centuries. In the C.E. twelfth century, Moses Maimonides, court physician in Cairo, reported at least three Chinese synonyms in his glossary of drug names, as well as dozens of Berber, Persian, Sanskrit and Spanish terms (Rosner 1979). The survival of Coptic plant names in the Egyptian dialect of Arabic has already been noted in the previous section of this chapter. Regarding a C.E. ninth-century Arabic manuscript from Baghdad, Levey (1966:20) states the following:

I find that about 33 percent of the names of the materia medica mentioned . . . come originally from ancient Mesopotamia, usually through Syriac, Aramaic, Hebrew, Persian, and other intermediaries. . . . About 23 percent come from Greek sources, 18 percent from Persian, 13 percent from Indian, 5 percent from Arabic, and 3 percent from Egyptian sources.

Traveling Physicians, Traveling Texts

The wide-spread distribution of drug-plants and “ideas” associated with them is an important concept to grasp in consideration of the interactions contributing to the development of Islamic medicine. However, memes (i.e., healing ideas) associated with particular plants do not spread without vehicles to carry them. Fortunately, there is a long history of meme- bearing, wayfaring physicians and medical texts in the Middle East. Some of the earliest evidence for interactions comes from Egypt. There, in the temple of Amun at Karnak, the artists of Thutmosis III carved pictures of plants (some of them medicinally efficacious) collected during their king’s expeditions in Asia Minor (Manniche 1989:13). Herodotus provides one of the more ancient written (as opposed to pictorial) accounts of the interaction of medical traditions. According to his report, when the native Egyptian physicians were unsuccessful at treating Darius I, the

34 Persian ruler of conquered Egypt, he summoned Democedes of Croton, a captured Greek from southern Italy, whose ministrations ultimately proved successful where others had failed (Nunn 1996:207 and von Staden 1989:31). Some of the most renowned and influential “Greek” physicians hailed from quarters now considered part of the Middle East. Most of these men traveled extensively to gain wisdom from the diversity of experiences afforded by travel. Hippocrates, the “Father of Greek Medicine,” came from Cos in Asia Minor (King 2001:9). One of the best-known Alexandrian physicians, Herophilus, was born in Chalcedon in the same region (King 2001:27). Galen came to Alexandria from Pergamum after working in Smyrna and Corinth, and he also practiced intermittently in the Roman capital as well (King 2001:39, 40). Dioscorides, the most famous pharmacologist of antiquity, traveled widely (Gunther 1959:2) and provided first-hand accounts of the medicinal properties of plants he observed in what were, for his contemporaries, far flung and exotic locales. The “great names” of Greek and Roman medicine practiced mostly within the confines of the provinces of the empires to which they belonged (however far their borders had been extended). As noted above, “Greek” physicians (many of them from Asia Minor and what is Syria today) were taken as slaves to Persian-occupied lands to the east (as in the case of Democedes). The C.E. fifth century Sassanian king who founded the academy of Gundishapur in Persia relied on medical knowledge from further east as well: he sent an envoy to bring Hindu physicians and medical treatises back to his school from India (Browne 1962:21). Roman doctors served in military outposts on the frontiers of the Empire (as far north as Scotland), where they had opportunity to exchange plants and knowledge with “native” practitioners of the medical traditions they encountered (Majno 1975:383). After the rise of Islam, the level of geographic mobility of physicians increased in proportion with the expansion of Islamic dominion, and manuscripts of early and late dates circulated freely. Initially, the professors of the school at Gundishapur were largely unaffected by the Arab conquest of the Sassanian Empire (Browne 1962:22). However, in C.E. 765, al- Mansur (the second Abbasid Caliph) sent for the chief physician of Gundishapur (Gergis Bakhtishu) to come to his capital at Baghdad and attempt to heal him of what seemed to his own physicians an incurable illness (Browne 1962:23 and Esposito 1999:271). After Bakhtishu’s successful ministrations, his descendants served the Caliphs for at least the next 250 years (Browne

35 1962:23). They established the first hospital in Baghdad, modeled on the one at Gundishapur (Esposito 1999:271). The C.E. twelfth century rabbi, philosopher, and physician, Moses Maimonides, was born in Cordova, Spain, spent several years in Fez, Morocco, briefly lived in Acre, Palestine, and spent the remainder of his life serving Saladin’s court in Egypt (bar-Sela et al. 1964:4). Ibn al-Baytar (d. 1248), endowed with the title “Chief of Botanists” by Sultan al-Kamil Muhammad of Egypt, collected and studied medicinal plants in a number of lands, including, according to Esposito: “North Africa, Greece, Anatolia, Iran, Iraq, Syria, Arabia, and also Egypt” (1999:212). Medical books were transported and transmitted in a range at least as broad as the domains traversed by physicians and other medical literates. After the works of Galen, Dioscorides, and other Greek medical writers were translated into Arabic, their words and ideas circulated freely throughout the Islamic world (Ullman 1978:10-13). Intellectuals made copies of translations of ancient works and medieval Islamic syntheses of them, which they passed from one urban center to another. To this day, the Galenically- inspired Qanun (Canon of Medicine), composed by Avicenna (a resident of southern Arabia in the tenth century), remains influential in locales as far removed from one another as Tunisia and India. An old Chinese Muslim book of medicine (titled Hui Hui Yao Fang), written in Arabic, seems to be “identical in parts to the Canon of Avicenna” (Jing-Feng 1996:197). While the perambulations of a few particular scholars and the transmission of their works make interesting food for thought, a review of their travels cannot be the sole basis for following directions of influence in Islamic medical thought and practice (such an approach would be as anecdotal, and ultimately, as futile, as some of the efforts of ethnographers of Islamic medicine which were critiqued earlier in this chapter). The main virtue of the examples given above is that they draw our attention to the high degree of transmissability medical ideas have enjoyed throughout the recorded history of this region.

Trade and Pilgrimage Routes

As already noted, trade in medicinal drugs facilitated the transmission of medical knowledge throughout the Middle East, and, as we have just seen, the mobility of physicians and their medical texts also played their part. Other mass movements of goods and people surely also had their impact. While

36 physicians might reside in a handful of cities over the course of a lifetime, the process of traveling between cities and towns is perhaps as important as being established in a stable locale (in fact, a whole form of literature, the rihla, is dedicated to travel, see Dunn 1989). Over the course of religious pilgrimage or while traveling in trade caravans, physicians and other medical scholars and practitioners have historically had the opportunity to interact and exchange ideas with people from all corners of the Arab and Islamic world, sometimes living together on the road for weeks or months at a time (see Dunn 1989 and Wolfe 1997 for examples of travel narratives of such journeys. Also see figure 1.3 for the route followed by one medieval Arab scholar on his return voyage from China). Baghdad, Cairo, and Damascus were long-time, major nexuses of trade, as well as primary gathering stations for mustering caravans for the pilgrimage to Mecca (Dunn 1989). The expansion and consolidation of the Islamic “Imperium” under the Abbasids (C.E. 750-1258) meant that Arabs controlled commerce between China and India in the East, and Spain, Morocco, and Italy in the West (Levey 1967:23). Sea and land routes to the peripheries of the Islamic world facilitated “the migration of significant numbers of . . . skilled or educated ‘pioneers’” (Dunn 1989:76). Lines of trade and pilgrimage internal to the Middle East further enabled the movement of people and ideas from one major Islamic center to another. It is difficult to narrate the shifting currents of the lanes of commerce and devotion that brought Islamic societies across the Middle East into a network with one another and with peoples on the fringes of and beyond the Islamic world. A strong general impression can be formed by surveying maps of the various routes and how they connect different regions both internal and external to Islamic lands. At least as far back as the rise of the Abbasids, external overland routes across central Asia mirrored sea routes through the Indian Ocean and Arabian Sea (Rahman 2002:24, see Figures 1.4 and 1.5). Sea and land routes had a complementary relationship: when political instability threatened land routes, sea trade increased; when pirates or other hazards restricted commerce by sea, overland traffic compensated (Rahman 2002:24). The central Asian and Indian Ocean routes connected with the Middle East’s internal paths of pilgrimage and trade in Iraq and around the Arabian Peninsula. From

37

Figure 1.3. Ibn Battuta’s Return itinerary from China to North Africa, 1346-1349. Reprinted Dunn 1989:267.

38

Figure 1.4. External Trade Routes of India. Reprinted from Rahman 2002:41.

39 Figure 1.5. Trade Routes connect the Middle East and Asia. Reprinted from Rahman, 2002:43.

40

Figure 1.6. Some Principal Lines of Trade in Africa and the Middle East. Davidson 1991:74.

41 Figure 1.7. African Pilgrimage Routes to Mecca, ca. 1300-1900. Reprinted from Wolfe, 1997:562.

42 there, they fed into the trade networks of the Levant and Egypt, and then onwards into the Maghrib or northern Africa (see Figures 1.4, 1.5, 1.6, and 1.7). It is evident from a perusal of maps of trade and pilgrimage routes that much of Asia’s traffic with the Middle East must have passed through Baghdad (either via the overland route through Persia, or up the Persian Gulf via Basra). Baghdad then connected these eastern routes to Aleppo, Alexandretta, and Damascus (in Syria), from whence they were linked with points still further west. Merchants and pilgrims from India and other parts of Asia also made their way directly to Yemen and the horn of Africa by sea during periods when the overland or Persian Gulf routes were less secure (or lucrative) options. The Levant (Jordan, Lebanon, Syria, and Palestine) and the southern extremities of the Arabian Peninsula (Yemen, Oman, etc.) were connected by overland routes converging on Mecca. These pathways were in a complementary relationship with the Red Sea trade that connected Yemen with the Sinai Peninsula and the port at Gaza. Egypt and the Levant were connected by both sea and land routes. The major cities of the Maghrib articulate to one another, to the central Middle East, and to the Horn of Africa primarily via the trans-Saharan trade networks and pilgrimage routes, shown in Figures 1.6 and 1.7.

Socio-Political Factors

Various socio-political factors also had an impact on the movements of people, products (including drug-plants and medical treatises) and ideas. Slaves and displaced minority peoples brought ideas from their home cultures into the ruling courts of Islamic societies and sometimes exerted a great amount of influence “from below.” For example, it is relatively certain that the zar possession cult, endemic throughout the Middle East, arrived with Ethiopian slaves in late eighteenth or early nineteenth century Egypt, from whence the cult diffused to the rest of the Muslim world (Fakhouri 1968:49). As already noted, at key points in the history of Islamic societies, the most influential court physicians have come from the ranks of minority peoples (predominantly Jews and Christians), a significant number of whom had arrived as a result of flight from persecution in other, sometimes far removed, territories.

43 Historically, Muslim physicians were often leading authorities in religious law (Esposito 1999:207). For this reason, it is also important to be aware of the distribution of various schools of Islamic law, which in a very real sense link the worldviews of noncontiguous locales (see Figure 1.8). The tendency of traveling scholars to spend a great deal of time studying a diversity of topics (including medicine and pharmacology) with like-minded co-religionists wherever they can be found may be a factor of influence worth more serious consideration than the matter has previously received. Similarly, the distribution of political entities (including large regions under the control of various cities or states) also limits or facilitates the commerce of ideas and the movement of people. Times of political instability could make overland trade and pilgrimage so unsafe that alternative sea routes needed to be used (or new overland paths had to be forged to circumvent the trouble spots). During times of stability, new areas could be brought into contact under the aegis of a uniting authority drawing them into the same socio-cultural “orbit.” Thus the oscillations of borders as empires rose and fell over the millennia brought new territories into contact or cut them off from one another. The spread of Islam, which united formerly disparate peoples over a huge geographic expanse, is perhaps the best example of this phenomenon (see Figures 1.9 and 1.10). The Arab success in uniting the former Persian Empire with a significant portion of the territory that formerly belonged to the Byzantine Empire created a cultural zone that linked societies together with one another as far west as what is today Libya, as far east as the Hindu Kush mountains, and as far north as the Caucasus, and a significant portion of the Arabian Peninsula by C.E. 690.

Linguistic Factors

Linguistic factors have both facilitated and (alternately) impeded intercourse and the exchange of ideas among Middle Eastern societies. While other languages have also had their periods of dominance, Arabic has, over the long term, exerted a tremendous unifying influence. The region is characterized by three major groups of languages (Bates and Rassam 2001:97-98 and Eickelman 2002:17): Semitic, Indo-European and Altaic (Turkic) families.

44 Figure 1.8. Muslim schools of law and Sufi brotherhoods: c. 1500. Reprinted from Lapidus 2002:212.

45 Figure 1.9. The Eastern Mediterranean and the Middle East, C.E. 600. Reprinted from Ochsenwald and Fisher 2004:27.

46 Figure 1.10. The Spread of Islam. Reprinted from Ochsenwald and Fisher 2004:40.

47 Semitic languages are represented by Arabic throughout the greater area of the Middle East and Northern Africa, Hebrew in Israel, and Neo-Aramaic in pockets of Syria. Indo-European languages include Persian in some areas of Iran and Iraq; Kurdish in Iran, Iraq, and Turkey; Baluchi and Luri in Afghanistan and Pakistan, and Armenian and Greek in minor enclaves). Altaic languages include Azeri in Iran, Turkish in Turkey, and Turkmen in Turkmenistan. Other, more minor languages are in common use among ethnic minorities including Berber in North Africa, Kazak and Tartar in Central Asia, and Circassian and Georgian in Turkey (Eickelman 2002:17). There is a good deal of Arabic-Persian bilingualism in Iran and Iraq, as well as some trilingualism including Kurdish (Eickelman 2002:17). Many speakers of the more minor ethnic languages throughout the culture area are also bilingual in Arabic, and those residing in Turkey generally speak Turkish as a second language (Eickelman 2002:17). See Figure 1.11 for a map of the distribution of the current principal languages in the region. In the course of the initial expansion of Islam, Arabic was introduced across the growing empire. However, it did not immediately take hold, to any considerable degree, outside of the military encampments and urban centers until some time later (Eickelman 2002:17). It has been posited that the major dialect differences in Arabic result from either the influence of the indigenous languages of areas conquered by the Arabs or from natural divergences from a shared parent stock (a process to which all languages are eventually subjected through the vicissitudes of time). It is uncertain whether the similarities between them result from the initial uniformity of the early Arabic influence (again, dialects are seen as natural developments from a common source) or convergence towards the Classical “standard” form of the language that would ultimately become so influential in religion, science, and law throughout the Islamic world (Bateson 1967:94-95, Versteegh 1984, Versteegh 1997:112). The major dialect areas of Arabic are divided into Eastern and Western varieties. Western Arabic is spoken in the Maghrib region (Northern Africa) and Eastern Arabic elsewhere. Egypt represents something of a transitional zone, where elements of both are used on opposite sides of the Nile Delta (Versteegh 1997:161). The main Eastern groups are those of the Arabian Peninsula, the Syro-Lebanese dialects, Mesopotamian dialects, and (in some classification schemes) the Egyptian dialects (Versteegh 1997:148-164). Some Peninsular Bedouin varieties are spoken well into the Syrian Desert and

48 Figure 1.11. Major Middle eastern and Central Asian Languages. Reprinted from Eickelman 2002:18.

49 Mesopotamia and are considered “Syro-Mesopotamian,” while others only continue their range into southern Jordan (Versteegh 1997:148). The Syro- Lebanese dialects developed in the cities of Damascus and Aleppo, the centers where the first varieties of the convergent Arabic koine that would eventually spread throughout the Islamic Empire were first spoken (Versteegh 1997:152). The modern Syro-Lebanese dialects are spoken in Syria, Lebanon, Jordan, and Palestine. Mesopotamian dialects are spoken in Iraq, parts of Iran, and into northeastern Syria as well (Versteegh 1997:153, 158). Numerous Persian loanwords occur in the Mesopotamian dialects closest to (and in) Iran (Versteegh 1997:158, 232). Egyptian Arabic has worked south into Sudan and Chad and is the most likely source for the varieties of Arabic found in West Africa, as far as Nigeria (Versteegh 1997:159). Arabic has not been the sole unifying language of Islamic civilization throughout the course of its history. The Seljuk expansion in the C.E. eleventh century brought Turkic languages from the Central Asian steppes into Iran, Mesopotamia, and Byzantine Anatolia (Lapidus 2002:119). The Turkic dynasties took Persian for their literary and scientific language and Arabic for their language of religion (Lapidus 2002:234). With the rise of the Ottoman dynasties (1280-1453), Turkish became the official language of the Islamic Imperium, while Arabic and Persian were maintained as the languages of the learned elite (Versteegh 1997:234, 236). In the eleventh century, Islamic dominion stretched as far as the Indus Valley, bringing with it the strongly Arabicized Persian of the Ghaznavid conquerors from Afghanistan (Versteegh 1997:236). In Pakistan, formerly a Muslim-dominated region of India and currently an independent state, a dialect of northern India (called Hindi by Hindus and Urdu by Muslims) is widely spoken. It contains a large number of loans from the Arabicized Persian of the Ghaznavid and later Mughal empires (Versteegh 1997:236). Throughout the ebb and flow of languages and dialects across the Middle Eastern landscape, expansions of one variety at the expense of others have unified previously disparate societies. Other languages, now no longer spoken, may still exert some posthumous influence on the Arabic that overlaid them and, in most cases, drove them into oblivion (or obscurity, as in the case of languages, like Coptic, that survive today only as “ethno- religious quasilects,” see Glinert 1993:249-250). The media of communication (languages and scripts) allowed the messages of medical theories and practices to pass from one people to another with greater or lesser degrees

50 of ease, depending on the dynamics of peoples’ linguistic “habit” (Dawood 1989:347, 431).

Shifting Intellectual Centers

Trade and pilgrimage routes allowed healers and medical texts to circulate with relative freedom about the Middle East and various linguistic and socio-political factors alternately facilitated or inhibited interactions. However, the main currents of medical theory and practice are characterized by a succession of traditions anchored in geographic realities of population shifts and the rising and falling prestige of centers of learning. After the decline of the great Egyptian and Mesopotamian empires, and before the Islamic expansion, Alexandria in Egypt and Nestorian Christian schools in Antioch (located in Syria) and Edessa (in northern Mesopotamia) were the dominant sources of medical knowledge in the Middle East. As noted above, their medicine was primarily Greek, with some local influences descended from more ancient Near Eastern systems. In earlier centuries, Alexandria was the greater influence of the two axes, but, with time, Edessa and Antioch became more important (Browne 1962:114). As already noted, many of the greatest minds at Alexandria had come from Asia Minor, or had taken up practice there after their studies were completed in Egypt. So, with time, Antioch and other Syriac-dominated locales influenced and (perhaps more significantly) were influenced by the lore of Alexandria (Esposito 1999:271). With the exodus of Syriac Nestorian Christians from the Byzantine Empire and their relocation to Gundishapur in Persia, the balance began to shift further east. Gundishapur inherited the knowledge of Alexandria through the Nestorians of Syria and northern Iraq (Browne 1962:114) and ultimately became the primary conduit through which Greek medicine passed on to the Arabs (Esposito 1999:271 and Majno 1975:420), sometimes “tempered with [elements of] Hindu medicine” (Majno 1975:400). This line of transmission is illustrated by countless examples, but the translation of an Alexandrian priest’s medical handbook into Arabic from Syriac, commissioned by Caliph Umar ibn Abd al-Aziz between C.E. 717 and 720, is one of the earliest and best attested. When Bakhtishu’s descendants replicated the medicine of the hospital of Gundishapur in Baghdad, they paved the way for a further intellectual shift in power. By the end of the tenth century, Gundishapur experienced a medieval “brain drain,” as the finest of its physicians were recruited to the

51 capital of the Caliphate (Majno 1975:421): Baghdad became the new locus of Middle Eastern medical thought (Browne 1962:114 and Esposito 1999:200). Its ascendancy only lasted a few hundred years, however. Over the course of the ninth and tenth centuries, five hospitals were built in Baghdad (Esposito 1999:208). Shortly thereafter, hospitals on their model began springing up in Damascus, Cairo, Mecca, Medina, and other major centers (Esposito 1999:208). By the twelfth and thirteenth centuries, another shift was underway. According to Esposito, “. . . physicians from all over the Muslim world sought medical careers in the institutions of Damascus and Cairo” (1999:209). While the hospitals of both Cairo and Damascus were influential, Damascus boasts the first school in the Islamic world founded exclusively for the teaching of medicine the Maristan al-Atiq teaching hospital, established in the thirteenth century (Esposito 1999:207 and Sonbol 1991:5). Some of the administrative heads and graduates of this school were prestigious members of society. Their influence ultimately led to a rise in the status of physicians allowed for their uncontested entry into the “social elite” of Islamic culture during this period (Esposito 1999:208). Cairo’s first great maristan, the Maristan al-Qalawun, was modeled after al-Atiq in Damascus (Sonbol 1991:5). After its inception, Damascus and Cairo quickly became legendary centers of medicine (the collection of stories, Alf Layla wa Layla, or ‘the Thousand and One Nights,’ includes the tale of a Jewish physician who studied at both maristans, see Zipes 1991:358-367) and maintained their reputation as the major medical centers of Islam into the eighteenth century and the introduction of Western (e.g., European) medicine (Sonbol 1991:5).

The Significance of Historical Interactions for the Development of Islamic Medicine

The portrait of historical interactions of members of Islamic and non- Islamic societies presented above makes it clear that numerous factors have been involved in the development of Islamic medicine. The true story is far more complex than the simplistic two- to three-source origins posited in previously published medical ethnographies of the Middle East. Local systems of Islamic medicine are not simply based on only two or three “high” traditions (typically given as Galenic medicine and Prophetic medicine, perhaps with a nod to survivals from a local pre-Islamic system or the intrusion of Western biomedicine).

52 It is clear that influences on localized expressions of Islamic medicine are highly complex, involving a multiplicity of interactions of varying time-depth, length, and intensity. On one hand, the history of localized systems may involve long periods of relative stability characterized by low-level interactions of low to moderate intensity, primarily with immediate neighbors, resulting in the constant introduction and seeping spread of ideas picked up from others in the vicinity. Carriers might be travelers, refugees, slaves, visiting medical practitioners, or traders. On the other hand, such periods may be punctuated by the shifting of political boundaries (usually by invasion or collapse) or some other extreme situation putting what were previously weakly-linked groups into more intense contact. The results are a flood of borrowings or even complete replacement of the native system. Not only do local expressions of Islamic medicine vary, but it is apparent, from a careful review of Middle Eastern history, that the sources (such as Greek medicine or Prophetic medicine) are likewise anything but monolithic. Indeed, the multiple sources of each local tradition have multiple sources. This chapter has presented the basic ethnographic, historical, and geographic background that will serve as grounds for an evaluation of the efficacy of the application of a phenetic approach to the cognitive domain of Islamic medicine. The next chapter consists of a review of the most relevant theoretical and methodological literature for approaching this particular problem, focusing on (a) categorization in cognitive anthropology and (b) approaches to taxonomy, diachrony, and the evolution of systems in various scientific pursuits. It will also discuss a discussion of the implications of these models for the present project.

53 CHAPTER 2

LITERATURE REVIEW: THEORIES AND INSTRUMENTS RELEVANT TO A COGNITIVE DIALECTOMETRY OF ISLAMIC ETHNOMEDICINE

The primary purpose of this chapter is to show why this research is important from both a theoretical and a methodological perspective. The review which follows provides the necessary background for appreciating how the methodologies to be outlined in Chapter 3 were chosen for undertaking the investigation, and how this project builds on research already carried out by others. It focuses on paradigm shifts within cognitive anthropology and changing perspectives on historical taxonomy as it relates to theories of evolving systems. The chapter concludes with a discussion of the implications of these models for a cognitive dialectometry of Islamic ethnomedicine.

Cognitive Categories: G1 and G2 Perspectives and Instruments

While first- and second-generation (G1 and G2) studies of cognition differ in both their underlying theories and in the methodologies they employ in elucidating mental aspects of cultural categories, they also share some basic presuppositions and techniques. Both G1 and G2 approaches are primarily concerned with examining culture in terms of the relationship between (a) thought (knowledge and beliefs) and (b) behavior toward and the organization of material phenomena (D’Andrade 1995:1, Foley 1997:107, and Tyler 1969:3). Consequently, both G1 and G2 theorists see culture as a logic-based “mental system” or grammar of behaviors, emotions, events, and things (Foley 54

1997:108, Goodenough 1957 and Tyler 1969:3). It should be noted that G2 studies have not replaced, nor made irrelevant or obsolete the earlier G1 studies. The approaches can be considered mutually complementary perspectives from which researchers may look at various mental codes and processes of the highly complex entity anthropologists call “culture.” In addition, G2 approaches were not without precedent: many of their basic features were already latent in a number of G1 studies, and some of their instruments were taken over in toto from earlier approaches (e.g., pile sorting and triads tests, to be described below). The G1/G2 distinction is, in a sense, merely a convenient rubric for discussing a still in-progress paradigm shift within the larger paradigm of cognitive anthropology (see D’Andrade 1995:1-15 and also Kuhn 1970). The immediately following pages will provide an overview of the basic theoretical assumptions and methodological instruments through which adherents to the two main branches of cognitive anthropology engage in the study of cultural categories “in the mind” (Shore 1996).

First Generation Cognitive Studies

The discussion of first-generation cognitive anthropology that follows is organized around themes prevalent in Roy D’Andrade’s (1995) history of cognitive anthropology and Stephen Tyler’s (1969) seminal anthology of articles from the later part of the formative period of the paradigm. Much of the methodology described herein has already been presented in concise “cook-book” style formats by Susan Weller and A.K. Romney (1988) and Gary Martin (1995). In formulating this overview of the G1 paradigm, I have drawn heavily upon these sources, as well as from early papers by Paul Kay (1966) and Charles Frake (1961, 1962 and 1964) relating to ideas about contrast and hierarchy and how to discover them. Some of the discussion on methods in first-generation cognitive anthropology is based on the writings of Conklin (1962) and Metzger and Williams (1966), whose discussions of the elicitation frame model are indispensable to an understanding of the practical working out of the paradigm. I have also drawn on J. Bright and W. Bright (1969), B. Berlin (1992), and E. Berlin and B. Berlin (1996) regarding alternative forms for representing taxonomic structures. The section dealing with denotation and “minimal defining features” of categories is largely derived from works by

55

William Sturtevant (1964) and Floyd Lounsbury (1969). Charles Osgood (1964), C. Osgood et al. (1957) and A.K. Romney and Roy D’Andrade (1964) are the main sources behind my description of the semantic differential technique. Agar’s “sort table” technique is described in his book The Professional Stranger (1995). From the very beginning, there was some discontent within the G1 paradigm. Wallace and Atkins (1960) presented one of the earliest critiques of componential analysis and Tyler recognized its limitations in his introduction to Cognitive Anthropology (1969). Others have presented alternatives to hierarchical models of taxonomy (i.e., non-hierarchical models), including Brent Berlin (1992), Claudine Friedberg (1979), Eugene Hunn (1985), E. Hunn and D. French (2000), Robert Randall (1976) and R. Randall and E. Hunn (1984), whose critiques form the basis for the discussion of the limitations of the paradigm that follows. Cognitive ethnoscience was born out of the convergence of ethnoscience (the study of what traditional peoples know about topics like biology, astronomy, etc.) and linguistic anthropology in the 1950s and 60s. With a flurry of research, this merger inspired two roughly simultaneous (and ultimately, mutually contingent) developments in anthropology: the investigation of (a) “native” (rather than ethnographer)-generated categorizations (taxonomies) of names (or “lexemes”) for things and (b) “componential analyses” of the natively salient semantic features of the lexemes so categorized. From a G1 theoretical perspective, natives of a culture are thought to share a worldview where they construct classes of objects (also referred to as “segregates” or “taxa”) that group together items that are contrasting alternatives in the same situation(s). These segregates may then include, or be included within, other segregates or taxa, which are usually named (however, cf. Berlin, Breedlove, and Raven 1968 for a discussion of the classification of unnamed, or “covert,” taxonomic categories). Insofar as the groupings are systematically arranged into relationships of inclusion on successive levels of sets of contrasting categories (i.e., a segregate or taxon is considered a “kind of” another segregate or taxon), they represent hierarchically arranged “taxonomies.” Cognitive anthropologists holding to the G1 paradigm argue that such taxonomies are fundamental to human thought. Figures 2.1, 2.2, and 2.3 are representative schematic presentations of the

56 kind of folk taxonomies G1 studies typically attempt to elicit. While other conventions in the diagrammatic representation of inclusion and contrast are occasionally utilized, these box and tree diagrams are typical exhibits. The most prevalent alternative is the use of Venn-like diagrams (cf. Berlin 1992, Berlin and Berlin 1996, and Bright and Bright 1969 for examples). Contrast sets and relations of inclusion are generally elicited either through successive pile sorting techniques or the use of native-language “frames.” Successive pile sorts may be undertaken using a “top-down” or “bottom-up” procedure. In the top-down approach, a group or list of items belonging to the same cultural domain (“a class of objects all of which share at least one feature in common which differentiates them from other semantic domains” [Tyler 1969:8]) is presented to the “native” consultant, who is then asked to divide the items into two, three, four or more piles of objects whose members are more similar to one another than they are to objects in the other piles. Each of these groups is then repeatedly subdivided into smaller groups until the consultant arrives at groups with only one or two members.

Figure 2.1. Levels of Contrast in “Skin Disease” Terminology. Reprinted from Frake 1961 (in Hymes 1964:196).

57

Figure 2.2. Levels of Branching Diagram. Reprinted from Tyler 1969:8.

Figure 2.3. Comparison of diagrammatic representation of the semantic structure of a fragment of Hanunoo classification of peppers using a tree diagram (top part of figure) and box diagram (bottom part of figure). Reprinted from Berlin 1992:38.

58

In a bottom-up sort, the consultant starts with a pool containing all of the objects to be classified and is then asked to put items into groups of things that “go together.” These groups are then successively lumped into larger, more inclusive groups, until all of the items are brought together into one large pile. Figure 2.4 is a taxonomic structure elicited from an American English speaker through a top-down sort of objects belonging to the domain “garden produce.” Each level of the taxonomy represents one of the successive sorts made by the consultant. Frame elicitation of a folk taxonomy is achieved through the use of native language questions or incomplete “fill-in-the-blank” sentences, designed to reveal relations of contrast and of inclusion in contrast sets at successive hierarchical levels. Questions typically take the form of statements like, “What are the names of the kinds of X?”, “What is X a kind of?”, or “is X a kind of Y?” Once answers to these questions are recorded, the ethnographer repeats the question at each new level of the hierarchy that

Figure 2.4. Taxonomic structure derived from a pile sorting of home garden produce. Reprinted from Martin 1995:134.

59 is generated. As Frake (1964) and Metzger and Williams (1966) demonstrated, queries may link segregates in other non-taxonomic relationships as well (usefulness or part-whole relations, etc.). However, when the goal is the diagramming of hierarchical taxonomies based on frame elicitation, only “kind of” questions are used to generate box diagrams and schematic trees of the sort shown in figures 2.1-2.4. G1 theory contends that the cultural logic underlying taxonomic classifications is based on a formal pattern of meanings that is apprehended by native speakers of a language in terms of contrastive semantic features that define each object in a folk taxonomy. Dimensions of meaning underlie each domain. Each item (segregate or taxon) possesses certain features of meaning that are values along these dimensions. Features of meaning make an item (a) a member of a particular contrast set (i.e., a higher-level segregate or a semantic domain), and (b) distinguish it from other members of the contrast set of which it is a member. The minimal defining features of an item are its “criterial attributes.” Each feature (i.e., criterial attribute) along a dimension is believed to have two or more possible contrasting values or “components,” that define a segregate by their presence or absence. In sum, the possession of a particular configuration of components is thought to provide the necessary and sufficient conditions for membership in the class of which that item is a member. “Componential analysis” (or feature analysis) is the name of the technique cognitive anthropologists developed for investigating the formal pattern of semantic meanings underlying a domain. The technique involves asking questions of a native consultant that are meant to reveal which attributes are distinctive features (also known as “denotative,” “intensional,” or “referential” meanings). Distinctive features place a segregate into a particular contrast set and distinguish it from other members of the contrast set to which it belongs. For example, Frake (1961) undertook a componential analysis of Subanun skin diseases, in which he diagrammed the referential definitions (i.e., the “distinctive features”) of types of “sores” by mapping these attributes onto the taxonomy he had previously elicited from his consultants (see figure 2.5 for his diagram). Similar semantic-based componential analyses of criterial features have been conducted in a large number of domains, including animals, plants, kin terms, pronouns, illnesses, cures, and foods, among others.

60

Figure 2.5. Critical Contrasts Differentiating the ‘Sores’. Reprinted from Frake 1961 (in Hymes 1964:204).

A limited number of G1 studies have involved the development of techniques for quantifying the degree of perceptual similarity of items in a domain, the validity or “psychological reality” of componential analyses, and the degree of agreement among a group of consultants belonging to the same culture (or subculture) in regard to how they classify objects into taxa. Romney and D’Andrade (1964) used Osgood’s semantic differential technique (Osgood et al., 1957 and Osgood 1964) to investigate whether or not the connotative (i.e., salient, but non-critical) similarities of American English kin and personal terms reflect the denotative (i.e., critically definitional) features extracted through componential analysis. The technique involves asking subjects to “judge a series of concepts...against a series of bipolar...scales defined by verbal opposites” (Osgood 1964:172- 173). In the case of the Romney and D’Andrade study, these were limited to the pairs good-bad, nice-awful, kind-mean, heavy-light, hard-soft, and fast- slow. Subjects were asked to rate each of the kin and personal terms on a

61 five point scale for each pair of connotations. After this, an intercorrelation matrix and factor analysis was computed and rotated to determine which adjectives went together. A second factor analysis was performed to find out which terms went together (i.e., were considered to be similar) based on these affective, non-denotative meanings. The mean scores for each concept were then plotted in a two-dimensional semantic space to elucidate the tendency for sets of concepts to cluster together, potentially in response to the terms’ components (see figure 2.6). Before interpreting their results, the Romney and D’Andrade argued as follows (1964:158):

To the extent that subjects are responding to the components [identified in the feature analysis], then, a factor score for an item which represents the response to a particular component will predict the factor score for another item which shares the same component.

While the graphic mapping into two dimensions did not clearly show that consultants were responding to the components of the terms, the factor scores taken by themselves (regardless of the somewhat opaque nature of the graphic output) seemed to show that the terms were still rated with a fair degree of consistency, as if consultants were rating the items’ definitional components. However, there were still some significant anomalies, and the researchers found it necessary to investigate further the cognitive structuring of the set of kin terms, using a triads test. Triads tests may be used either to determine the degree to which respondents perceive items in a domain to be similar to one another or to examine the degree to which a componential analysis fits with the “psychological reality” of the cultural consultants being interviewed. In a triads test, the subject is presented with objects (the members of a domain or segregate) in sets of three and is then asked to select which of the three objects in each set is most different from the other two. It is assumed that the decision to retain two items and reject the third is based on the criterial features of the objects, and that the two which remain must therefore share a criterial attribute that the third does not. By calculating the mean number of times a pair of terms is kept together by a group of respondents and then dividing by the number of subjects, it is possible to generate similarity ratings for the set of terms. When the terms most frequently paired together are plotted on a componential analysis diagram that is close to the psychological reality of the subjects,

62

Figure 2.6. Mean semantic differential factors. Reprinted from Romney and D’Andrade 1964:159.

the highest frequency pairings will be between terms which differ by only a minimal number of components (see figure 2.7 for an illustration of this phenomena, based on Romney and D’Andrade’s English kin terms triads test). Romney and D’Andrade were not confident that they had proven that the decisions made in the triads test were based solely on the presence or absence of criterial attributes. They were not entirely certain that the componential analysis that they had produced was psychologically real. However, citing interviews with “a handful of subjects,” they concluded that (in this particular case) referential meaning was the predominant basis upon which subjects made their decisions (1964:161). Just as the results of a triads test given to multiple respondents can be used to compute mean similarity ratings, Agar (1996:190-195) has demonstrated a technique, the sort table, for agglomerating the pile sorts of multiple subjects into a shared contrast set. Such a shared contrast set is really an averaged “master” pile sort. His technique is based on observing equivalent responses to items identified using frame elicitation. This is

63

Figure 2.7. Diagrammatic representation of Romney feature analysis of English male kin terms. Reprinted from D’Andrade 1995:53.

achieved by (1) comparing the results of the respondents’ pile sorts and locating, by inspection, the sorts of multiple respondents that share more than one item; (2) listing these sorts on a table where the nth column lists the responses of the nth subject (see Figure 2.8 for an example from Agar’s study of difficulties faced by “junkies” in the street), and (3) indicating the responses that are shared in the sorts of all of the respondents (in the case of Agar’s study, the items underlined in Figure 2.8). Residuals (those items in a sort that are not shared) are partitioned out and treated as a new pile. These “residual sets” are then compared with one another until (a) all residual sets that can be treated as shared responses have been extracted (through a continual partitioning of the original sorts between “shared” and “residual” elements), (b) all “error residuals” that cannot be used as shared responses have been identified, and (c) a final result is reached (see figure 2.9 for Agar’s final “equivalent response sets” in the “junkie” study). Agar goes on to develop two measures that can be applied to sort table results: (a) the U or usability function (which indicates the percentage of objects in the original pool that are “usable,” i.e., not error residuals, in the final table of shared sorting responses, which thereby serves as a measure of the overall agreement of the subjects regarding the agglomerated

64

Figure 2.8. Example of Sort Table. Reprinted from Agar 1996:192. Note: “1.1” represents the first sorter’s first pile, “2.5” represents the second sorter’s fifth pile, and column “3.4” represents the third sorters forth pile.

Figure 2.9. Equivalent Response Sets. Reprinted from Agar 1996:195. Academic Press: San Francisco. 65 sort as a whole), and (b) the A or agreement function (which measures the degree to which the respondents agree on each individual agglomerated taxon). Since the present study is only concerned with overall similarity between subjects or sources (and not with agreement on individual taxa), it is unnecessary to further consider the A function (which is quite a bit more complicated than the U function) in detail. Interested readers should consult Agar (1996:196-197) for further elaboration of the A function, if desired. The U function, however, is very easily computed by dividing the number of items used in the final table of equivalent responses (nu) by the total number of original items sorted (nt), a calculation which may be represented by the formula U=nu/nt. The theory developed within the G1 paradigm has limitations that become problematic for the enterprise of uncovering “true” cognitive realities. In particular, neither the results of semantic differential analysis as practiced by Osgood and used in the early Romney and D’Andrade study (1964) nor the results generated by triads tests are based solely on the componential features so critical to G1 theory. Romney and D’Andrade themselves recognized that “other conditions [including, but not limited to] . . . frequent contiguity in normal speech are also likely to increase association frequencies” (1964:161, italics added). In fact, most of the early proponents of G1 theory ultimately came to terms with the likelihood, pointed out in an early critique (Wallace and Atkins 1960), that componential analyses only allow access to the “minimal meaning” of a term, which, in reality, should instead be thought of as including both denotative and connotative dimensions. By the time Tyler edited the seminal volume that brought together some of the most important papers in the formulation of G1 theory (1969), it was commonly acknowledged that componential analyses are only concerned with what is expected and appropriate from a denotative, definitional perspective. Tyler and others realized that larger domains are, in fact, “characterized by discontinuous and partial combinations of a large number of [not necessarily solely denotative] features”, so they “should probably not be analyzed [only] in terms of [denotative] features and feature organization” (1969:15-16). In time, Romney, D’Andrade, et al. (1972) and others realized that the defining features of an object may not always be its most salient properties and that criterial attributes do not necessarily determine how something is

66 categorized or reacted to. In peoples’ real, lived experience, connotative features of a domain are often of greater interest and import than denotative features. The validity of the assumption that “native” classifications always organize contrast sets into hierarchically ranked, tree-like structures by contrast and inclusion has also been called into question. Other, non- hierarchical logical relationships for organizing folk taxa (by functional categories of utility, “kinship” among taxa, locomotion and habitat, or other forms of coordination) have also been identified. In addition, some have argued that the impressive multi-leveled hierarchies elicited on the basis of the premises underlying the G1 paradigm may potentially be as much a product of the method as they are an accurate picture of the “native’s” psychological reality. Indeed, such a critique suggests that the richness of some folk taxonomies could be little more than an illusion resulting from ethnographers’ cultural predisposition toward finding such hierarchical structures. Since it is possible that nonhierarchical modes of organization may be in operation, it is also feasible (in some cases) that it is only “the anthropologists’ queries which create the total [hierarchical] structure [of a taxonomy]” (D’Andrade 1995:101, in reference to the findings of Randall 1976:545). Once cognitive anthropologists realized that both componential analyses and the modeling of hierarchical folk taxonomic trees were more limited in their applicability than originally theorized, researchers’ interests began to shift. The shortcomings of the first-generation approach ultimately led to the development of more flexible and inclusive theory in G2 cognitive anthropology. In tandem with this theoretical metamorphosis, cognitive anthropologists began to be more deliberate in using their instruments, with a greater awareness of and sensitivity to the importance of both denotative and connotative meaning.

Second-Generation Cognitive Studies

As in the previous paragraphs, the discussion of G2 theory and method that follows adheres to the general historical outline developed in D’Andrade’s The Development of Cognitive Anthropology (1995) and draws on Weller and Romney’s overview of quantitative methods given in their book Systematic Data Collection (1988).

67

The prototype-extension theory underpinning the G2 paradigm was largely developed by Eleanor (Heider) Rosch in the 1970s, in reaction to Berlin and Kay’s 1969 Basic Color Terms. My discussion of the theory behind the paradigm is primarily an elaboration of D’Andrade’s (1995) description of it, augmented by reference to Rosch’s early works (1971, 1973 and 1978), as well as contributions by Berlin (1978) and Lakoff (1986) regarding the Gestalten or prototypical “images” so central to the paradigm, and of Kronenfeld et al. (1985) regarding the “fuzzy” nature of category boundaries. The discussion of methods for measuring perceptual similarity (and distributional similarity in particular) largely derives from a review of D’Andrade (1972), Fillenbaum and Rappaport (1971) and Stefflre (1972). Finkler (1984) and Garro (1986) are the two main authorities on measuring sharing and variation in cognition. Hierarchical clustering methods for modeling similarity have been described by D’Andrade et al. (1972) and Wilkinson et al. (1996) and multidimensional scaling techniques for representing the same by D’Andrade et al. (1972), Garro (1986), Stefflre (1972) and White (1985). The revolution in cognitive theory began in earnest with Berlin and Kay’s 1969 revelation that color term categorizations are universally centered on generic, basic-level foci, with less salient (sometimes unnamed) colors referred to by extension from these focal points in “color space”. It was suggested soon after by Eleanor (Heider) Rosch (1971), that semantic reference in other perceptual domains (besides color) might involve generalization from focal exemplars as well. In 1973, Rosch was the first cognitive anthropologist to clearly articulate the idea that domains (in general) develop around salient prototypes, and that categories in real languages may not simply consist of combinations of already learned attributes. Rather, instead of categories consisting of a catalog of necessary and sufficient criterial features, they are conceived of as a gestalt. Soon after Rosch’s initial formulation, Berlin, Breedlove, and Raven (1974:53) reported observing focal (i.e., “basic” or “generic”) and peripheral (i.e., “extended”) ranges of botanical categories in Tzeltal plant classifications. The gestalten or prototypical “images” of G2 theory are thought to focus categories around “prototypical instances that contain the attributes most representative of items inside [the category] and least representative

68 of items outside the category” (Rosch 1978:30). The most prototypical category members have more properties (possibly, but not necessarily, criterial attributes) in common with other members of the category. Consequently, categories are “gradient” or “fuzzy,” rather than rigid, clearly delimited, box-like structures, as in G1 theory. Their discrete features undergo “configurational recoding,” where they are “‘chunked’ together to form a single [gestalt] attribute” that indicates whether something is “cat-like, dog-like, has flowers, or berries, or whatever” (D’Andrade 1995:92-93, 102). As Maurice Bloch puts it, “we consider something as a ‘house’ by comparing it to a loosely associated group of ‘houselike’ features, no one of which is essential, but which are all linked by a general idea of what a typical house is” (1990:185). This theoretical flexibility surmounts the primary obstacle that hobbled G1 theory (namely, the fact that definitional features are not necessarily the most salient), by taking into account the influence of both referential, denotative meanings, and of connotative, non-criterial meanings in classification. A number of G2 studies were conducted in the 1970s and 80s (see Burton 1972, D’Andrade et al. 1972, Fabrega 1970, Garro 1986, Kirk and Burton 1977, Kronenfeld, et al. 1985, Rosenberg and Sedlak 1972, Stefflre 1972, Young 1978, Weller et al. 1987, and White 1985 for some key examples outside the genre of color term studies). Although researchers do not always explicitly refer to the prototype-extension theory underlying their approaches, their instruments are generally tuned to identify the most salient attributes (rather than the defining features) of a domain. G2 methodologies for investigating categories can be divided between those that measure perceptual similarity between items and those that measure similarity between subjects. The results of both kinds of analyses can be modeled using the same numerical techniques (hierarchical clustering and multidimensional scaling). Perceptual similarity between items (or categories) is usually measured through item-item or item-attribute comparisons. Item-item comparisons typically involve (a) asking respondents to rate the degree of similarity (or dissimilarity) of all of the possible pairings of all of the items within a domain (see Fillenbaum and Rapoport 1971 and Stefflre 1972 for examples); (b) requesting subjects to undertake a single pile sort of the domain’s members, or (c) having respondents complete a triads test. If multiple consultants are interviewed using the paired item approach, their

69 similarity scores are aggregated. In the pile sort approach, subjects are requested either to make as many piles as they want (an “unconstrained sort”) or to sort the items into a specified number of groups (a “constrained sort”). Results from all respondents are tabulated together in an item-by- item similarity matrix, where each time two items co-occur in a sort, they are given a “point of similarity.” Once all of the subjects’ responses have been tallied, those items that share the most similarity points may be considered to be those that are most similar overall. The triads test is identical to that outlined in the discussion of G1 methods and is tabulated across consultants in the same way as the pile sort just described (i.e., the remaining pair of items in a triad are given a similarity point in the matrix each time a subject keeps them together). Item-attribute comparisons measure the distributional similarity of items in a domain. The first step involves (a) eliciting a list of items in a domain, and (b) collecting native-language sentence frames (analogous to those discussed above under the G1 heading) which make some belief statement about an item (e.g., its attributes, uses, etc., see Figure 2.10 for examples of terms and belief frame statements from D’Andrade et al.’s (1972) study of disease categories in American-English and Mexican-Spanish). Next, the researcher systematically pairs each item in the list of domain members (i.e., categories) with each frame statement. Informants are then asked whether each statement is true or false for each item or they are asked to rank the degree to which a statement is accurate in relation to each category. For each co-occurrence of an item with an attribute (i.e., a sentence frame), one similarity point is entered on an item-by-attribute matrix, or in the case of a ranking approach, the rating is entered in the appropriate cell. As in the other methods already described, those items with the most points (i.e., attributes) in common are inferred to be those that the consultants judge to be the most similar. Sharing and variation in cognition among members of a society have been investigated using data derived from item-attribute comparisons (Garro 1986 and cf. Finkler 1984). In Linda Garro’s study of Tarascan healers and non- healers, similarity scores for each pair of informants were calculated on the basis of the number of times that the informants responded in the same way to item-belief combinations (1986:355). Similarity scores were entered into an “interinformant” matrix analogous to the item by item matrix described above.

70

Figure 2.10. American-English Belief-Frames. Reprinted from D’Andrade et al., 1972:12.

71

The more agreements between two subjects, the more the two could be said to share a common cognitive structuring of the domain (Garro 1986:355-356). The results for all of these computations of similarity (between items and between consultants) may be modeled using hierarchical clustering or multidimensional scaling techniques. Hierarchical clustering graphically represents similarity (perceptual, in the case of a consultant’s similarity judgments, and actual, in the case of similarity between informants) in the form of a “hierarchical clustering tree.” Such trees are generated by (a) clustering the most similar objects, and then (b) continuing to cluster items and clusters in a systematic progression. Multiple linkage methods, or “amalgamation rules,” are possible (see Wilkinson et al. 1996:611-612 for further details). Ultimately, hierarchical clustering techniques arrive at a finished branching tree-structure by combining objects and clusters of objects in a bottom-up fashion (see Figure 2.11 for the resulting “taxonomy” of American-English disease terms investigated in the D’Andrade, et al. 1972 study). The multidimensional scaling procedure produces graphic output nearly identical in form to that of Osgood’s semantic differential technique. However, in the G2 approach, the mapping of similarity in semantic space is based on a much wider array of possible similarity measures. Computer programs for performing the scaling process use a similarity matrix, elicited through any one of the methods outlined above, to plot coordinates for each item in Euclidean space. In the resulting figure, items with the highest similarity ratings are represented as more proximate, while those that are more dissimilar are farther removed from one another (see Figure 2.12 for a multidimensional scaling, or “MDS,” plot from White’s study of perceived similarity of A’ara verbs for interpersonal conflict). Cognitive scientists’ means of theorizing about and modeling similarities bear interesting resemblances to certain theories, methods, and instruments that have been developed in other academic disciplines for classifying (a) biological forms, (b) languages and dialects, and (c) human societies (both living and extinct). The preceding sketch of what G1 and G2 approaches to categorization and classification typically involve will allow us to consider more carefully what kinds of approaches may be most fruitfully used to investigate cognitive change and continuity over time. It provides a critical backdrop for the following discussion of the development of

72

Figure 2.11. Taxonomy for American-English Disease Terms. Reprinted from D’Andrade, et al., 1972:27.

73

Figure 2.12. ‘Two-Dimensional Scaling and Clustering Model of 33 A’ara Verbs of Interpersonal Conflict’. Reprinted from White, 1985:353.

classificatory and historically oriented theories and methodologies in other sciences.

Taxonomy, Phylogeny, and Change

The following pages constitute a review of the history and current status of theory and method of classification in biology, linguistics and cultural anthropology (including archaeology, the study of non-contemporary material culture). In particular, the focus will be on how models of categorization have been used to address issues of history and evolutionary theory in these domains. Each subsection generally follows the chronology of the development of theory and methodology in each of the three sciences with which we are here concerned.

74

Biological Models

Biological approaches to classification and the historical interpretation and import of taxonomies may be roughly divided into three main streams of theory and method: (a) traditional evolutionary taxonomy, (b) phenetics, and (c) cladistics. Each has its own special emphasis and valid range of applications, despite the fact that the proponents of one school or another may laud the supremacy of their own approach over and against the others, standing ever vigilant to denounce the tenets and results of competing dogmas as outmoded, unenlightening, or simply “wrong-headed” whenever the opportunity presents itself. Nevertheless, the three systems are each worth serious consideration as models capable of contributing to the development of a historical approach to G2 cognitive science. Much of the discussion that follows is a dialectic construction based on the history of classification in biology as it is portrayed by proponents of two opposing paradigms: the cladists and the pheneticists. While either’s account may be biased when seen in isolation, when they are taken together and synthesized, a fairly objective description becomes feasible. The basis for identifying taxa and the history of the shift from descriptive to explanatory emphases in traditional evolutionary taxonomy are chronicled by H. Gee (1999), M. O’Brien and R. Lyman (2003), G. Simpson (1945) and P. Sneath and R. Sokal (1973). Their observations are the backbone of the description that follows. The same authors, as well as C. Hill and P. Crane (1982), and O’Brien and Lyman (2003) are the main sources for the summary of instruments and procedures in the traditional model. A. Yablokov (1986) and Sneath and Sokal (1973) provide the critique of traditional taxonomy which initially justified the advent of phenetics. V. Heywood and J. McNeill (1964), Sneath and Sokal (1973) and O’Brien and Lyman (2003) describe the theory behind the paradigm, and Hill and Crane (1982), Sneath and Sokal (1973), Sneath (1964), Sokal and Sneath (1963) and Sokal (1986), the procedures used to actualize it. Its emphasis on accretion of pattern over theories of evolutionary explanation is discussed by Hill and Crane (1982) and O’Brien and Lyman (2000 and 2003). R. Blackwelder (1964), Sneath and Sokal (1973) and Sokal and Sneath (1963) have pointed out that traditional and cladistic approaches are largely phenetic at bottom. The description of phenetic theory and method is based upon their summaries.

75

The summary of cladistic theory is based on Sneath and Sokal (1973), Kitching, et al. (1998), and O’Brien and Lyman (2003). The outline of cladistic methodology is primarily drawn from Kitching, et al. (1998) and is supplemented by O’Brien and Lyman (2003), unless otherwise noted. The critique of cladistics is derived mainly from pheneticists Sneath and Sokal (1973) and the reflections of two prominent cladists, O’Brien and Lyman (2000). Traditional evolutionary taxonomy. Traditional evolutionary taxonomy has roots in classical antiquity (with the systems of Aristotle, Plato, and others), but blossoms in its first “modern” expression with the work of Carol von Linné (Linnaeus) in the mid-1700s—the Systema naturae. As originally devised, the Linnean system was nonevolutionary. Taxa were fixed and unchanging natural units organized in a hierarchical taxonomy based on their natural morphological affinities to one another. They represented archetypes defined by the sum of fixed morphological characters (actually, character states) shared by members of the group, much like the taxa of G1 cognitive studies. While the Linnean taxa were based on the essence or “essential characters” of archetypes, the advent of Darwinian evolutionary theory of descent with modification in the mid-nineteenth century shifted the emphasis from essential characters and the making of taxonomies to a search for “primitive” or ancestral characters (supposedly those features least likely to change over time) and the discovery of categories. Darwin’s theory gave a new explanation for why organisms were similar. In his On the Origin of the Species (1859), he posited that those organisms that were most similar were likely descended from a common ancestor. Although the general acceptance of Darwin’s theory revolutionized the interpretation of biological taxonomies, it did not have much of an impact on the instruments or techniques of classification. The formal method that was recalibrated in the wake of Species would now lead researchers to work in two directions at once. From the perspective of the traditional Darwinian approach, classification should be made on the grounds of both the phylogenetic relationships of organisms and of the evolutionary divergences that succeeded their shared ancestral form. Traditional evolutionary taxonomy thus entails an investigation of both the pattern of relations among specimens believed to be descendants of a common ancestral form and of the inferred process of ancestor-descendant evolution by which they developed.

76

In evolutionary taxonomy, classification is undertaken in two phases of three stages each. The first phase involves sorting individual entities (not archetypes, as in the pre-Darwinian incarnation of the system) into groups (species) “such that within-group formal variation is minimal relative to between-group variation” (O’Brien and Lyman 2003:28). The first step of the first phase is the identification of the unique properties of each taxon (at any given level) through a comparison with other taxa. The second step involves the determination of the common properties shared by certain taxa and the postulation of biological explanations for the similar and different characters of the taxa. The third step is an appraisal of the amount of variation possible within each taxon. The second phase of classification in evolutionary taxonomy is the actual categorization of specimens into a phylogram or taxonomic tree. First, the individual specimens are sorted into variously-sized arrays of related “species” based on phenotypic (behavioral, genetic, morphological, or physiological) similarity in their homologous characters (i.e., those character states present in their presumed common ancestor). Next, the resulting groups are placed into a hierarchy of categories. Finally, analogous characters resulting from parallelism or convergence are excluded from consideration and a phylogram showing both formal similarity and shared ancestry is completed. Analogous characters are independently evolved adaptations for a common purpose. Parallelism is similarity in two distantly related taxa due to adaptation to historical and environmental factors that channel development in a common direction. Convergence occurs when unrelated taxa exhibit identical characters resulting from similar adaptive pathways. Much is assumed a priori in the operationalization and instrumentation of traditional evolutionary taxonomy. Both before and after Darwin, the “authors” of taxonomic systems have selected only those characters of the taxa that they have presupposed are the most important (essential characters in the earlier perspective or ancestral characters in the post-Darwin era), and weighted the significance of each character as they deem most appropriate based on their subjective personal experience and expertise with a particular set of specimens. Even if classifications are based solely on supposed homoplasy (homologous characters), Darwinian Theory provides no real means for objectively recognizing which characters (or character states) are in fact homologous. So, even classifications based on apparently “homologous

77 characters” are necessarily subjective and must always remain hypothetical rather than be taken as statements of fact (O’Bryan and Lyman 2003:30, 31). We will encounter this same difficulty again in the review of cladistic theory and methods, below. The dual-purpose nature of traditional evolutionary taxonomy has made it the target of critiques by both pheneticists and cladists. Both hold that the phenotypic and the phylogenetic cannot be adequately addressed simultaneously and given equal weight, as in the traditional system. According to biologists outside the camp of traditional evolutionary taxonomy, it is necessary to separate the phenetic and cladistic aspects of taxonomic relationships. Phenetics. Biological pheneticists focus their attention solely on the phenotypic resemblances (external markers of similarity) among taxa, excluding phylogenetic approaches to the origin of resemblances on several grounds. They primarily argue contra evolutionary taxonomy and cladistics that (a) the evolutionary branching sequences of phylogeneticists (both traditional evolutionary taxonomists and cladists) are in actuality only inferred from largely phenetic relationships of resemblance among organisms; (b) that phylogenies require inferences about evolutionary directionality, even when the actual history of a group is unknown or unknowable based on current fossil evidence, and that (c) even the “criteria for choosing the ancestral forms in a phylogeny are phenetic and are based on the phenetic relationship between putative ancestor and descendant” (Sneath and Sokal 1973:10). As a result, whatever traditional or cladistic phylogenetic hypotheses may be put forward, they are ultimately founded on what are essentially phenetically-based inferences. Pheneticists are not primarily concerned with evolution, although their classifications may be interpreted from an evolutionary perspective. Since pheneticists generally forego subjective consideration of the origins of resemblances to focus solely on character correlations among phenotypes, it is not necessary or even desirable to make a distinction between analogous and homologous characters. Instead, the maximal number of characters of both types (which together yield the greatest possible information content) is used to classify objects based on their overall phenotypic similarity. While the individual and aggregated units or phenomena to be classified (called operational taxonomic units or OTUs at either level) are describable in terms

78 of multiple characters and their states (i.e., they are polythetic), not every individual classified within an OTU will share every character of the larger OTU. As with prototypes in G2 cognitive theory, there is no set of necessary and sufficient characters required for membership in a taxon, since categories are gradient aggregations whose members are more similar to one another than they are to those of any other taxon, based on a whole constellation of attributes or characters. The process of phenetic classification may be either qualitative or quantitative, but pheneticists have, for the most part, converged on the construction of taxonomies by numerical means. There are five main steps in constructing a numerical taxonomy. First, as arbitrary as possible a collection of specimens (OTUs) is selected. Second, the characters of the OTUs are discovered, listed, measured, and coded. Characters may be “two- state” (binary) characters that are either present or absent (coded as 0s or 1s) or “multi-state” characters that are measurable in orders of magnitude (coded as continuous variables, rank orders, or percentages). Third, overall phenetic affinity (i.e., similarity) is calculated between specimens by comparing the coded measurements of each in turn with every other. Fourth, specimens are then sorted into clusters (by cluster analysis or some other similar, and generally automated, method) called phenons and are displayed in a dendrogram (see figure 2.13) or a multidimensional scaling scatter plot. Last, the characters are re-examined to find those that are most constantly present among clusters of taxa and therefore most useful for developing taxonomic keys. Phenetic analyses yield results that may be ordered in a phenogram (a phenetically generated dendrogram) that displays overall phenotypic similarity between aggregates in hierarchical, nested units. While traditional evolutionary taxonomists and cladists acknowledge the usefulness of phenetic taxonomies in regard to arranging specimens in groups by phenotype, they also take frequent opportunity to disparage phenetics as less important or interesting than their own approaches. This tendency derives from the fact that phenetics is not as deliberately historical a scheme of classification as the other two approaches, and, in its emphasis on accretion of pattern irregardless of character origins, it is generally divorced from theories of evolution. However, as already alluded to above, pheneticists have shown that phenetic analyses (even if they are non- numerical) are a necessary (though frequently slighted or unacknowledged)

79

Figure 2.13. Phenogram of 81 species of gulls and terns based on UPGMA cluster analysis of correlation coefficients of 51 skeletal measurements. Reprinted from Sneath and Sokal 1973:262. 80 first step before more evolutionarily-oriented investigations can be undertaken. Without some kind of prior grouping (almost universally based on phenotype), there would be no set of specimens to compare and base a phylogeny upon. Indeed, the data for cladistic and traditional evolutionary inferences are “invariably phenetic” at bottom (Sneath and Sokal 1973:60, 319 and also see Sokal and Sneath 1963:220). Despite the criticisms, phenetic similarity may be taken as an initial indicator of where to look for phylogenetic relationship, because, while phenotype-based classifications may include non-homologous characters, “patristic similarity” (homoplastic affinities resulting from common ancestry) “comprises the major part of phenetic similarity” (Sokal and Sneath 1963:220). Cladistics. Cladistic methods of classification also group taxa into hierarchical structures based on comparison of characters. However, cladists argue (in opposition to pheneticists) that homologous characters should carry greater weight than analogous ones, and further (in opposition to classical evolutionary taxonomists), that a particular sub-class of homologous characters are the only characters upon which a classification should be based. Following Hennig (1965 and 1966), the founder of cladistics, cladists distinguish between symplesiomorphic homologous character states (symplesiomorphies) and synaphomorphic homologous character states (synaphomorphies). Symplesiomorphies are character states that are identical in two or more descendants as well as in their preceding common ancestor(s). Synaphomorphies are “evolved novelties”; adapted character states shared by two or more descendant taxa and their immediate common ancestor, but not with more remote common ancestors. Cladistic theory relies solely on synaphomorphic shared derived characters for the development of phylogenetic taxonomies, because (since modification “goes hand in hand” with descent) cladists believe that reliance upon symplesiomorphic characters can be misleading. Presumed phylogenetic relationship (i.e., recency of common ancestry) is the only criteria for cladistic classification. In Hennigian cladistics, the apomorphies (derived character states), thought to signal phylogenetic descent in monophyletic lineages, are identified a priori, even before construction of a cladogram begins. A monophyletic lineage, or clade, is a complete set of taxa sharing a common ancestor, plus that common ancestor. Apomorphies and plesiomorphies

81

(symplesiomorphies) for each individual in the study set (also called the ingroup) are recognized and coded (apomorphies as 1s and plesiomorphies as 0s). To be included in the ingroup, taxa must share at least one presumed apomorphic state (i.e., a synapomorphy). The study set is initially determined by comparison with one or more presumed outgroup taxa (which may share plesiomorphic characters with the ingroup taxa), that do not share the apomorphic state of the ingroup and through their contrastive nature thereby polarize the ingroup taxa into a unit, or monophyletic group. The distinction between ingroup and outgroup creates the first branch in a tree-like cladogram, a nested representation of character relationships, with taxa (composites of the characters) as the terminal nodes [see figure 2.14(a)]. The approach is then repeated within the ingroup node (creating another, secondary ingroup and outgroup) and is again repeated with the newly derived ingroup, and so on, until all taxa are united into monophyletic groups by a continual process of exclusion of plesiomorphic character states. See figure 2.14 (d) for a completely resolved cladogram of the taxa shown in 2.14 (a). Multiple solutions may be possible, so cladists generally use computerized heuristic methods to arrive at the most parsimonious solutions-those with the least steps, or fewest presumed evolutionary events. According to Kitching et al. (1998:41), six taxa produce 105 fully resolved cladograms, and O’Brien and Lyman (2003:63) note that a typical cladistic analysis may produce thousands to millions of solutions. Aside from the problem of selecting the one optimal cladistic representation of phylogenetic descent from among possibly millions of choices, cladistics is also confronted with the same problem of homology with which traditional evolutionary taxonomy is faced. Methods employed by cladists for determining which characters are homologous (as well as ways for distinguishing between apomorphic and plesiomorphic homologies) and even the means by which species are selected to be compared as “related” are based largely on phenetic considerations. Cladistic outgroup comparison is based on inherently circular reasoning because it “presumes that we know a priori that the chosen outgroup is no more closely related to one group than it is to any other group under consideration” (O’Brien and Lyman 2000:225). The three main biological approaches to classification (traditional evolutionary taxonomy, phenetics, and cladistics) each have their merits.

82

Figure 2.14. Determination of the most parsimonious cladogram using Hennegian argumentation. Reprinted from Kitching et al. 1998:39.

While the stated goal of evolutionary taxonomy and cladistics (to determine the phylogeny of a set of taxa) seems loftier at first glance than that of phenetics (simple classification based on overall similarity), it is clear upon closer inspection that the two main phylogenetic approaches cannot be operationalized without reference to phenetics. Numerically based cladistics, which has more than once been heralded (by cladists) as the triumphant arch-rival of phenetics, has also been roundly criticized as being little more than an intuitive form of “statistico-phenetic taxonomy” (O’Brien and Lyman 2003:46). Other scientific disciplines (linguistics and cultural anthropology, to name those most relevant to the current study) have developed theories and methods similar to those outlined above. At times linguists, archaeologists, and others have directly borrowed models from biology (the “biological metaphor” is rife within the social sciences, see Hoeningswald and Weiner 1987). In some instances, theorists’ similar ideas have been the results of interdisciplinary cross-pollination and at other times have arisen from common concerns (instantiated in the prevailing habitus or scientific

83 zeitgeist) leading to convergent, but still independent innovations along parallel lines. Although biological metaphors are prevalent and have proven useful in the social sciences, an analogy is not the same as identity. Languages, cultures, and artifacts are not the same kinds of organisms, in all respects, as those studied by biologists. If we hope to move closer to developing a method for examining cognitive categories in diachronic perspective, a review of historically oriented taxonomic approaches parallel to those found in the biological sciences is also warranted—but in the disciplines of linguistics and cultural anthropology (the pursuits most closely associated with cognitive studies, historically).

Linguistic Models

Potentially relevant linguistic approaches to classification and change can, for the sake of convenience, be divided into three main groups: (a) traditional models (including the phylogenetic approach pioneered by the Neogrammarians and the dialectological approaches of adherents to the diffusionist “wave model” of language relationship); (b) phenotypically- based, quantitative models (primarily those of numerical dialectologists and, less so, of lexicostaticians and mass comparativists), and (c) variation and contact-sensitive models (including those put forward by creole theorists, developmentalists, and other variationists). It is fair to say that phenotypic and variationist approaches derive from, and to a large extent are contingent upon, various aspects of the traditional models. They simply represent different (broader or more narrowly focused) emphases and the elaboration of particular aspects of the more basic models. The following descriptions and critiques of theory and method in traditional “tree” and “wave” approaches to taxonomy, history, and reconstruction in linguistics are derived entirely from Leonard Bloomfield (1933), Lyle Campbell (1999), Terry Crowley (1996), Roger Lass (1997), and R.L. Trask (1996), unless otherwise specified. Phenetic and phenetic-like models include dialectometry, lexicostatistics, and multilateral mass comparison. Sheila Embleton (1991), Hans Goebl (1991), and John Nerbonne and William Kretzschmar are the basis for the discussion of dialectometry that follows. The description of lexicostatistics comes from Campbell (1999), Crowley (1996) and Isidore Dyen

84

(1975). Campbell (1999), Lass (1997) and Merritt Ruhlen (1987) are the authorities underlying the summary of mass comparison. The discussion of pidgin and creole theory is derived from Crowley (1996), Sarah Thomason (2001), and S. Thomason and T. Kaufman (1988). Some of the key figures in variationist and developmental linguistics, including C.M. Bailey (1978 and 1966), S. Mufwene (2001), P. and Mühlhäusler (1985), are the primary sources underlying the description of this recently emerging paradigm. William Croft (2000) and Crowley (1996) have also contributed. Traditional. Of the two “classical” approaches to linguistic classification (the genetic and the diffusionist models), the theoretical and methodological ideas of the phylogenetic school have maintained pride of place since their definitive formulation in the nineteenth century. Like traditional evolutionary taxonomy and cladistics in biology, the “family tree” or Stammbaum model in linguistics (August Schleicher, circa 1861-1862) is based on the twin principles of descent with modification. It has been remarked that Schleicher’s ideas parallel those of Darwin, and debate has gone back and forth about whose ideas may have influenced whom (Jankowsky 1973:100, 499). However, both of these scholars had a penchant for Hegelian philosophy, and it is just as likely that their ideas were inspired and independently arrived at as a result of their shared intellectual predilection for his work (Jankowski 1973:100). The family-tree model begins with the assumption that currently existing languages deriving from a common origin (i.e., from a single, uniformly homogeneous ancestral proto-language or parent language with no dialect variation) and sharing a history together are genetically related, and therefore may be classified as belonging to the same “language family.” The initial identification of a language family is much like the selection of specimens for comparison in traditional evolutionary and cladistic models of biology. The idea of a set of languages’ relatedness is initially posited on the grounds of what are essentially phenotypic resemblances in “marked and pervasive” similarities of characters among a group of languages that intuitively seem to go together. It is phenotypic resemblance that first permits the assembly of a data set for further consideration. A second assumption of the family-tree model is that proto-languages experience sharp and sudden splits, where the parent language splits up into “daughters” (“sister” languages to each other), which may later diversify in

85 the same fashion into daughters of their own. In a third, but related assumption, it is assumed that the related daughter languages have no further contact subsequent to their divergence from the parent stock. These three assumptions lead adherents to this model to rely solely upon homoplastic characters and do not permit the consideration of analogous features. Anticipating the cladists of biology by several generations, the Junggrammatiker, or Neogrammarians (the group of linguists who initially adopted and then further developed Schleicher’s model), and after them, modern adherents to the Stammbaum theory, distinguish two sources of heritable similarity. First, there are symplesiomorphic “shared archaisms” or “shared retentions,” which are features that have been held over from the proto-language and are therefore thought to be of little or no value in establishing groupings based on common descent because of the likelihood that a large number of features would be so retained. Second, there are synaphomorphic “shared innovations” (i.e., shared departures from some trait in the proto-language) which appear in some members of the family but not in others. It is presumed that shared innovations first appear in a daughter of the proto-language that subsequently splits into daughters of its own, each of which then inherits the change from their shared immediate common ancestor, i.e., the intermediate parent between the sisters and the original proto-language. Other languages of the family that do not have the innovation are presumed to have split off (or be descended from an ancestor that split off) at an earlier point in the family’s history. A shared innovation is therefore seen as evidence for recency of common ancestry and is thus, for this model, “the only generally accepted criterion for subgrouping [languages of a family]” (Campbell 1999:104). The Stammbaum theory is operationalized through the paired techniques of (a) “the comparative method” (alternately called “comparative reconstruction,” by some historical linguists) and (b) subgrouping languages based on the shared innovations identified through the application of the comparative method. The following paragraphs provide a simplified description of the basic techniques that make up the method. The first step in the comparative method is deciding by inspection (i.e., based on phenotype) that a group of languages are likely to be genetically related. Next, potential cognates (words or morphemes of similar

86 shape and meaning in the proposed sister languages that are thought to derive from a single form found in the shared proto-language) are assembled for comparison. These are then examined for systematic (i.e., non-chance), recurring (regular) sound correspondences for “each set of sounds that appears to be descended from the same original sound” (Crowley 1996:93). Then, a plausible hypothetical reconstruction of the ancestral proto-sound for each set of sound correspondences is postulated. Once the phonology (the sound system and the rules for combining sounds) of the proto-language has been “reconstructed,” the lexicon of the proto-language is retrieved by putting reconstructed sounds together “following the order in which they appear for the cognates” (Campbell 1999:131). Although analogous methods exist for reconstructing a proto-language’s syntax and semantics, historical linguists generally tend to base their classifications of a family’s members on phonological and lexical features. Once shared innovations of a subset of sister languages descended from the same proto-language, but presumed (based on the sharing of innovations) to share a more recent intermediate ancestor, have been reconstructed, subgrouping is a fairly straight-forward matter. Those languages that are believed to derive from the same intermediate proto-language are placed in the same subgroup on a branching family-tree diagram of the larger proto- language. Intermediate ancestral languages may in turn be descended from other, higher level intermediates between themselves and the original proto- language. Subgroups may be nested into higher-order subgroups of greater presumed time-depth, themselves. See Figure 2.15 for a family tree diagram of the languages descended from Proto-Indo-European (Trask 1996:184). Because innovations may be likened to copying errors in biosystematics, and shared errors therefore define new taxa or nodes on a branching tree, some linguists have in recent years adopted the full panoply of vocabulary, representation, and argumentation from biological cladistics for addressing problems in historical linguistics (see Lass 1997 for a prototypical example). This allows for the quantification of relationships based on synaphomorphies (shared innovations) through the standard cladistic means described earlier in this chapter. Frequently, the comparative method reconstructs intermediate nodes (i.e., intermediate proto-languages) in a family tree that are known to have never existed. This rather serious problem is a natural outcome of a dire

87

Figure 2.15. The Indo-European family tree. Reprinted from Trask 1996:184.

88

flaw in the application of the theory. According to Leonard Bloomfield, the earliest students of the Stammbaum approach (who applied it to the reconstruction of the Indo-European proto-family) accepted the historical reality of uniform proto-languages, sudden and clear-cut splits, and non- interacting daughters as common-sense realities, without ever recognizing that the family trees resulting from these assumptions are, circularly, “merely a statement of their method” (1933:311). In point of fact, no speech community is ever completely uniform; languages seldom split suddenly, and dialects frequently interact. As a result, in the words of Bloomfield: “the comparative method cannot claim to picture the historical process” (1933:318). Subsequent generations of linguists have since refined the family-tree model and now allow for non-catastrophic change, variation (with the simultaneous coexistence and competition of plesiomorphies and apomorphies), incomplete changes resulting from various environmental conditions, gradual diffusion of changes throughout the system, and conditioning effects of various social factors on linguistic changes. Many (if not all) of these allowances were made in reaction to the valid challenges of the wave model of language change and will be addressed in the subsequent pages. Even so, many traditionalists argue, as Roger Lass does, that “the essential and privileged pattern for any genealogy is [in spite of these deficiencies, still] the Stammbaum . . .” (1997:113). The wave model (or Wellentheorie, as it was first termed in German in 1872 by its best known early proponent, Johannes Schmidt) is an alternative historical model that deals primarily with contact-induced changes resulting from the interaction of languages and dialects (an important and widespread phenomena for which the Stammbaum model, as originally presented, makes no provision). Even historical linguists whose primary work utilizes the family tree model acknowledge that the wave model is a more realistic portrayal of language change than the Stammbaum model (if not “the reality” underlying linguistic diversification) (see Campbell 1999:188, Crowley 1996:241, and Trask 1996:185). Based on the existence of “special resemblances” that “link [geographically proximate] languages which in the tree diagram are shown as having split apart very early, and as belonging to separate branches”

89

(Bloomfield 1933:317), adherents of wave theory contend that linguistic changes diffuse or emanate from centers of dispersion “as waves on a pond do when a stone is thrown into it” (Campbell 1999:189). These linguistic “waves” may travel different distances at different rates. They are seen as a consequence of continuing contact and mutual influence among dialects which have begun to gradually diverge. Rather than the exceptionless sound rules that apply to all words in a cognate set propounded by the family-tree model, each word has its own history. In essence, the affirmations of wave theory contradict each of the three main assumptions of the Stammbaum theory. Rather than a homogeneous parent stock, we find a heterogeneous one. Dialects are recognized as only gradually becoming more dissimilar and not splitting from a parent suddenly, sharply, and decisively. Finally, wave theory holds (accurately) that dialects and languages continue to interact and influence one another, sometimes long after they have “diverged.” Thus, languages are thought to be most realistically grouped on the basis of their sharing (or being delimited by) bundles of ‘isoglosses’ (dialect features or linguistic variants and the wave-like lines on a map that represent their geographic boundaries. Wellentheorie has engendered its own methodology for approaching linguistic variation, a kind of dialectology that produces “wave-model diagrams.” In standard dialect geography, isoglosses are drawn on maps which represent actual geographic relationships between the locales where given linguistic varieties are spoken. In the production of wave-model diagrams, language names are simply placed in “some convenient arrangement” on a page (Trask 1996:185). Iso-glosses (linguistic features graphically represented by a curved enclosing line) surround those languages or dialects which share a particular feature. In most incarnations of the method associated with the wave model, the isogloss lines represent shared changes since divergence from the proto-stock that occurred in only some of the languages (especially those changes not concordant with the standard family tree diagram for the family). However, in other instances, the isogloss lines may simply represent significant shared features in general which are not necessarily identified as the general). The ‘bundling’ of numerous isoglosses typically reveals coherent results of change (cf. the two diagrams below in figures 2.16 and 2.17; the

90

2.16. A wave diagram of the Germanic family. Reprinted from Trask 1996:186. 91

Figure 2.17. Some overlapping features of special resemblance within the Indo-European languages, conflicting with the family-tree diagram. Reprinted from Bloomfield 1933:316.

first is based on shared changes, the other on special resemblances in groups that generally reflect the same major splits that a family tree diagram does, while simultaneously providing a more nuanced and intricate picture of the continuing interactions of languages and dialects that have previously ‘diverged’ from one another (as the Stammbaum model would have it). In some instances, this allows the dialectologist to draw a family tree based on the strongest relationships attested by an abundance of isoglosses. Sometimes, isogloss lines do not bunch together into ‘bundles’ that are clear enough to draw a family tree. We may speak of “dialect chains,” or even “language chains” of varieties that appear to belong to more than one subgroup of the same family in cases where “characteristic provincial peculiarities [are found] . . . prevailing in a solid core, but shading off at the edges (Bloomfield 1933:340). Similarly, we may come to the same

92 conclusion in cases where “immediately neighboring dialects only exhibit slight differences from each other,” but “as geographical distance increases, so too does the extent of difference between dialects” (Crowley 1996:247). One way of dealing with this issue is to quantify the degree of similarity between such non-discrete groups using the techniques of dialectometry, a phenetic-based approach that will be explored in the next subsection of this chapter. In an even more extreme manifestation of the diffusionary phenomena central to the wave model, under some circumstances, bundles of numerous isogloss lines (representing phonological, morphological, lexical, or syntactic features) may show coherent groupings of geographically proximate languages already known to be genetically unrelated or at least genetically distant. Such a convergent grouping is known in German as a Sprachbund and in English as a “linguistic area.” Such areas support the wave theory of diffusion but may (a) make genetic classification more difficult, and (b) cause the confident representation of relationship by means of the Stammbaum model to be further suspect in these cases. The multilateral interactions of languages in a Sprachbund, however, only represent a “greater than average degree of contact between languages” (Crowley 1996:317). An even more intensive degree of contact and mutual influence will be considered in the discussion of pidgins and creoles under variationist approaches, below. The wave model of taxonomy has been charged with three main short- comings. The first, and perhaps the most damning, is that its diagrams are too “tedious and cumbersome to prepare and draw” in comparison to “the simpler and more vivid trees” of the Stammbaum model (Trask 1996:187). The second criticism is that researchers who draw wave diagrams representing exceptional changes (i.e., those that contradict the standard Stammbaum of a family) would be unable to recognize the exceptions apart from accepting the changes as normative (as the comparative method they so often denounce would declare them to be). Of course, this charge does not apply to those who draw isoglosses for features “in general,” without necessarily distinguishing (a) innovations since the divergence of the proto-languages from (b) features resulting from other causes of similarity. The third and most serious accusation, from a technical perspective, is that wave diagrams are incapable of representing multiple stages of development simultaneously (as the tree model so aptly does).

93

It is clear that both the family tree model and the wave model each have serious defects, and that neither is completely sufficient in and of itself. Various resolutions to the conflict between these schools of thought have been proposed, including primary reliance on the tree model, with supplemental insights provided by the wave model (the typical response). Other scholars are less explicitly partisan, as when Lyle Campbell notes that “both are needed,” even though his handbook gives only a handful of pages to Wellentheorie in comparison with his lengthy treatment of the family-tree approach (1999:191). Perhaps the most creative (and potentially useful) rapprochement has been that argued in an essay by R. M. W. Dixon (1997). He proposes that historical linguists should consider a “punctuated equilibrium” model of language change:

[Where,] during a period of [socio-political and sociolinguistic] equilibrium, linguistic features tend to diffuse across the languages of a given area so that—over a very long period—they converge on a common prototype. Then, during a period of punctuation [due to natural causes, technological developments, movements of people or secular or religious imperialism]— characterised [sic] by expansion and split – a series of new languages will develop, diverging from a common proto-language. (1997:3-4)

Dixon’s punctuated equilibrium model suggests that the family-tree model is only useful for periods of punctuation (which are, in reality, few and far between) and that the diffusion of areal features (as per the wave model) and the convergence of languages “on a common prototype” during long- term periods of equilibrium is more typical of the long march of linguistic history (Dixon 1997:4, 67). He argues further that, as languages in an equilibrium situation continue to “converge toward a common prototype”, the convergence will eventually “obscure the original genetic relationships [of the languages in the area]” (1997:96). If he is correct, and periods of equilibrium truly are the normative state of languages, then perhaps a totalistic emphasis on genetic relationship as the primary form of historical relationship worth studying may be misplaced. Phenetic. Phenotypically-based quantitative models of language history, like other more recent models, derive much of their theory and instrumentation from the traditional tree and wave approaches. Dialectometry primarily operates on principles characteristic of Wellentheorie, although one of its subbranches, lexicostatistics, operates under principles derived 94 from the Stammbaum model. The third phenotype-centered approach we will address, multilateral mass comparison, also makes its claims based on the family tree model. Generally, the consensus on phenetic models is that, as in biosystematics, they are potentially good places to start looking inductively at interrelationships between taxa. Lexicostatistics and mass comparison, as avowed “phylogenetic” approaches, are particularly emblematic of a troubling trend where models whose use may properly extend only so far as a “beginning” step in investigating taxonomic relationship are also frequently (and unwisely) put forward as producing final results when they should not. Dialectometry, on the other hand, seldom makes strong claims about phylogenetic relationship. Instead, it typically treats both phylogenetic and diffusional input on an equal footing to paint a picture of overall historical relationship between linguistic varieties. Dialectometry, as its name implies, is “the study of quantitative measures of distance between dialects” (Embleton 1991:267). It is specifically used to address the problem of non-discrete subgroups in dialect and language areas and language chains through the application of principles derived from numerical taxonomy (i.e., phenetics) to linguistic data. Its main emphasis is on elucidating “our perception of dialect areas” (Nerbonne and Kretzschmar 2003:246, italics added). Dialectometry primarily leans toward wave theory in its ability to incorporate non-genetic special resemblances into its analyses. As in the more general isogloss-based forms of dialectology, it is typically impartial to the source of inheritance (i.e., whether a shared linguistic feature is originally genetic or borrowed). Dialectometrists hold that the systematic numeric combination of numerous “low ranking [geolinguistic] patterns” will aid in the discovery of previously unseen “higher ranking patterns” (Goebl 1991:277). As in biological phenetics, overall similarity in phenotype ought to be indicative of a higher degree of underlying patristic (inherited) similarity. As a case in point, recent work by Kessler (1995) has shown that results derived dialectometrically may parallel traditional genetic results (generated through application of the comparative method) rather closely. Dialectometry is, in the end, only a supplement to other forms of dialectology and is not meant to be a replacement of them.

95

The theory and method of dialectometry is almost identical to that of the wave model. However, its unique contribution, as first noted by J. Séguy (1971), is in the aggregation of individual similarities and differences for the global comparison of linguistic varieties by paired comparisons with one another. Séguy’s original proposal (upon which all subsequent dialectometry has been based) was that “one [should] simply count the number of overlapping features between any two data collection sites” using a technique where sites that show the same feature “are counted one point more similar” than sites that do not share that same feature, thereby generating similarity scores between dialects or languages based on their aggregated features (Nerbonne and Kretzschmar 2003:246). While the method has since evolved in its finer details, the general principles put forward by Séguy remain the core of the endeavor. Various techniques are used to represent similarity between linguistic variants, including specialized maps, hierarchical clustering methods, and various kinds of scaling (including multidimensional scaling techniques). By aggregating similarity into scores that may be easily ranked, dialectometrists surmount the “tedious and cumbersome” problems faced in preparing and drawing wave-model diagrams. One of the strongest advantages of dialectometry is that it does not require the positing of proto-forms or of monophyletic lineages (though it may suggest where to look for them). The non-assumption of monophyly becomes all the more important in light of creole and variationist theories about language genesis and development, to be discussed in the following subsection. Other phenetic approaches to historical relationships between languages (lexicostatistics and mass comparison) have (rightly) suffered a considerable amount of criticism based on their claims to represent phylogenetic realities. Dialectometry, however, is immune to such charges as long as it focuses on the sharing of influences in general. It is fairly plain that both genetic features (those received through “vertical transmission”) and borrowed characters (those received through “horizontal transmission”) are heritable. As an attempt to look at shared overall history, dialectometry is eminently successful in quantifying degrees of mutual influence among a set of languages. Lexicostatistics, another quantitative means of measuring similarity between dialects or languages, is a special kind of vocabulary-based dialectometry (although it had been widely practiced for years before Seguy

96 coined the term dialectometry). However, unlike adherents to its much broader adoptive parent, lexicostaticians deliberately attempt to exclude inheritance by borrowing as a “quick and easy means” of arriving at a genetic classification. It is classed herein as a phenetic approach because its practitioners often score words as cognates based on a kind of loose inspection method where, if “two forms look cognate” they are treated as such without necessitating further inquiry (Crowley 1996:180-181). Terry Crowley provides a succinct description of the major premises (1996:168-170):

Lexicostatistics is a technique that allows us to determine the degree of relationship between two languages, simply by comparing the vocabularies of the languages and determining the degree of similarity between them. This method operates under two basic assumptions. The first of these is that there are some parts of the vocabulary of a language that are much less subject to lexical change than other parts . . . [namely, the] core vocabulary . . . .

The second assumption that underlies the lexicostatistical method is that the rate of lexical replacement in the core vocabulary is more or less stable, and is therefore the same for all languages over time . . . .

If these assumptions are correct, then it should be possible to work out the degree of relationship between two languages by calculating the degree of similarity between their core vocabularies. If the core vocabularies of the two languages are relatively similar, then we can assume that they have diverged quite recently, and that they therefore belong to a lower level subgroup. If, on the other hand, their core vocabularies are relatively dissimilar, then we can assume that they must have diverged at a much earlier time, and that they therefore belong to a much higher level of subgrouping.

There are several major problems with the lexicostatistical approach to taxonomy. First and foremost, lexicostatistics suffers from the same deficiencies of the Stammbaum model which it is attempting to emulate: languages simply do not split suddenly and sharply into daughters with no further subsequent contact (as the tree makers would have it). In addition, it has been discovered that the proposed universal rate of retention is inaccurate and unfounded. Its core vocabulary has also been shown to be highly subject to borrowing and liable to a high degree of inexactitude in the one-to-one matching of vocabulary items to the proposed “core” notions. Nevertheless, lexicostatistics is worth noting briefly, since it involves the application of quantitative measures with an eye for diachrony over the long

97 term (often lacking in some of the most recent work in ‘general’ dialectometry). Another somewhat problematic phenetic approach (again, worth at least passing attention) is that of multilateral mass comparison. Joseph Greenberg and Merritt Ruhlen are two well-known proponents of this approach. Mass comparison is based on the comparison of “a few words” across numerous languages (hence the adjectives “multilateral” and “mass”). It is, like lexicostatistics, an approach based on superficial visual inspection and the idea that the sharing of greater numbers of assumed “cognates” so discovered is representative of a greater degree of genetic relationship. As a reconstructive method, it suffers from many of the same questionable practices as lexicostatistics: reliance on a core vocabulary, a method of fairly superficial nonsystematic inspection, the positing of phylogenetic assertions based on phenetic relationship without reference to ancillary historical support, and the assumption of monophyly and sudden splits. The strongest critique is the accusation that adherents to this approach “pick and choose” a word here and a word there from different languages to reconstruct proto-forms. However, as Ruhlen notes, at least the major players in the field generally limit themselves to comparing already established “language families, not individual languages” (1994:279). Theirs is not intended to be a method of reconstruction, but rather, of taxonomy. Aside from the warning that the less careful application of multilateral mass comparison provides for developing a method of studying cognitive diachrony, mass comparison gives a hint of another complicating factor with which we may be faced: the existence of “macrophyla.” Language families (and the proto-languages that they supposedly descend from) may themselves group into many more levels of higher-order relationships. If such is the case with cognitive systems, it may be a considerable challenge to settle on a “cut-off point” for identifying the optimal degree of inclusion in the positing of genetic relationships. This serves as yet another warning that the pursuit of a phylogeny before attempting to establish a basic phenetic taxonomy for its own sake and for what it can tell us about degrees of overall relationship in general may be something of a garden path. Variation and contact. Variationist and contact-centered models may be seen as two sides of the same coin, since contact between mutually dissimilar

98 linguistic varieties (of whatever degree of dissimilarity; whether it be between dialects or languages) is impossible without there being enough variation present to consider a situation to be one of “contact” between entities that vary enough to be identifiable as “non-identical.” These approaches to language history raise several more interesting concerns for linguistic taxonomy. Earlier, the wave model drew our attention to the problem of contact among “lects” in general (lect is a catch-all shorthand term for linguistic varieties including languages, dialects, sociolects, and other sorts of variants). We also reviewed how these contacts might affect a genetic “signal.” It was noted that the Sprachbund, or linguistic area, is only one outcome of a high degree of contact among variant linguistic forms. Sarah Thomason (2001) has proposed a typology of contact-induced linguistic results and processes, in which she refers to cases of even more “extreme language mixture” among language variants: what linguists have termed pidgins and creoles. Pidgin languages involve a process of linguistic negotiation arising in situations where speakers of two or more languages are in close contact and urgently need to communicate regularly with one another for limited purposes (such as trade). They do not have a widely shared single language in common (nor the social opportunities to acquire any significant degree of competence in each others’ languages or in a dominant language present in the contact situation). The communicative systems developed under these circumstances are essentially negotiated simplifications of the languages in contact— socially and linguistically restricted codes—with less “linguistic material” than the contact languages involved. The vocabulary of a pidgin language is typically drawn from one of the interacting languages, most often the socio-politically “dominant” language (also called the “superstrate language” or “lexifier”). The grammar does not necessarily come from any single language, and generally involves greater regularity and less redundancy than the grammar of the dominant language. However, there are often strong structural parallels between the pidgin’s grammar and that of the substrate (or subordinate) language(s), so much so that this has led some linguists to conclude that pidgin lexicons are primarily derived from the superstrate language and pidgin grammars from the substrate language.

99

Pidgin languages are typically “nobody’s native language.” However, when a pidgin becomes the native language of a community (typically the children of its first speakers), it undergoes “rapid lexical and structural expansion” and can thereafter be termed a “creole language” (Crowley 1996:267, also see Thomason 2001:262). Creoles are hybrid or mixed languages spoken as a “mother tongue” by a speech community (not by individuals who have acquired the pidgin as their native tongue). Even after the genesis and stabilization of a creole, the superstrate/substrate distinction may still be observable. While the superstrate may seem the most obvious contributor to the mix, there may still be a significant degree of “substratal interference,” or influence, on the fully developed creole. As inherently “mixed” languages, pidgins and creoles present a particularly knotty problem for linguistic taxonomists: if their lexicon is in actuality derived from one language and their grammar from elsewhere, then pidgins and creoles have multiple ancestors, i.e., they are polyphyletic rather than monophyletic. This has led some to the conclusion that, since creoles do not fit in with the phylogenetic (i.e., monophyletic) model of linguistic evolution, they are simply not able to be classified genetically. Others, however, have argued that there is “no obvious yardstick for determining the point at which influence from other languages makes the resulting variety nonclassifiable genetically” (Mufwene 2001:129). A few linguists have therefore defended the applicability of the family tree model to pidgins and creoles, while others have questioned whether its principles should be re-examined and ‘tuned-up’ in light of the “complex process of language speciation” behind “blending inheritance” (Mufwene 2001:141). Yet another contingent has emphatically stated that the Stammbaum model is completely invalidated and made defunct by pidginization and creolization phenomena, and is, in the final analysis, inapplicable to any group of languages. Those who argue that contact phenomena nullify the assertions of the family tree model altogether (whom I shall refer to as “extreme variationists” in the following discussion) do so largely on the grounds that the variation and interaction between dialects or subvarieties of a language that leads to “normal” linguistic change is not ultimately very different from that observed between languages in a pidginization situation. They assert that the process of creole genesis is actually the normative path

100 through which languages develop, basing their argument on (a) the fact that every individual has membership in multiple speech communities and is therefore at least minimally “multilingual,” (b) that all speakers replicate their target language imperfectly (just as pidgin and creole speakers imperfectly replicate the superstrate language), and (c) that no languages have ever developed in total isolation. There is always some degree of contact and mixture between language systems and sub-systems. Extreme variationists point to the fact, first noticed by the wave theorists, that there are “numerous syncretisms in language-mixing which can only be assigned to two or more parent languages at the same time”; they also observe that mixing “often leads to new constructions which, apart from properties belonging to both parents, exhibit reinterpretations and universally motivated developments” (Mühlhäusler 1985:80). One of the best known variationists, Charles Bailey, has pointed out that mixture among lects is always present and rarely absent in any and all language systems (1978:99). A few representative quotes should suffice to paint a fuller picture of the extreme variationists’ viewpoint:

Creole vernaculars are not outcomes of abnormal, unusual, or unnatural developments in language evolution. Rather, they make more evident restructuring patterns that must have taken place in the evolution of other languages. (Mufwene 2001:192)

All internal and external . . . factors considered, there is no particular reason for assuming that creoles developed by any restructuring process that distinguishes them from other languages. (Mufwene 2001:197)

[No] . . . changes within or across systems occur without contact with another language variety. (Bailey 1996:258)

Indeed, it should be asked whether there is any other way for a new language to come into existence than as a pidgin becoming a creole . . . (Bailey 1996:288)

[The] only legitimate case for [the family tree model’s] . . . application—the development of two or more distinct systems out of a single system by the way of splitting up or diffusion, has in all likelihood never occurred. (Mühlhäusler 1985:81)

[Any new system] . . . is created by heterosystemic mixture. Thus, every legitimate node (every node representing a new system) on a so-called family tree must have two or more parents. (Bailey 1978:99)

101

While some of the theory that has developed out of creole linguistics and variationist studies calls for a radical reappraisal of the traditional family tree and wave theories, even the more extreme variationists are hesitant to discard traditional methods. Bailey and Mühlhäusler, the two variationists who have been the most visible (and vociferous) opponents of traditional theory, have also stated that the comparative method remains indispensable despite their critique of the theory underlying its interpretation. However, they argue further that the comparative method is inadequate without the use of supplemental descriptive methods and a better understanding of “discontinuities and contact-phenomena” (Mühlhäusler 1996: 6-7). The two main methodological contributions of variationists are found in the descriptive techniques advocated by Bailey and Mühlhäusler. Reacting against the application of tree-diagrams to creole languages, Mühlhäusler has advocated the use of diagrams that show the continuing and mutual influences among lects in situations of pidginization or creolization (see figures 2.18 and 2.19). Bailey has adopted and expanded the method of Elliot et al. (1969) for arranging variant lects into a “polylectal grid” showing patterns of implicature. The polylectal grid aids in identifying which prior features are implied to be already present in a lect from the presence of a particular feature that requires them to already be present (see Mühlhäusler 1996:7-14 for an overview of Bailey’s reworking of this technique). The implicational scales so derived allow lects to be classified in terms of a “grammatical system” by bundles of features that imply the prior presence of more basic, constituent features (see figure 2.20 for an example from Bailey 1996:21). Like biotic models, linguistic models of history and taxonomy provide another angle from which to consider possible approaches to the elucidation of patterns of cognitive continuity and change. Traditional, phenetic, and contact-oriented linguistic approaches each raise new concerns and offer elements of alternative strategies related to the enterprise at hand. Just as biological and linguistic models intersect with and parallel one another, they likewise interrelate with and mirror approaches to cultural variation and development.

102

Figure 2.18. Pidgin English in the Pacific around 1880. Reprinted from Muhlhausler 1985:79.

Figure 2.19. Linguistic Influences on Tok Pisin around 1900. Reprinted from Muhlhausler 1985:80.

103

Figure 2.20. Listing, classifying, analyzing, and system: data map. Reprinted from Bailey 1996:21.

Cultural Models

Several streams of thought dealing with cultural change and taxonomy are particularly relevant to the current research. While early cultural anthropologists (and recently, cladists as well) have used what they call the “comparative method,” their approach has generally been quite different from that applied to linguistic data. G1 cognitive anthropologists have come the closest to replicating the successes of historical linguistics in this matter, though there have been almost no attempts to use the new-found results to construct taxonomies of thought-systems among a group of societies. Early and mid-twentieth century diffusionists and cultural geographers made strides toward measuring cultural variation and similarity, 104 although their methods rely more on etic, “outsider” approaches to data than cognitivists (of either stripe) can comfortably allow. Recently, researchers have developed cladistic approaches to cultural taxonomy in archaeology and ethnology. However, their phylogenies are problematic for a number of reasons. Memeticists (scientists who study units of cultural transmission) are developing theoretical models of the various ways in which ideas are transmitted. Although their approach is non-taxonomic, memetic theories about intellectual “lineages” and epidemic-like “thought contagion” provide interesting food for thought when considering how cognitive systems vary and develop. The discussion of unilinealism and its version of the comparative method is primarily drawn from Ralph Beals and Harry Hoijer (1958), Marvin Harris (1968) and Larry Naylor (1996). The outline of diffusionism and historical particularism is based on M. Harris (1968) and L. Naylor (1996), unless otherwise indicated. R. Mace and M. Pagel (1994) and M. O’Brien and R. Lyman (2003) are the basis for the description of phylogeneticism and cladistics in cultural studies, respectively. The subsection on cognitive taxonomy and historical reconstruction of mental categories is based solely on descriptions of individual works, which will be cited as they are discussed. The portrait of memetic approaches to continuity and change is synthesized from R. Aunger (2002), R. Blackmore (1999), J. Cartwright (2000), R. Dawkins (1976 and 1989), and A. Lynch (1996). Unilineal evolution and the “comparative method.” The earliest “modern” anthropological models of culture change and classification (developed by Sir James George Frazer (1922), Lewis Henry Morgan (1877), Sir E. B. Tylor (1871), and Herbert Spencer (1877), among others) were largely based on the principle of unilineal cultural evolution. Under the influence of Spencer’s concept of the “survival of the fittest” and Charles Darwin’s ideas about the retention of “primitive” characters in systems of descent with modification, these theoreticians extrapolated evolutionary logic into the cultural sphere and applied it to both living cultures and archaeological data. They had observed that contemporary cultures bear varying degrees of resemblance to extinct sociocultural systems. Coupling this insight with the then-current Victorian rationale that all cultures have developed progressively over time and their belief that cultures continue to progress teleologically toward some ultimate conclusion, these early anthropologists

105 were compelled to construct sometimes elaborate sequences of cultural stages of progress from simple to more complex forms of social organization and technology. The notion of the “doctrine of survivals” played a considerable role in these models. In 1871, Tylor defined “survivals” as follows:

[Survivals are] processes, customs, opinions, and so forth, which have been carried on by force of habit into a new state of society different from that in which they had their original home, and they thus remain as proofs and examples of an older condition of culture out of which a newer has been evolved (quoted in Tylor 1924:17).

The relic traces of earlier times “carried on by force of habit into a new state of society” provided an explanation for the troublingly anomalous “primitive” and “superstitious” features found in more “advanced” societies (like Victorian England and nineteenth century America). However, the unilineal evolutionists were not interested in processes of change, per se, so much as in where any given society could be placed on their evolutionary continua (i.e., how each group fit into the “family tree” of human cultural development). Their investigations were not meant to discover the history of a particular people or culture, but to provide grounds for generalizing about the evolution of culture “as such.” The unilineal evolutionists relied mainly on what they termed the “comparative method.” Their version of the comparative method was much like that applied in early traditional evolutionary biology. It involved (a) collecting examples of cultures and (b) deducing by inspection their relative level of evolution based on perceived complexity of their institutions. In what was, for the time, a major methodological advance, E. B. Tylor applied statistics to data gathered by the “comparative method” in an attempt to “tabulate and classify” cultures by the presence or absence of particular institutions (1889:245). He proposed that particular customs accompany one another and that the recurrence or absence of them is dependant on causes, i.e., the presence of other institutions prior to their development. This is an early development of an idea similar to the concept of ‘implicational ordering’ currently in vogue among some of the more influential variationists in modern linguistics—most notably Bailey and Mühlhäusler. Tyler’s statistics were a matter of simple correlational arithmetic: he estimated the probable closeness of causal connection by counting the number of times particular institutions (such as teknonymy, in-law avoidance, or 106 matrilocality) occurred in the same sample of cultures (approximately 350-400 in his 1889 paper). The product of the fractions x/350 · y/350 · z/350 (where x, y, and z represent different institutions and the numeral 350 represents the number of cultures surveyed for these features) was expected to be lower if the institutions were not causally related and higher if they were. We now know that the unilineal evolutionists were wrong about a great many things. Their theories were to a large degree ethnocentric (especially in their labeling of whole societies as “primitive” or “advanced”), and their work was often based on extremely inaccurate data gathered from unreliable sources. Their chronological arrangements frequently went far afield from the realities of actual historical developments. Most significantly for the present concern, their efforts produced little of worth for arranging cultures or societies into taxonomies showing relationships between cultures— they only ever showed relationships between cultural elements, but could not say anything about the history of a particular society in relation to others. Nevertheless, unilinear models provided two foundational theoretical premises for models to be developed later: those of regularity and continuity in change. In addition, the recognition that cultural features may correlate with one another in relations of prior and subsequent ordering (as do implicationally ordered linguistic features) raises an important consideration for the historical analysis of cognitive systems. Critically, for the development of anthropological theory, the assertions of the unilineal evolutionists precipitated the twin responses known as diffusionism and historical particularism, two developments linked together in a kind of “feedback loop,” which together provide a much more clearly relational taxonomic model for classifying historically related cultures. Diffusion and historical particularism. Although diffusionism is now treated as an “out of date” model in the majority of anthropology textbooks, this is more a consequence of the shifting tides of interest and paradigmatic emphasis than of any inherent flaw in the general idea of diffusion. As with any model, diffusionism can be (and has been) taken to ludicrous and sometimes dangerous extremes, but the general concept (as in linguistics) of the reality of the horizontal transmission of culture traits is sound. As long as it is understood at the outset that it is only a portion of culture traits that are so passed on, and not all culture traits (as in some earlier

107 extreme expressions of the model), the worst of the opprobrium that clings to the title “diffusionism” should be dispelled. Diffusionism is, in fact, strongly linked with the rise of the historical and particularistic perspectives of culture that are still central to current anthropological theory and method. Historical particularism provided a strong critique of unilineal evolutionism. Its focus on diversity and relativity remains an important contribution for all later formulations of culture change. At the same time, it also provided useful data for diffusionist models. Franz Boas, indisputably the most important figure in the development of the historical particularism movement, argued against unilineal evolutionism as follows (Boas 1988:89):

[Anthropological] research which compares similar cultural phenomena from various parts of the world, in order to discover the uniform history of their development, makes the assumption that the same ethnological phenomenon has elsewhere developed in the same manner. Here lies the flaw in the argument of the . . . method, for no such proof can be given.

Instead, Boas proposed an approach (only later termed “historical particularism”) which would look at a group’s customs in terms of their geographical distribution among neighboring groups, with the aim of determining the historical causes leading to the formation of those customs. This method was firmly entrenched in the model of diffusionism, which held that culture traits could be traced historically and geographically. To the diffusionists, cultures were accumulations of ideas, techniques, and artifacts, borrowed and spread (essentially in wave-like emanations, much like linguistic features in Wellentheorie) from one or more geographic centers. Diffusionary models center on kulturkreis (“Culture-Circles”) or “Culture Areas” (analogous to linguistic areas defined by bundles of isoglosses), which serve as heuristic devices for mapping and classifying societies. Two of the most outstanding early twentieth-century American proponents of this model were Clark Wissler and Alfred Kroeber. Wissler summarized the main import of the diffusion model in the following law: “[Cultural] traits tend to diffuse in all directions from their centers of origin” (cited in Harris 1968:376). In this model, the oldest culture traits are thought to be those most widely distributed around a center of diffusion, 108 what Kroeber called a “culture climax.” Following this logic, culture areas have frequently been defined in terms of comprehensive lists of cultural items. Diffusionism eventually gave birth to the field of cultural geography, which, like dialectometry in linguistics, does not always emphatically ascribe one or another origin to the phenomena it maps. It does not necessarily argue whether a trait has diffused through vertical transmission from one generation to another or through horizontal transmission via diffusion between neighboring groups—it simply maps the distribution of the features. One cultural geographer, D. W. Meinig (1969), has suggested that the morphology of a culture area may be viewed as consisting of three broad zones. C. Williams and J. Ambrose (1988) offer a refinement of his model. According to Williams and Ambrose, a culture area is made up of the following: (a) “the Core” (“the centralized zone of settlement where the most virulent characteristics of the culture . . . would be displayed” and where “the chief decision-making institutions and groups” are located); (b) the “Domain” (where “the original culture is still dominant, but . . . the bonds of connection are fewer and more tenuous, and where regional peculiarities are evident”), and (c) the “Sphere” (“the zone of outer influence, . . . where the culture is represented by only certain of its elements, or where its people reside as minorities . . .”) (Williams and Ambrose 1988:94, also see Meinig 1969 and Zelinsky 1973). A schematic of this core-periphery morphology (which may prove useful in discussing the relative ‘proximity’ of sources utilized in the present investigation of cognitive variation and similarity) is presented in figure 2.21. Diffusionists (and cultural geographers), like the unilineal evolutionists they so roundly criticize, claim to utilize “the comparative method.” However, unlike the unilinealists, they use these tools in tandem with statistical measures intended to deduce degree of similarity, and by it, to determine the “relatedness” of groups to one another. Harold Driver and A. L. Kroeber, whose 1932 essay “Quantitative Expression of Cultural Relationships” was a watershed development in the comparison of “culture wholes” (1932:211), provided one of the earliest and most clearly expressed examples of such statistical measurement. Driver and Kroeber’s work is based on formulas using three main values: c, the number of identical positive traits in the cultures to be compared

109

Figure 2.21. The morphology of a Culture Region (after D. Meinig). Reprinted from Williams and Ambrose 1988:95.

(i.e., those traits held in common), b, the total number of traits known for or possessed by one of two groups (labeled “tribe B” in the representational scheme they adopt for their analysis of the Plains Indian Sun Dance), and a, the total number of traits known for or possessed by the second of the two groups (“tribe A”). They developed several statistical measures using these variables. The simplest measure, based on earlier work by Wallis (1928) is to ascertain the proportion (P) of traits (c/a = Pa and c/b = Pb) of two or more groups, and then proceed to compare the proportions through a “piecemeal” method (1932:219), ranking the similarity of each group in a list (i.e., B, C, D) in relation to the first group listed (i.e., to A). Initially, Driver and Kroeber continued to follow Wallis in (a) using the correlation coefficient as a mean of two proportions and (b) applying the measures successively to tabulations of traits in a given area, as an attempt to discover the distances of their relationships (1932:220). However, they further developed the method by utilizing the arithmetical mean (A) of Pa and

Pb: (c/a + c/b)/2, as well as the geometrical mean (G) of Pa and Pb: c/√a·b (1932:256). Use of these means allowed them to overcome the difficulties of the piecemeal method developed by Wallis. These two measures permitted 110

Driver and Kroeber to plot relatedness of groups (in terms of similarity distances) both geographically and socially. This way of measuring the presence or absence of traits and comparing culture groups on their agglomerated similarity scores is essentially a method of phenetic classification almost identical in principle to those outlined above under biological and linguistic models of taxonomy. Although Driver and Kroeber only conceived of the similarities they measured as resulting from horizontal transmission (i.e., diffusion), in reality, the method captures traits possibly inherited through vertical means (not to mention, analogous traits) as well. Really the only new features are the formulas used (the arithmetic and geometric means), the kinds of characters compared (cultural traits), and the form of diagrammatic representation Driver and Kroeber utilized: maps of “principal influences.” Driver and Kroeber’s maps model “the leading intertribal influences indicated by the numerical values” obtained through their statistical method (1932:228, see figure 2.22 for their map of principal Sun Dance influences; also see their 1932 essay for analyses of Polynesian, southern Northwest Coast, and Northeast Peruvian cultural complexes). Groups are shown in their geographic positions and connected by arrows according to the “influences” captured by the arithmetic described above. The directionality of influence between groups is determined by “ethnological as much as [by] statistical considerations” (1932:228). Driver and Kroeber generally found that the maps so produced (and the history of influences so “discovered” or inferred) were in accord with the known historical and geographic facts of the phenomena in question (1932:255). Wisely, however, they conclude that these methods are only ancillary to non-statistical methods of examining culture history and that the computations “merely corroborate the ordinary ethnological findings made by students who know their field comparatively” (1932:255). The disallowance of independent invention is one of the major flaws of the diffusion model in its most extreme manifestations. A more serious difficulty related to this approach (even in the more moderate forms of diffusionism) is its proponents’ inability to identify systematically culture traits on the same level of detail, e.g., one social institution might count for one trait, while some particular item of material culture might count for four or five. Coupled with a reliance on etic categories rather than on emically defined ones, these difficulties seriously detract from the

111

Figure 2.22. Principal Sun Dance influences. Reprinted from Driver and Kroeber 1932:229. 112

usefulness of the traditional diffusionist approach. However, more positively, diffusionism calls attention to the possibility of intercultural borrowing and models one way (however imperfect) of quantifying similarity (and concomitantly, possible relationship) between cultures. Phylogeneticism and cladistics. Cultural anthropologists’ concern with evolution has ebbed and flowed. A first peak was reached during the nineteenth and early twentieth centuries (unilinealism and the diffusionary reaction against it). Another crest occurred in the mid-twentieth century (with the rise of neo-evolutionism, which—even more than unilinealism—does not really provide much “grist for the mill” for constructing taxonomies of relationship and is therefore excluded form the present discussion). Questions of history and relationship fell out of the limelight in the late 1950s and have only recently experienced a resurgence, with limited attempts to apply phylogenetic models and cladistic methods to ethnographic data (see Mace and Pagel 1994 for a seminal article) and archaeological assemblages (see O’Brien and Lyman 2003:93 for citations of major applications to date). The theory and method behind cultural phylogenetics is essentially identical to that of cladistics as applied to biological and linguistic forms. Like other cladists, cultural phylogeneticists (whether they are ethnographically or archaeologically oriented) assume that the cultures they study have evolved by descent with modification and that descendants of a common ancestor will have many characters in common. Behaviors and their results (including objects of material culture) are seen as phenotypic realizations of underlying genetic histories of socio-cultural selection and drift. Cultural phylogeneticists, like the linguists who have adopted the cladistic method, also make allowance for horizontal transmission of some traits, although (also in line with the theory of cladistically oriented linguists) they suggest that cultural phylogenies will still be predominantly vertical. The most significant unique feature of cultural phylogenetics is found in its methodology. Startlingly, its practice is often not based on culture at all. Almost universally, cultural phylogeneticists start with taxonomies (of individual characters, bundles of characters, and rarely, of whole cultures) generated through genetic, linguistic, or historical analyses.

113

They use these to “create the basic cultural phylogeny” (O’Brien and Lyman 2003:94). More often than not, cultural phylogenies are initially drawn up solely using linguistic data. Of the many possible “true” cladograms, the linguistic cladogram that best reflects the facts attested by ancillary genetic, cultural, or archaeological data is selected as the best representative. Afterwards, if parsimony is still an issue, horizontal transmission is postulated. Parsimony involves arriving at the least possible number of hypothetical changes by branching. If ‘too many’ branching changes have to be posited to explain the distribution of a character or character complex, then an explanation of horizontal transmission is invoked. Figure 2.23 is Mace and Pagel’s (1994) proposed phylogram of camel herding among Kenyan pastoralist cultures, which was arrived at in the above manner. It should also be noted that they recommend that such phylogenies are best based on the linguistic phylogenies proposed by Ruhlen, Greenberg, and other mass comparativists. While cultural phylogenetics seems like a noble pursuit, it has some problems. O’Brien and Lyman do a fair job of rebutting the main critiques commonly leveled against it. The first criticism they address is that artifacts do not carry phylogenetic information because they do not interbreed. However, their producers may. They then address the charge that any phylogenetic signal will be “swamped” in a “blur of interrelated forms,” because culture is reticulate (i.e., branches can merge). However, horizontal transmission does not—according to their report—result in 50/50 cultural “offspring”. Last, they address the concern that culture does not follow along in “lockstep” with language and biology, so cultures are not likely to possess a measurable genetic heritage. They answer that “If there is a high correspondence among the three variables, then perhaps the genetic [or linguistic] unit is an analytically useful device [for hypothesizing a cultural phylogeny]” (2003:97-115). Although O’Brien and Lyman have done well with deflecting the most common charges, several more serious flaws do not seem to have been adequately addressed. First and foremost, cultural phylogenies, when based on cladograms from biology or linguistics, suffer from the same deficiencies of any other cladistic phylogeny. They assume monophyly and sudden branching as the primary form and mode of inheritance, are largely circular in their

114

Figure 2.23. A phylogeny of nine Kenyan pastoralist cultures. Reprinted from Mace and Pagel 1994:553.

selection of “in” and “out” groups (which are decided upon a priori by largely phenetic means) and are ultimately based on subjectively selected homologous characters (which are not even cultural). Furthermore, if they are based on linguistic taxonomies generated by lexicostatistics or mass comparison (as Mace and Pagel advocate), they suffer from all of the defects of those linguistic models as well. Nevertheless, cultural phylogeneticists at least acknowledge that more than one kind of inheritance is possible and that relationships may be quantitatively assessed based on sharing of character states. Cognitive reconstruction and taxonomy so far. Cognitive anthropologists operating under the G1 paradigm have approached the problem of the history of cognitive systems from several interesting directions. Some have attempted to reconstruct proto-terms for categories and the underlying componential features and contrasts assumed to define them. Others have observed area-wide parallels in structures of contrast and hierarchical inclusion (Hopkins 1978 and 1980). Still others have made broad comparisons among classificatory systems that are “only partly hierarchical” (Bright and Bright 1965—a paper that also represents an important step in the direction of thinking in terms of G2 prototypes). A review of some of their key ideas, successes, and limitations will “set the stage” for a concluding 115 discussion of how the models of continuity, change, and taxonomy may best be used to investigate the cognitive phenomena analyzed in the G2 paradigm. The comparative method of historical linguistics has been applied in the G1 cognitive paradigm to reconstruct the shape of terms for categories of several domains for a number of proto-languages. In his 1970 book Proto- Indo-European Trees, Paul Friedrich reconstructed the phonology, morphology, and the semantic references of Proto-Indo-European (PIE) tree terms using the comparative method. His lexical reconstructions of tree terms followed standard procedures in historical linguistics. To access the semantic system, he abstracted the denotata of a proto-form from the tabulated correspondences in meaning of the reflex forms in the daughter languages (1970:2). Studies similar to Friederich’s have been undertaken with kinship terms, including Proto-Central-Algonquian (Hockett 1964), Proto-Central-Yuman (Frisch and Schutz 1967), and Proto-Otomanguean (Merrifield 1981), among others. These usually compare the features of terms in the modern reflexes (i.e., the vocabulary of a proto-language’s daughter languages) in an attempt to reconstruct the features for the reconstructed proto-term. However, Friederich went further than the others in positing taxonomic hypotheses based on his reconstructions as well (i.e., hypotheses about the Proto-Indo- Europeans’ taxonomic classification of trees). He argued that the PIE parent stock recognized at least eighteen named (super-ordinate or “life-form” level) major categories of trees that subsumed or included approximately twenty to thirty names for trees at the “generic” or “species” level (1970:1, 4). Friederich’s “taxonomic hypotheses” were more concerned with (or at least revealed more about) reconstructing contrasts than they were with identifying correspondences in multiple levels of taxonomic inclusion (1970:4, 6-8) or of deducing taxonomic relations between the cognitive systems of various PIE daughter stocks. Other researchers have made some interesting observations on such matters in different domains and cultures. Through a comparison of various “ethnoscientific” (i.e., G1) reports on systems of plant taxonomy for Nahuatl, Yucatec Maya, and Amuzgo (Chuj was also considered, though its data were not investigated ethnoscientifically), Nicholas A. Hopkins observed phenomena that can be interpreted as revealing a “common Mesoamerican classification system” (1978:2-3 and 1980:9). The

116 system he observed shows general agreement across all of the languages groups in question, in that they all place the majority of plants in taxonomies that agree substantially in their placement of specific botanical entities at the ‘life form,’ ‘generic,’ ‘specific,’ and ‘varietal’ levels of classification (1978:2-3 and 1980:9). Hopkins did not posit whether these common structural affinities resulted from a process of descent with modification (i.e., that they are genetic), or of diffusion, although his drawing the reader’s attention to the fact that the common system is shared by members of different language families (1978:2) may be a telling hint at presumed areal (diffusional or convergent) phenomena. Bright and Bright (1965) are more explicit in attributing a biotaxonomic classificatory system shared by speakers of different language families in Northwestern California to diffusion through areal convergence (analogous to Dixon’s convergence on a linguistic prototype, noted above). They found that the Yurok, Karok, and Smith River tribes shared some linguistic influences (predominantly in their grammatical systems) that seem to parallel, and possibly may condition, similarities in their biotaxonomies (1969:74-75). However, it appears that the biotaxonomies of the tribes are even more similar to one another than their languages are (1967:73). This seems to point to a situation of evolutionary convergence (less likely), areal diffusion (more likely), or both. The most interesting aspects of Bright and Bright’s study are that they are making statements about relationships of similarity between tribes’ classificatory systems and that at least two of the systems are not predominantly hierarchically-oriented, but instead class some of their objects as “like such-and-such” (1969:71). While they note similarities among the tribes, they only arrange them on a rather linear continuum of similarity rather than in terms of which two of the three are most like one another (1969:74). They come closest to positing a taxonomy of the tribes when they note that both Smith River and Karok exhibit a “poverty of taxonomy” when compared with Yurok (1969:75). Perhaps even more significantly though, their assertions about similarity between the tribes’ biotaxonomic systems incorporate the observation of phenomena very similar to the core-referent/peripheral-referent relationships central to G2 prototype studies.

117

Bright and Bright never quite managed to arrange the tribes into any kind of taxonomy. However, their acknowledgement that classificatory systems may be more or less related to one another and that these relationships may be exhibited by similarity in potentially overlapping (or fuzzy-bounded) G2- like relations between entities holds out promise for a more fully elaborated attempt at classifying taxonomic systems based on their overall global similarities. The work of Friederich and other “reconstructionists” confirms that at least some genetic aspects of vertical descent in taxonomic systems are able to be reconstructed through the comparative method—when it is applied to contrastive semantic feature analyses of denotative meaning. Still, however “language-like” (and therefore subject to the comparative method of phylogenetically-oriented linguistics) cognitive structures may be, horizontal transmission is at least as significant (if not more) a factor as vertical transmission in their development over time and across geographic and cultural space. Studies by Bright and Bright (1965) and Hopkins (1978 and 1980) show that both genetic and non-genetic forces may shape similarities between whole systems of classification. In their studies, we may, in fact, be confronted with the cognitive equivalents of the linguistic Sprachbund: areas of convergence in thought that are not necessarily completely isomorphic with the finer divisions of either linguistic or cultural areas. In addition, Bright and Bright’s work in particular suggests that similarities may be found in both denotative and connotative kinds of classification, i.e., that a prototype-oriented approach (which includes both kinds of meaning) may capture more information about interrelationships among groups. Memetic continuity and change. Memetics is not actually a taxonomic approach (nor have memeticists proposed any particular method for directly measuring its subject: memes), but its primary ideas may help us to consider more carefully the kinds of transmission that may affect cultural (especially cognitive) relationships of descent and borrowing. According to memetic theory, the sharing of inherited cultural character states may best be explained in terms of processes of ideational replication contingent upon interactions among carriers (“vehicles”) of various idea units (memes). Two main schools of memetic thought have generated theories based on genetic and epidemiological models of transmission. Several qualities favoring

118 transmission of memes have been posited, and memeticists have noted various environmental factors that may determine the primary kind of transmission that will be prevalent under certain societal conditions. Oxford zoologist Richard Dawkins coined the term ‘meme’ to refer to a kind of replicable idea unit, analagous with the gene concept of biology (1976). A meme is “an idea or thought that occupies brains in the same way as a gene occupies bodies” and is able to replicate in other bodies like a parasite going from host to host (Cartwright 2000:89, 313). As proposed by Dawkins, memes (abbreviated from the Greek mimeme) are units of cultural transmission or imitation such as “tunes, catch phrases, clothes fashions, ways of making pots or building arches” (Dawkins 1989:192) as well as “visual images ... facial or hand gestures, [or various] skills” (Dawkins 1999:109). The concept of idea replication through imitation, “broadly defined”, is central to memetics. As gene-like ‘replicators’, memes are propagated through replication. Replication is simply a relationship between a source and its copy exhibiting the characteristics of (a) causation, (b) similarity, (c) information transfer, and (d) duplication. Causation means that the source is causally involved in the copy’s production. Similarity simply means that the copy is like the source in relevant aspects. Information transfer is the process that generates a copy similar to the source from that same source. Duplication is the process by which one entity becomes more than one. All replicators (including memes) are sustained by an “evolutionary algorithm” which Susan Blackmore says is based on “variation, selection, and retention (or heredity)” (1999:14). In the case of memes, variation arises because ideas are rarely communicated in exactly the same way twice; selection is determined by how well a meme “grabs attention” and how easily it can be remembered, and retention simply implies that something of the original meme has been copied to another host, or “vehicle.” The vehicle is an entity in interaction with an environment that, by its behavior, provides circumstances allowing copies of the replicator to be made. Even though both genes and memes are replicators, most memeticists have taken pains to point out that they do not necessarily behave or work in the same way in all respects (Blackmore 1999:18).

119

Memes use a special kind of replication which differentiates them from all other potential replicators (and genes in particular): they transmit culture through imitation. Since memetic replication is based on imitation and not solely on direct transfer (as it is with genes), copying is not always exact. Imitation in Dawkins’ “broad sense” means, for example, that in passing on a story, it is not necessary to imitate “every action and word,” but only that the gist of the story is copied to someone else (Blackmore 1999:6-7). Therefore, two modes of memetic replication exist, which Blackmore has termed “copy-the-product” and “copy the instructions.” “Copying-the-instructions” is analogous to Darwinian genetic transmission. In biology, the genes are the instructions to be copied, which are then manifested in an outward phenotype. For memes, copying-the- instructions arises in situations where literacy is prevalent and instructions or recipes are written down and circulated. The instructions are then the “genotype” and the behavior or product the “phenotype.” “Copying-the-product” involves the transmission of memes from brain to behavior to brain as a result of observation and imitation of behavior. In this sense, memetic transmission is potentially more Lamarckian, in that imitators will pick up phenotypic variation and pass on these acquired characteristics to the next observer/imitator. Both processes are utilized by memes. However, in the world of memeticists, only the physical manifestations of a memetic trait—its outward physical manifestation, or phenotype (sometimes called the ‘memotype’)—can be measured. The existence of evidence for both copying-the-product and copying-the- instructions processes leads memeticists to the use of two major schemas (see Shore 1996) for describing memetic transmission: the genetic and the epidemiological. The genetic or “meme-as-gene” perspective that Dawkins has popularized sees memes as replicators endowed with their own “evolutionary interests”, like the “selfish” genes that act only to get themselves replicated into the next generation. Memes-as-genes may be spread vertically in lineages from parent to child, accompanying genes from one generation to the next. Copying-the-product, however, has led the epidemiological “meme- as-germ” school to make use of a contagion metaphor for better understanding memetic transmission. Aunger explains as follows (2002:17-18):

Memes are the equivalent of a cold virus that, by causing sufferers to sneeze . . . succeeds in infecting everyone in the 120

vicinity. So memetics is the cultural analogue to the study of how disease-causing pathogens diffuse through populations. The striking metaphor of memes as ‘mind viruses’ . . . takes memes as particles of culture that parasitize human hosts, causing them to behave in ways conducive to getting copies of their information into the heads of other people . . . . [Both] memes and viruses undergo vigorous competition for survival. Viruses must overcome the immune system and induce the host to transmit new virus particles to uninfected hosts; memes must overcome those memes previously existing in the host’s mind and induce her to transmit the meme to new potential hosts. (2002:17-18)

Dawkins observes that memes are usually passed on in altered form either through the development of copying errors or as a result of interaction with other memes. Ideas are not repeated in the same words. “[We] twist them round for . . . [our] own purposes, changing the emphasis, blending them with ideas of . . . [our] own and of other people” (Dawkins 1989:195). Because of these two phenomena, Dawkins argues that memetic transmission is “subject to continuous mutation” (like genes), and also to “blending” (1989:195). There is, however, no clear-cut preponderance of mutation over blending, or vice versa. Neither has there been substantial evidence produced that shows that either the genetic or epidemiological “contagion” metaphor is most accurate overall. Both modes, however, are expressed phenotypically and may only be measured through observation of cultural behavior and its products. The transmission and evolution of memes is driven by selection of and competition between memes in the “meme-pool.” Those memes with the highest “survival value” will exhibit the qualities of “longevity, fecundity, and copying-fidelity” (Dawkins 1989:192). Another key factor in memetic competition is that of compatibility. Since “[memetic] selection favors memes that exploit their cultural environment to their own advantage” and the “cultural environment” largely consists of other memes also engaged in a struggle for survival (and thereby, selection), compatibility with the already existing “corpus of ideas” prevalent in a particular society plays a major role in a meme’s success (Dawkins 1989:110-111, 199). While memes may be successful because they are “useful” to their hosts, it is often just as critical for their survival and propagation that they be memorable. In fact, several avenues for getting themselves copied are available to memes, including (but not limited to) the following: (a) “their benefits to human happiness or to human genes,” (b) “they appear to provide advantages even

121 when they do not,” (c) “they are especially easily imitated by human brains,” or (d) “they change the selective environment [i.e., the “meme-pool”] to the detriment of competing memes” (Blackmore 1999:27). The principle of memetic selection dictates that people will imitate the people who are best at imitating others. This necessarily entails horizontal transmission, since a society cannot persist for long solely on the interaction of parents with their own children alone. Under some social circumstances, horizontal transmission becomes even more prevalent. In small-scale “hunter-gatherer” or horticultural societies, “life changes slowly, [memetic] transmission is largely vertical, and a meme is most likely to succeed if it benefits (or at least appears to benefit) the . . . success of its carrier” (Blackmore 1999:133). However, in ‘larger-scale’ societies, memetic success is more often achieved by the meme which “can get quickly from host to host, and never mind how well each host does in terms of its own survival or its reproductive success—as long as there are more hosts around to infect” (Blackmore 1999:133). The development of city life and of various modes of transport, as well as the establishment of interlinking roads and waterways, speeds up the process of competition among memes by carrying vehicles (people, writings, goods) to distant places and thereby creates a more varied meme-pool as it increases the number of people in contact with one another. Just as higher population densities and improved transportation favor the “rampant . . . spread of biological contagions, so too do they favor the spread of thought contagions” (Lynch 1996:25). When people live in or come to a city, or move (or dwell) along major routes of human interaction, they “have more communication partners from whom to catch communicable ideas, and more potential retransmitting contacts as well” (Lynch 1996:25). However, there is an imbalance in directionality of transmission. People on the periphery have fewer contacts, unless they travel to an urban center where they will encounter many new memes. When the denizens of cities travel, however, they will meet fewer people in rural areas than rural people will meet in the cities, and they will also encounter fewer new memes to pick up. While memes may spread piecemeal, they can often replicate better as part of a larger group, or “memeplex,” than they can individually. Such a “co-adapted stable set of mutually-assisting memes” may be treated as something of a “macro-meme” consisting of an evolving set of participating

122 memes (Dawkins 1989:197). Thus, large groups of memes may be transmitted together in a tightly bound group, but mutations or invaders worming their way in may gradually change its composition. Dawkins gives the example of an organized church with “architecture, rituals, laws, music, art, and written tradition” (1989:197). New memes and meme-combinations may come and go under the aegis of the overarching macro-meme. It is likely (if we can hazard to speak in terms of “memes” at all) that cognitive systems—like ‘Islamic Medicine’—are just this sort of memeplex.

IMPLICATIONS OF PRIOR METHOD AND THEORY FOR THE PRESENT RESEARCH

In the previous pages we observed that biologists, linguists and cultural anthropologists have found the results of phenetic approaches to their subjects to be good indicators of where to begin looking for historical relationships. As with languages and other components of cultures, cognitive structures may be shaped by processes of punctuation and equilibration, and their morphology may result from various kinds of “mixing” among multiple parental memeplexes. Cognitive systems include both denotative and connotative elements, and frequent contiguity of memetic units likely affects judgments regarding their similarity. Dialectometry in linguistics and numerical taxonomy in biological and cultural studies provide key models for how similarity of cognitive structures defined in the second-generation paradigm might be measured. The following chapter will describe the methodology for cognitive dialectometry developed in reaction to these observations.

123

CHAPTER 3

METHODS: MEASURING RELATIONSHIPS BETWEEN COGNITIVE DIALECTS

This chapter presents the application of the principles of numerical taxonomy to patterns of drug-plant prescription in Islamic societies and their antecedents. It includes an overview of (a) the research design of the project, (b) the derivation of specific hypotheses and their associated variables from the problem under consideration, (c) the sources and sampling procedures used to obtain it, (d) the data collection and analysis procedures followed, and (e) the limitations that affect the study. Together, these serve as the structural framework for obtaining the results that will be presented in chapter 4.

Design Parameters of the Research Project

The research design is correlational in nature, utilizing the methods of G2 cognitive anthropology and numerical taxonomy. It includes the use of hierarchical clustering, multidimensional scaling, K-means clustering, and factor analysis techniques. The purpose of the design was to (a) measure the interrelationships between patterns of drug-plant prescription in documents hypothesized to represent different cognitive dialects of medical practice in the modern and pre-modern Middle East, and (b) to correlate these inter- relational similarities with the known facts of geography and history. The research design was based on an “interactive model”. The different design components are viewed holistically as interconnected and interacting in a flexible nonlinear fashion. This allows for the continuous assessment of the implications of the elements for one another (Maxwell 1996:3-5). The primary

124

Purposes Conceptual Context

Research Questions

Methods Legitimations

Figure 3.1. The interactive model of research design used in this study. After Maxwell 1996:5.

elements of the design are seen as logically influencing or implicating one another in an interactive network of two-way ties, as shown in figure 3.1. Correlational research design is one of several alternatives to traditional experimental research designs. Like quasi-experimental and other ex post facto designs, correlational designs are useful in situations where variables cannot be controlled because (a) the events under investigation have already occurred in the past or inherently cannot be manipulated, and (b) it is infeasible for subjects to be randomly assigned to treatment groups (Kerlinger 1979:116-117, and Rudestam and Newton 1992:25). The present study necessitates such a non-experimental design because (a) it is intended to address historical questions with primary sources, some of which are of substantial antiquity, and (b) it investigates the natural history of a group of evolving systems for which sources of data are severely limited. Correlational designs differ from quasi-experimental ones in that they do not include a treatment, or there are only minimal grounds for identifying one variable as the treatment. In addition, their measurements use continuously

125 scaled variables rather than discretely scaled “categorical” values (Mark and Reichardt 2004:79). This project uses continuously scaled variables to represent the degree of overall similarity between sources and does not involve experimental treatments so much as the application of observational techniques. It represents, in essence, a discovery procedure, rather than an experimental one.

Purposes

The purposes of the study helped to guide and justify the other design choices made in its construction and realization. They represent elaborations on both personal and public goals. The researcher’s personal interests in and experiences with ethnobotany and the Middle East were the initial impetus for their development. The practical need to develop methods for addressing historical questions in the G2 cognitive paradigm was an additional motivating factor. The widespread lack of any real understanding of the interrelationships underlying Middle Eastern medical systems also necessitated further investigation. The project was undertaken with three primary purposes in mind:

1. To provide a more elaborate portrayal of the diversity of the system(s) of Islamic medicine and its/their antecedents in the region than has been heretofore presented;

2. To provide ancillary means for supporting and illustrating previously posited historical “hunches” regarding inter- and intracultural intellectual influences and relationships among Islamic societies and their forebears, and

3. To introduce a more overtly historical element into G2 cognitive research through the use of a theoretical and methodological approach that may serve as a foundation for the future elaboration of G2 applicable reconstructive techniques.

Conceptual Context

The conceptual context of the research design entails the theoretical framework of the study and was (in a sense) already elaborated on at length in the previous two chapters. It entails theories of cognition and the evolution of cognitive systems and the history and nature of Islamic medicine and the culture area where it is most prevalent.

126 In this study, cognitive categories are seen from a G2 perspective (i.e., they are prototype-centered and are delimited by fuzzy boundaries). One of the best ways to access the actual cognitive structure of such categories (like drug-plants) is through an investigation of their distributional similarities. Character state correlations among external markers of similarity (phenetic analyses) are the most conservative option for categorizing cognitive systems. Distributional similarity is based on largely phenotypic data (behavioral patterns). In addition, the meme-as-gene versus meme-as-germ controversy has yet to produce methods for distinguishing between vertically inherited (i.e., cognate or symplesiomorphic) and horizontally diffused (i.e., borrowed or synaphomorphic) memeplex components. Cognitive structures are conceived of herein as evolving systems with both genetic and areal features. They are of potentially polyphyletic (“mixed” or multiple) parentage and may share continuing mutual influences with other “cognitive dialects.” Cognitive dialects are likely to be subject to creolization processes similar to those undergone by linguistic lects. In addition, cognitive systems possess spatially and socially distributed traits. They should therefore reflect (to a significant degree) the primary features of the subjectively defined culture-area morphology posited by historians and anthropologists for the same societies. The “ecumene” (see Canfield 1991:xiii) of which Islamic medicine is a part may be thought of as a culture area shaped by specific geographic and historical facts. Like other culture areas, it can be described in terms of a morphological structure divisible into Core, Domain, and Sphere regions. However, there is a disconcerting lack of consensus among scholars on the full extent of this culture area’s boundaries and on its subdivisions. The following synthesis is derived from Bates and Rassam 2001, Canfield 1991, Eickelman 2002, Held 2006, and Spencer 2000, although it does not agree with any of them in all respects. For the purposes of this project, the Core minimally includes the Arabian Peninsula, the Levant, Mesopotamia, and the territory of the modern nation-state of Egypt. However, some regions of southern Arabia are geographically isolated and have long histories of limited contact with other areas in the Core. Such areas, traditionally ascribed to the “core Middle East” but only peripherally linked to its heartland, will be considered to have membership in the “distal Core” or “proximal Domain.” The Domain of the culture area includes North Africa (the Maghrib) and the plateau area of Turkey and Iran. However, much of North Africa should be considered more

127 proximal than Turkey or Iran, as the North African region is characterized by a stronger and more persistent “Arabic” ethnic and linguistic identity, traits central to identity formation in the Core area. Conversely, the plateau areas of Turkey and Iran will be considered to be “peripheral/distal Domain” or “proximal Sphere” members. The culture area also includes a “double periphery” zone of convergent traditions located between Iran and the sphere of Inner Asian civilization. Primary sources for studying practices of drug-plant prescription in the region come from multiple time periods and different streams of tradition. However, they may be divided into three main categories: (1) modern traditional, (2) medieval traditions thought to be at least minimally Galenically-inspired or influenced, and (3) non-Hellenic. Non-Hellenic sources may or may not have had a significant influence on Galenic practices or local traditions. Modern sources may or may not derive from the medieval “high tradition” of Arab medicine and may or may not reflect substratal influences from pre-Arabic, non-Galenic local systems.

Research Questions

The research questions form the heart of the design used in this study. They are intimately tied to each of the four other components of the interactive model diagrammed in Figure 3.1 (above) and thus mediate and are mediated by all of them. The questions developed out of the theory outlined in the conceptual context section (and chapters 1 and 2), and their answers are expected to a) contribute to theory and method in cognitive anthropology and numerical taxonomy and b) to deepen our understanding of the heritage of Islamic medicine and its forebears. In addition, the questions parallel the purposes of the project, detailed above. The three queries that follow, taken together, are the core component of this research design:

1. What do the cognitive structures of Islamic medical systems and their antecedents “look like” when taken as individual wholes? 2. What do the relationships between twentieth-century Islamic and antecedent medical systems “look like,” based on objective numerical measures? 3. Through which instruments and measures can numerical taxonomy/ dialectometry best be applied to second-generation cognitive categories in a historical perspective?

128 Methods

The basic methodology of the research design involved two groups of sources. Group I consisted of modern sources only. Group II consisted of pre-modern sources thought to be either Greek-influenced or non-Hellenic, as well as modern sources from the Core of the culture area (included for comparative purposes). For each source, distributional similarities of drug- plants were modeled using the MAX or “Complete” method of hierarchical clustering, based on the Pearson coefficient (or on Euclidian distance, where the Pearson coefficient was inapplicable due to the use of negative numbers to define “cold” and “moist” humoral properties). The resultant cluster trees were compared within each group. For Group I, hierarchical clustering, multidimensional scaling, K-means clustering and factor analysis techniques were applied to the arithmetic mean, geometric mean, and the arithmetic and geometric means of the usability function (UA/G) of shared clusterings of drug- plants. For Group II, hierarchical clustering, multidimensional scaling, K- means clustering and factor analysis techniques were applied only to the arithmetic and geometric means of the usability function (UA/G) of shared clusterings. Figure 3.2 shows the basic relationships among the groups, techniques (i.e., instruments), and measures underlying the methodology. Sampling, data collection, and data analysis procedures will be treated at greater depth later in the present chapter.

Group I Modern Sources Only Measures: Arithmetic Mean Geometric Mean UsabilityA/G

Multidimensional Hierarchical K-Means Factor Analysis Scaling Clustering Clustering

Measures: UsabilityA/G

Group II Modern and Pre-Modern Sources

Figure 3.2. Basic relationships among groups, techniques, and measures.

129 Legitimations (Validity, Reliability, and “Truth Value”)

Classic descriptions of the validity and reliability of experimental research design are based on the validity of the measurements used. In turn, the internal validity of the measurements (and concomitantly, the external validity or reliability) is based on control of all variables affecting the dependent variable. Since ex-post facto research designs are used in situations where all variables cannot be controlled or manipulated, traditional notions of experimental validity do not apply to them. Rather than dealing with questions of causation (as in experimental designs), ex- post facto designs are primarily correlational. Therefore, instead of possessing “validity” in the traditional sense, they are subject to the more relative tests of legitimation, or “truth value,” that are applicable to qualitative research designs (see Newman and Benz 1998:27 ff.). A research design is legitimate if it adequately deals with threats to the “truth” of its description, interpretation, and theory (Maxwell 1996:89 ff.). In historical document-based research, invalidity in description results from the use of secondhand data. Illegitimacy in description is reduced through the use of original sources (Newman and Benz 1998:72). Since this project utilized primary sources in written form, it comes closer to legitimate description than non-verbatim (e.g., participant observation based) descriptions of local medical practices. Threats to legitimate interpretation mainly derive from the imposition of an “outsider” etic framework. Since this design was founded on a thoroughly emic, practice-based conception of drug-plant categories (defined by their distributional similarity), it has much greater interpretive legitimacy than traditional diffusionist studies of etic cultural categories. Theoretical legitimacy is threatened by lack of attention to discrepant data or alternative explanations. In the present study, the main perceived threat to theoretical legitimacy was the possibility that ecological factors (realized as shared presences or absences of medicinal plants between sources) and/or the physiological (phytochemical/pharmacological) properties of plants might affect similarity judgments between sources. While these possibilities cannot be completely excluded, they are mitigated by two important facts. First, shared plant presences most likely result from geographic proximity and/or shared history (e.g., neighbors in the same ecological zone or members in the same trading network will share many of the same plants). Second, recognition of medicinal qualities is cultural, and

130 therefore it is constrained by the same factors of transmission as the larger cognitive system of prescription. For these reasons, it is important to remember that this is a phenetic approach and not a phylogenetic one. No claims can be made in regard to the exclusion of similarities resulting from convergent evolution. With this in mind, the research design of this project follows Driver and Kroeber in presuming that, insofar as the results “fit” the known geographical and historical facts of the situation, they possess validity, reliability, and significance (1932:217).

Derivation of Specific Hypotheses and Associated Variables

Two general research hypotheses were presented in the introductory chapter of the dissertation. Each of these general hypotheses has its corollary in a specific research hypothesis with operationalized variables. The hypotheses and their variables interrelate with the research problem, the purposes of the study, the research questions asked, and the contextual context of the project. The review of theoretical and methodological literature in chapter 2 leads one to the conclusion that the methods of numerical taxonomy (i.e., dialectometry) are relatively well suited to an exploratory consideration of the problem of developing quantitative mechanisms for making inferences about the history of cognitive systems in the G2 paradigm. The overview of the case study itself (the ethnography and history of medicine in Islamic societies), in collusion with the methodological review, suggests that the key variables relevant to the research questions at hand can be seen on two levels. At the most basic or foundational level, the variables are simply “drug-plants” and “illnesses” (or “drug-plant properties”). Once these variables have been used to model distributional similarities for each source, comparison across sources yields a continuously measured derived dependent variable (DV) on a higher level of analysis, termed “degree of similarity.” Since the theoretical overview suggests that similarity in cognitive structure should correlate to facts of geography and history, “proximity” and “shared history” (as defined in the Introduction) are relevant independent variables (IV). At present, these cannot be easily measured directly and independently, rather, they are realized in the form of an intervening independent variable, “culture area morphology,” which is

131 itself subjectively evaluated. Relative strength of “antecedent-successor relationship(s)” is another key variable, which assumedly shapes “shared history” and surely correlates with the “degree of similarity” between related sources as well. Therefore, the following general and specific research hypotheses were generated: General Research Hypothesis 1. A numerical taxonomy approach to Islamic medicine will show that there is a relationship between (a) the proximity and shared history of the localities where two or more sources (i.e., collections of prescriptions for drug-plants, or descriptions of drug-plant properties) were produced and (b) the degree of similarity between overall use-based similarity judgments of drug-plants by those sources. Specific Research Hypothesis 1. Sources in the Core of the culture area will be shown, through numerical taxonomy, to be more similar to one another than to sources on the periphery. Peripherally located sources will be shown to be more similar to core sources and immediate neighbor sources than they are to non-neighbor peripheral sources. Exceptions will generally conform to known influences not subsumed by the culture area model (e.g., the influence of trading relationships, or membership in a “double periphery zone”). General Research Hypothesis 2. A Numerical taxonomy approach to Islamic and pre-Islamic Near Eastern medicine will show that there is a relationship between (a) the relative strength of antecedent-successor relationships of sources, and (b) the degree of similarity between the overall use-based similarity judgments of drug-plants by those sources. Specific Research Hypothesis 2. Where there are sources from three or more successive periods of an intellectual tradition, numerical taxonomy will show that the earliest and latest are more similar to the middle sources than they are to one another. The hypotheses regarding the status and associations among the variables are represented in the path diagram in figure 3.3. As in traditional path diagrams, boxes indicate variables for which there are scores. Ellipses indicate inferred variables for which there are no scores. Curved lines indicate correlations, and, in contradistinction to traditional usage, straight lines indicate assumed lines of causality. Dotted lines represent relationships of secondary strength, and solid lines, those of primary strength.

132 c1

Proximity (IV)

a3 e1 a1 b1

Shared a4 Culture Area Degree of Similarity History Morphology (DV) (IV) (Intervening Variable)

a2

Antecedent- c2 Successor Relationship (IV)

c3

Figure 3.3. Path Diagram of Hypotheses.

Paths a1 and a2 show the presumed “constructing” influence of proximity and antecedent-successor relationships on shared history. Paths a3 and a4 describe the subjective basis for positing culture area morphology. Path b1 represents the main thrust of General Research Hypothesis 1, and path c3, that of General Research Hypothesis 2. The circle labeled e1 captures variability in the degree of similarity (DV) not explained by paths b1 and c3 (see the discussion of “legitimations” in the previous section of this chapter). Paths c1 and c2 represent alternative influences which fall outside the culture area model. For example, they address problems of similarity resulting from membership in long-distance trade networks not describable in terms of Core, Domain, and Periphery.

Sources and Sampling Procedures

The sample is made up of two groups of texts, totaling fourteen sources (written collections of multi-drug-plant prescriptions). Group I consists of seven “modern” sources, six of which are from the Middle East plus one from a neighboring “out-group”—all of which were composed between 1977 and 1990.

133 Group II consists of nine sources, seven of which are “pre-modern.” The earliest pre-modern source dates from 1534 B.C. and the latest from 1208 A.D. One of these is a modern (1988) summary/distillation of a much longer and less accessible medieval text. Group II also includes two sources which are duplicated in Group I. The two modern sources in Group II represent locales in the proximal Core of the culture area and are retained for comparative purposes (see Specific Research Hypothesis II, above, for the rationalization for their retention). The data set was composed of a total of 176 drug- plants, 1,177 multiple drug-plant prescriptions (selected from an original corpus numbering over 2,600), and forty-one drug-plant properties. The selection procedures applied to drug-plants and prescriptions within each source will be described in the description of data collection and analysis procedures which follows the present sub-section. This project utilized a non-randomized, non-probabilistic approach to sampling known as “purposeful” or “criterion-based” sampling (see Weiss 1994:17 ff. and Maxwell 1996:70). Criterion-based sampling entails the selection of subjects (in this case, sources) that provide unique information that cannot be obtained as effectively from other sources. Purposeful samples thus stand midway between convenience samples (whatever the researcher can get hold of) and probabilistic samples (where each member of the population has a known probability of being selected). Unlike convenience samples, purposeful samples include only those subjects that can best provide the information needed for answering the specific research questions under consideration. These usually represent panels of knowledgeable experts (or “expert” sources) that are uniquely informative and significantly instructive. The heart of the sample (all of the members in Group I except for the “out-group” representative) was derived from collections published in the Studia Culturae Islamicae series of the Institute for the Study of Languages and Cultures of Asia and Africa. These sources were selected because they contained individual sets of prescriptions composed by named herbalists resident in specified cities of the culture area under study. Other potential sources that were excluded from Group I of the sample were prepared by academicians who had synthesized the practices of multiple informants and thus summarized the traditional medicinal uses of plants for large regions (e.g., Boulos 1983, Mossa 1987, and Ghazanfar 1994). Such summaries gloss over the nuances of individual variation and the grading of practices between locales, thus obscuring the differences that form the basis for a comparative

134 study. It should be noted, however, that the “out-group” source for Group I was a summary of a region located outside the culture area under consideration, chosen for its convenience. It was unnecessary for the purposes of this research that the out group be a uniquely authored (rather than a synthesized) source. The use of the Studia Culturae Islamicae series also gave the Group I segment of the sample a unity of presentation and information content that would have been more difficult to achieve using texts collected from diverse sources. The authors of the series presented both native language and English or French translations of four of the six sets of prescriptions, along with Linnean identifications of the botanical materials used in all six of the sets. These sources were therefore both uniquely informative and significantly instructive where others might not have been. The sampling procedure employed for Group II was slightly more opportunistic than that employed for Group I. This was necessitated by the fact that the archaeological and epigraphic records are limited to those documents that have survived up to the present. It is further constrained by what has been published and the form of publication. Some ancient sources were excluded because they did not constitute a large enough unified text. For example, Assyrian medicine is known through a large corpus of cuneiform tablets (Thompson 1923). However, no individual tablet contains enough multi-drug-plant prescriptions to be comparable in number or variety to those obtained for Group I or to the other pre-Modern sources ultimately selected for Group II. Rather than synthesize multiple (possibly cognitively conflicting) tablets composed by multiple authors into a “master” text, Assyrian texts were ultimately excluded from the sample. Other ancient texts (such as the hieroglyphic papyri P. Berlin and P. Hearst from Egypt) were excluded because specialists contend that much of their content (which is not presently available in adequate translation) is replicated in a larger unified text, the P. Ebers (which is also readily available in German translation with hieroglyphic plant names retained). Only one Coptic medical manuscript of substantial length survives (the P. Chassinat); other texts are preserved only on leaves of parchment or on ostraca. The same can be said for Syriac. As a result of their unique representation of extinct cognitive systems, the selection of these two manuscripts, P. Chassinat and the Syriac “Book of Medicines”, was guaranteed. Medieval Arabic sources were more abundant than either ancient or modern sources. However, not all of them were equally informative or

135 instructive. One of the sources (the medical formulary of Al-Kindi) was selected because it was composed in Baghdad during the initial flowering of Islamic medicine and thus represents the “root” of the intellectual “stock” from which the family tree of later Islamic medicine derives. While other sources from this locale and period were available, this particular one was chosen for its clarity and consistency of structure and the scholarly attention that has been paid to it (Levey 1966). Another source, the lengthy summary of ibn Wafid given in 'Abd al-'Aziz al-Sulamī’s thirteenth-century study manual, was selected because it was the clearest and most succinct presentation of traditional humoral qualities of drug-plants that the researcher was able to locate. Galen’s De Simplicibum and Book IV of Avicenna’s Qanūn fi al-Tibb, though the classic references for all later Islamic humoral medicine, are inconsistent in their presentation of humoral qualities: not all drug-plants are described in a comparable manner. A modern “handbook” of the medicine of Avicenna that summarizes and standardizes his attribution of humoral qualities to 117 drug-plants was used in their stead.

Data Collection and Analysis: Techniques and Measures

The data collection and analysis procedures are so intimately connected with one another that it is difficult to recognize where one ends and the other begins. This is a consequence of the fact that the dependent variable “degree of similarity” is a derived variable based on a set of still deeper derived variables: the sharing of characters between sources. The sharing of characters between sources is derived from the clusters of drug-plants (yet deeper derived variables) resulting from each individual source’s placement of each drug-plant in one of three to four clusters. These clusters are based on the distributional similarity of each source. Since each derived variable is the result of the analysis of deeper (or lower) level variables, it is simplest to describe the data collection and analysis procedures as a linear sequence of events. Both Group I (modern sources, N=7) and Group II (pre-modern and modern Core sources, N=9) were treated in a manner that followed the same general outline, diagrammed in figure 3.4.

136 1. Selection of Sources • Application of selection criteria • Identification of plant 2. Treatment of Individual terms Sources • Selection of recipes and plants • Derivation of pile sorts from hierarchical clustering of distributional similarity

3. Comparison of Sources • Discovery of characters • Coding and measurement of characters • Calculation and clustering of phenons 4. Evaluation of Resulting Phenons • Holistic overview • Particularistic overview

Figure 3.4. Diagram of procedures for data collection and analysis.

Selection of Sources and Identification of Plants

For both Group I and Group II sets, sources were selected using a criterion-referenced strategy (see the previous subsection for details of the sample and the sampling procedures employed). For each plant name found in the selected sources, every effort was made either (a) to ascertain the Linnean identity of the plant so termed, or (b) to obtain an identification of each plant (or plant product) name with a term occurring in one or more other texts where a Linnean identification had already been. This involved frequent reference to the indices and glossaries of translations of the sources (where available), dictionaries and “handbooks” with lists of multilingual synonyms (e.g., Crum 1950, Deines and Grapow 1959, Boulos 1983, Manniche 1989, and Ghazanfar 1994), and classical and medieval works including Dioscorides’ De Materia Medica and Moses Maimonides’ glossary of drug names. Once either of the two identification requirements was met (and preferably both), each native language plant name was given a Linnean taxon gloss.

137 Treatment of Individual Sources

After all possible plant names were identified with Linnean taxa or matched with their corresponding terms in the other sources and given a gloss, recipes and plants were selected for the data set to be analyzed. Plant drugs to be considered in the analyses were narrowed to those which were shared with one or more other sources. Plants were considered “shared,” if their glosses corresponded at either the generic or species level (see the Assumptions subsection in the introductory chapter). Only recipes with two or more drug-plant constituents were admitted to the data set. All recorded drug-plant properties were admitted to the data set. For Galenic-based sources, drug-plant properties included two variables: hot-cold and moist-dry continua rated in a range of -4 (for cold or moist in the fourth degree) to +4 (for hot or dry in the fourth degree). The Indian (out group) source used thirty-seven attributes that are elaborations on the four basic categories of Ayurvedic drug-plant properties: rasa ‘taste,’ guna ‘quality,’ veerya ‘potency,’ and vipak ‘metabolic property.’ For prescription-based (rather than property-based) sources, the plants (more properly, their Linnean glosses) and prescriptions were recorded in an item-attribute matrix showing which plants occur in which prescriptions. For property-based sources, the plants (i.e., their glosses) and their qualities were arranged in a similar matrix. Distributional similarities were modeled for each source by applying the complete linkage (also known as “maximum,” “MAX,” “or farthest neighbor”) method (technically, an amalgamation rule) of hierarchical clustering to the data matrices constructed in the above-described manner, using the SYSTAT 10 statistical analysis computer package. Complete linkage clustering was used (rather than other amalgamation possibilities like “single linkage clustering”) because it tends to produce compact, discrete, globular clusters that join other clusters only with difficulty (Sneath and Sokal 1973:222 and Wilkinson et al. 1996:611). Such a method encourages the clustering of a minimum number of large groups containing maximal amounts of variables. It was an ideal choice for this project, because the intent was to derive hierarchical clustering trees that would, in their last few “joins,” parallel traditional G1-style constrained pile sorts, while still being based on distributional similarities. The Pearson correlation coefficient was used as the basis for measuring the distributional similarity-based association between plants for all sources except the humoral property-based ones (which used negative numbers

138 to code degrees of “cold” and “moisture”). It was selected because the coefficient correlates similar patterns even when variables have different means and standard deviations (Wilkinson 1996:722). Normalized Euclidian distances were used for the sources explicitly using humoral qualities, since (when standardized) they represent an inverse transformation of Pearson correlations. The use of Euclidian distance was necessitated by the inability of the Pearson coefficient (i.e., the Pearson -1) to accommodate negative numbers. Hierarchical clustering involves the sequential joining of similar objects and the joining of resultant groups of similar objects into higher and higher levels of grouping. The last two joins of a cluster tree (the last three or four groups to be amalgamated into the tree) are therefore analogous to the last four piles made in a “bottom-up” pile sort. Each of the last three or four groups to join together were treated as a “pile” in subsequent analyses. Four groups were preferred. However, when one of the last four groups had fewer than four members, it was decided that the smallest cluster should be “lumped” with its nearest neighbor, and the source should be treated as having “decided upon” three rather than four “piles.” This approach emulates the freedom permitted informants undertaking constrained pile sorts where they are asked to keep putting items and groups of items together until they arrive at either three or four final groups (cf. Weller and Romney 1988).

Comparison of Sources

A modified version of Agar’s sort table technique (1996) was applied to the “pile sorts” derived from each individual source’s hierarchical clustering tree. Procedures differed for Group I (modern sources) and Group II (pre-modern and modern core sources). Group I procedure. For Group I (N=7), “pile sort” results were compared first in pairs, then in all permutations of groups of three, four, five, six and seven to identify the implicationally scaled sharing of characters among sources. Thus, given sources a, b, c, d, e, f, and g, sort tables were prepared comparing: {a, b}; {a, c}; {a, d}; {a, e}; {a, f}; {a, g}; {b, c}; {b, d}; {b, e}; {b, f}; {b, g}; {c, d}; {c, e}; {c, f}; {c, g}; {d, e}; {d, f}; {d, g}; {e, f}; {e, g}; {f, g}; {a, b, c}; {a, b, d}; {a, b, e}; {a, b, f}; {a, b, g}; {b, c, d}; {b, c, e}; {b, c, f}; {b, c, g}; {c, d, e}; {c, d, f}; {c, d, g}; {d, e, f}; {d, e, g}; {e, f, g}; {a, b, c, d}; {a,

139 b, c, e}; {a, b, c, f}; {a, b, c, g}; {b, c, d, e}; {b, c, d, f}; {b, c, d, g}; {c, d, e, f}; {c, d, e, g}; {d, e, f, g}; {a, b, c, d, e}; {a, b, c, d, f}; {a, b, c, d, g}; {b, c, d, e, f}; {b, c, d, e, g}; {c, d, e, f, g}; {a, b, c, d, e, f}; {a, b, c, d, e, g}; {b, c, d, e, f, g}; and {a, b, c, d, e, f, g}. “Characters” were defined as any shared grouping of two or more drug- plants. Some characters were only shared by a pair or triad of sources. Others were shared by larger groupings of sources. Following Bailey (1996) and Mühlhäusler (1985), those characters/clusters shared by sources in a larger sort table (i.e., one with more members) which are subsumed by larger clusters in smaller sort tables (with fewer members), are considered to be “implied” by the larger character/cluster. Thus, given the hypothetical characters A {Aloe, Boswellia, Commiphora, Myrrh} (shared by sources a and b) and B {Aloe, Commiphora} (shared by sources a, b and c), character B is implied by character A. After all implicationally ordered characters were identified, the Operational Taxonomic Units (the sources) and characters (implicationally ordered clusterings of drug-plants) were arranged in an OTU-Character data matrix. Characters were then coded by “string length.” For example, a four- drug-plant character/cluster shared by a pair of sources would be given a string length score of 4 representing the four plants included in the cluster. A three-plant character/cluster shared by a pair of sources would be ascribed a string length score of 3 representing the three plants in the shared cluster. Implicating characters (those that implicate or subsume less complex strings) were only given additive values representing the number of drug-plants added beyond the string length of the smaller (implied/subsumed) cluster. Thus, given a hypothetical character/cluster, A {Aloe, Commiphora}, shared by sources a, b, and c, and another hypothetical character, B {Aloe, Boswellia, Commiphora}, shared by sources a and b, character A would be given a string length score of 2 and character B would be given a string length score of 1 representing the addition of Boswellia to the already existing string (A). Where it was unclear which more basic character was implied by a more complex character, the relationship that required the lowest additive value was selected. Where a complex character could be viewed as stemming from either a) the addition of multiple individual drug-plants to a simpler character or b) the addition of an already existing character (cluster/string of drug-plants) or characters to that character, the addition of pre-existing

140 characters to one another was selected over the addition of a series of additional individual drug-plants. Each added character/string added a string length score of 1 to the resulting new (implicating) character. Imagine a hypothetical scenario where we are presented with the characters A {Aloe, Commiphora, Foeniculum, Pimpinella, Rosa} (shared by sources a and b), B {Aloe, Foeniculum, Pimpinella} (shared by sources a, b and c), C {Foeniculum, Pimpinella, Rosa} (shared by sources a, b, and c), and D {Aloe, Commiphora} (shared by sources a, b, and c). Given this scenario and following the above rule, character A would be seen as the addition of characters C and D to character B and given a string length score of 2. This decision would be preferred, over a solution that sees character A as simply character B plus Commiphora and Rosa. (Incidentally but insignificantly, in this hypothetical case the rejected decision would also have resulted in a string length of two.) The arithmetic mean (A) and geometric mean (G) were calculated for each pair of sources in Group I. The sum of all string length values possessed by the first source of a pair and the sum of all string length values possessed by the second source are represented by a and b, respectively, in the formulas A = (c/a + c/b)/2 and G = c/√(a · b). The sum of the string length values shared by both members of a pair is labeled c, common or shared characters. The usability function (U) was based on the results of paired comparisons only. Thus, in the formulas UA = (c/a + c/b)/2 and

UG = c/√(a · b), a represents the total number of plants the first source shares with any other source in Group I, and b, the number of plants the second source shares. The total number of plants for which the two sources are in agreement (those which they both group together) is again represented by c. Initially, only hierarchical clustering and multidimensional scaling (MDS) were used to model overall degree of similarity between sources (the dependent variable in figure 3.3). Hierarchical clustering trees (using complete linkage clustering and the Pearson coefficient) and MDS plots were generated based on the A, G, UA and UG similarity scores. For further confirmation of the patterns displayed in the resulting outputs, factor loading plots were generated for the same scores. In such a plot, sources that are nearly identical in their factor loadings (i.e., the covariances of the original variables with their components) will overlap. The more similar the factor loadings, the closer the sources will appear when plotted. In

141 addition, K-means clustering (see Wilkinson et al. 1996:614) was used to further confirm the grouping of sources inferred from the other graphical methods. K-means clustering is a method that uses an algorithm to search for the “best” separation into k groups among all objects in the set. It attempts to place each object (in this case, source) into clusters where the Euclidian distance of each object to its cluster center is as small as possible. K-means analyses were performed on the A, G, and U scores with k set at 2, 3, 4, and 5. Group II procedures. For Group II (N=9), “pile sort” results were compared in pairs only. The exclusion of triads, quads, etc., was precipitated by two factors. First, with the addition of each source, the number of comparisons necessary for a full treatment increases exponentially. With a greater number of comparisons, the probability of observer error increases concomitantly. The fifty-six sort table comparisons for Group I required over two weeks of dedicated effort, with multiple cross-checks and corrections taking an additional week. Until automated methods are devised for applying the technique, it is impractical to calculate sort table-based A and G scores for groups of sources larger than 6. Second, the Group I results for the U factor (which utilized only paired sorts, rather than all permutations) were not found to differ substantially from the results based on the full battery of possible comparisons. Consequently, only the arithmetic and geometric means of the U factor were calculated for Group II.

Once the UA and UG similarity scores were calculated for Group II, the results were modeled in the same manner as described for Group I (with the sole exception that, for k-means clustering, k was set at 2-8).

Evaluation of Resulting Phenons

Evaluation of the groups modeled on the A, G, and U similarity scores was both holistic and particularistic. The holistic evaluation entailed a subjective comparison of the hierarchical clustering, multidimensional scaling (MDS), factor analysis (using Varimax rotation), and K-means outputs to the culture area morphology of Group I and to the generally agreed-upon antecedent-successor relationships of Group II. The holistic evaluation was essentially a diagnosis of the realism of the quantified schemes of relationship viewed as a whole. Particularistic evaluations were centered on source-specific interrelations. They focused on correlating the most significant influences for each source to the known or presumed principal

142 influences (vertical or horizontal) upon them (based on a comparison of A, G and U scores).

Limitations of the Study

Although the data set includes over 1,000 recipes and the method is based on a synthesis of previously legitimated approaches, they still present some limitations. The overall scarcity of sources was one limiting factor. Another was a paucity of data (usable prescriptions and plants) in two modern sources (those from Gaziantep, Turkey, and from Karachi, Pakistan). Neither had more than twenty prescriptions that met the dual requirement of including more than one drug-plant that could be identified with drug-plants in other sources. In addition, they had the two lowest numbers of drug-plants meeting the selection criterion outlined above. Nevertheless, because they represent unique positions in the culture area morphology, they were indispensable for answering the research questions of this study. A second limitation is the indeterminacy of the number of “piles” ascribed to each source. The decision to “lump” a final cluster of three or fewer drug-plants in with its nearest neighbor was a somewhat arbitrary one. However, it was based on accepted practice in G1 cognitive studies and still reflects emic judgments of distributional similarity. A third limitation could be considered an issue of relativity. The results of this study are based on multiple “givens.” Given this set of plants and these particular illnesses for which prescriptions are needed, this particular herbalist will produce that pattern of distributional similarity. Nevertheless, no matter what the size or shape of the sampling universe, in the real world we can never be privy to all possible “bits” of information. Finally, one error in the construction of a hierarchical-clustering tree, modeling the distributional similarity of a source in Group I, was noted after analysis had been completed. The error involved the inclusion of one additional plant for the Indian source (Aristolochia), which was not shared with any other modern source. Correcting the error would have involved performing thirty-one of the fifty-six sort table comparisons for Group I over again. Since this source shared more drug-plants with other sources than any other source in its group, it is assumed that the addition of one irrelevant plant should not precipitate any significant changes for the final result.

143

Summary

This chapter has presented a methodology for measuring relationships between cognitive lects. It is based on a correlational design that synthesizes methods from G2 cognitive anthropology and linguistic dialectometry/phenetics. The General Research Hypotheses posited in the introduction of the dissertation were further elaborated and refined into specific hypotheses about how a numerical taxonomy approach should perform when applied to collections of drug-plant prescriptions (or properties) from Islamic societies and those historically affiliated with them. A sample of fourteen sources dating from 1534 B.C. to the present and entailing one hundred seventy-six plants, 1,777 prescriptions, and forty-one drug-plant properties was used to examine the legitimacy of these hypotheses. Procedures for the selection of sources, their individual treatment and comparison, and the evaluation of phenons resulting from their comparison were explained in detail. Finally, the limitations of the study were addressed and the general soundness of the methodology was affirmed.

144 CHAPTER 4

RESULTS: A NUMERICAL TAXONOMY OF ISLAMIC MEDICAL SYSTEMS AND THEIR ANTECEDANTS

This chapter presents (a) a description of the sample used in this study and (b) the results of testing the general and specific research hypotheses put forward in chapter 3. The description of the sample provides a more detailed picture of the individual sources used in each group. The results section shows the outcome of an application of the principles of a numerical taxonomy approach to the sources under consideration.

Description of the Sample

The sample was divided into two groups; one of modern collections of traditional Middle Eastern (plus India as an out group) prescriptions (or descriptions of medicinal properties) only, the other of pre-modern sources plus two modern sources from the Core of the culture area (for comparative purposes). The descriptions of each source include a brief overview of their geographic and temporal provenience; the total number of recipes or drug plant properties and drug plants present in the source and the number of drug plants and prescriptions or properties which met the selection criteria.

Group I, Modern Sources Only

In this dissertation, the sources in Group I are denoted by the name of the center (city or town) and modern nation state where the data were collected. Since the Ayurvedic Indian text (representing the “out group”) is not a primary source but a secondary one reporting drug-plant properties prevalent in a broad cultural area beyond the sphere/periphery of the Middle Eastern culture area, it is simply referred to herein as “India.” Language

145 prevalence and literacy rates are included for most of the countries from which the sources were derived. Statistics contemporaneous with the collection dates of all texts were not always available. Therefore, percentages given below may not be entirely accurate. However, the statistics and language information were gathered from reports as close to the collection dates as possible. The demographic and political data is synthesized from The World Factbook (1990, 2005), Country Studies/Area Handbooks (1988-1998) and The Ethnologue (2005). Aleppo, Syria. This set of prescriptions was obtained sometime around 1988 or 1989 by the academic research team for the “Comparative Study on the Islamic Societies and their Cultural Changes” of the Institute for the Study of the Languages and Cultures of Asia and Africa (ILCAA). The wasfāt (‘recipes’ or ‘prescriptions’) were provided in Arabic by Mohammad Faez Bawadiqji, an herbalist in Aleppo. Honda et al. (1990) do not provide a translation. Bawadiqji provided seventy-two recipes/prescriptions utilizing seventy-eight plants. Of these, fifty-nine recipes and sixty plants met the selection criteria outlined in the methodology chapter and were used for the analysis of relationships among the members of Group I. Aleppo is a major cultural and commercial center in northern Syria. In 1990, Syria had a literacy rate of 49 percent. Arabic is the official language of Syria, although Armenian, Aramaic, Circassian, and French are also present. Cairo, Egypt. This set of prescriptions was collected between October 1977 and March 1978 by the academic research team for the “Comparative Study on the Islamic Societies and their Cultural Changes” of the ILCAA. The wasfāt were obtained in Arabic from a Cairene herbalist named Surur Muhammad 'abd al-Hadi. Ahmed et al. (1979) provide an English translation. 'Abd al-Hadi’s text includes one hundred thirteen recipes using approximately one hundred fifty-three plants. Sixty-six recipes and fifty-four plants from this source were used in the final data analysis for Group I. Cairo is the capital of Egypt and a major urban center. In 1990, the country of Egypt had a 45 percent literacy rate, with Arabic as the primary language and English and French as important secondary languages. At present, literacy is at approximately 57.7 percent. It is likely that the percentage given for 1990 was significantly lower twelve years earlier when 'abd al-Hadi’s data were collected. Gaziantep, Turkey. This set of prescriptions was collected between 1983 and 1984 by the academic research team for the “Comparative Study on the Islamic Societies and their Cultural Changes” of the ILCAA. The recipes were

146 obtained in Turkish from Ilhan Arslanyürek, an aktar (“herbalist”) in the province of Gaziantep. Başer et al. (1986) include an English translation. Arslanyürek provided three hundred thirty-eight recipes using a total of approximately two hundred twelve plants. Of these, only twenty recipes and forty-six plants met the selection criteria outlined in the previous chapter. Gaziantep is an important industrial and commercial center located in southeastern Anatolia. In 1990, Turkey had 70 percent literacy, with Turkish as the official language and additional literacy in Kurdish and Arabic. At present, literacy is between 70-90 percent. India. This source is a reference book that was produced by L.D. Kapoor, Ph.D., a retired research scientist from the National Botanical Research Institute, Lucknow, and was initially published in 1989. Kapoor has undertaken research throughout India on various aspects of Ayurvedic medicinal plants and their uses. The text describes some two hundred fifty plants and their pharmacological and therapeutic uses in Ayurveda (traditional Hindu medicine, based on the doctrine of “signatures”). In addition, it provides the plants’ Arabic, Bengali, English, German, Hindi, Nepali, Persian, Sanskrit, Tibetan, and Unani (Greco-Arabic) names. Of the two hundred fifty-one plants included in Kapoor’s text, seventy were selected for the current project based on their being shared with other, non-out group sources (also see the “Limitations” section in the previous chapter). Ayurvedic pharmacology recognizes four basic categories of properties by which materia medica may be classified. For every plant, each of these categories is represented by one of several characters. By category, the number of attributes which Kapoor notes, and that were relevant to the drug plants selected for this project were rasa (6), guna (20), veerya (8), and vipak (3). Karachi, Pakistan. This set of prescriptions was collected between 1983 and 1984 by the academic research team for the “Comparative Study on the Islamic Societies and their Cultural Changes” of the ILCAA. The recipes were obtained in Urdu from Hakim Mohammad Said, who, in his youth, studied tibb (Islamic medicine) in Delhi, India, before relocating to Pakistan sometime around 1948. Ushmangani et al. (1986) provide an English translation. The recipes are from a technical manual composed by Said’s uncle that has been circulated widely throughout the subcontinent. The text gives approximately forty-one recipes, using a total of about four hundred forty-five plants. Of these, however, only eighteen recipes and forty plants met the selection criteria outlined in the previous chapter. Karachi is a major port city in

147 southern Pakistan, a Muslim nation that separated from British India in 1947. Literacy was at 26 percent in 1990. In 2005, it stands at approximately 45.7 percent. Urdu and English are the official and majority languages of Pakistan. In 1990, 27 percent of the population also spoke Balochi, Pashtu, Punjabi, Sindhi, or other minority languages. Marrakech, Morocco. This set of prescriptions was collected in October of 1980 by the academic research team for the “Comparative Study on the Islamic Societies and their Cultural Changes” of the ILCAA. The recipes were obtained in Arabic from Rahhal ben el-Hajj Mohammad, an herbalist resident in Marrakech. Bellakhdar et al. (1982) provide both English and French translations. The Arabic text gives thirty-two recipes utilizing a total of three hundred ten plants. Of these, fifty-two plants and twenty-seven recipes met the selection criteria and were used in the present analysis. In 1990, Morocco had a literacy rate of 28 percent. In 2005, the literacy rate was 51.7 percent. Sanaa, North Yemen. This set of prescriptions was obtained sometime around 1988 or 1989 by the academic research team for the “Comparative Study on the Islamic Societies and their Cultural Changes” of the ILCAA. The wasfāt were provided in Arabic by Ahmed al-Nashiri, the proprietor of the Nashiri Drug Store in Sanaa. Sanaa was the capital of Northern Yemen, which joined with Southern Yemen and became the Republic of Yemen in 1990 (retaining Sanaa as the new capital of the unified state). Honda et al. (1990) do not provide a translation from the Arabic. The text includes fifty-five recipes utilizing a total of approximately fifty-two plants. Of these, forty-two plants and forty-eight recipes were used in the current data analysis. In 1990, literacy in Yemen was estimated at approximately 15 percent. By 2005, it had reached 50.2 percent.

Group II, Pre-Modern and Modern Core Sources

The pre-modern sources are generally referred to by (a) their author, where known; (b) their traditional designation (e.g., “P. Ebers” for the ancient Egyptian Ebers medical papyrus), or (c) their language (e.g., Syriac). Since the Syriac text was divided into two sources, one Greek-based and the other native Syriac, it is described in two separate subsections under the headings Syriac α and Syriac β, respectively. The two modern Core sources (Syria and Egypt) were described under the Group I heading above. However, they both shared different drug plants in common with other Group II sources that were not shared with other sources in Group I. Fifty-three

148 plants from the modern Egyptian source and 58 from the modern Syrian source met the selection requirements for Group II. al-Kindi. This source is an Aqrābādhīn (registry of prescriptions, from Greek γραφιδιον) by Abū Yūsuf Ya’qūb ibn Ishāq al-Kindi (ca. 800-870 A.D.) containing both simple and compound remedies. Al-Kindi is thought to have been born in Kufa (in what is present-day Iraq) and to have studied Greek philosophy, mathematics, and medicine at Basra and Baghdad during the early Abbasid period. Levey (1966) provides an Arabic facsimile and English translation of this manuscript from the Aya Sofia library (number 3603, fols. 91b-139a). Al-Kindi’s materia medica includes approximately three hundred nineteen items, the vast majority of which are plants and plant-derived products. The text consists of two hundred twenty-six formulas. A total of one hundred fifteen drug plants and one hundred thirty-seven recipes were selected for further analysis based on the criteria given in the previous chapter.

ibn Wafid. This source is drawn from chapter V of Muwaffaq al-Dīn ‘Abd al-‘Azīz ibn ‘Abd al Jabbār ibn Abī Muhammad al-Sulamī al-Dimashqī’s Expert’s Examination for All Physicians. al-Sulamī (ca. 1155-1208 A.D.) practiced medicine in the bīmāristān (hosptital) of Nūr al-Dīn in Damascus. His fifth chapter (on simple drugs) is intended as a test of aspiring physicians’ knowledge of the humoral qualities that ibn Wafid al-Lakhmi ‘Abd al-Karim of Toledo (d. 1075 A.D.) ascribes to some one hundred fifty-five drug plants. al-Sulamī asks which plants ibn Wafid classifies as “hot and dry in the third degree,” “moist and cool in the second degree,” etc., and then proceeds to provide the answers for each possible permutation of degree and quality.

Leiser and al-Khaledi (al-Sulamī et al. 2004) provide an English translation and footnotes with the Arabic names given in the original manuscript. Of the one hundred fifty-five plants, one hundred two were selected for this analysis. Pseudo-Avicenna. This source was excerpted from a modern English “handbook” of Avicenna’s medicine, written by a hakim (i.e., a traditional healer), G.M. Chisti, who studied Greco-Arabic medicine in Afghanistan, Pakistan, and India. His work is closely based on the book Mizan al-Tibb (The Standard of Medicine), an eighteenth-century commentary on a slightly earlier Persian abbreviation of Avicenna’s (980-1037 A.D.) extremely lengthy Arabic classic, the Qanūn (“Canon of Medicine”). Chisti provides descriptions of one hundred seventeen drug plants. He gives their Latin and

149 Arabic names as well as their Avicennan humoral qualities. Sixty-nine of the plants so described were used in the current analysis. Syriac α. This set of prescriptions is based on the first section of the twelfth-century A.D. Syriac “Book of Medicines” acquired by the British Museum at Mosul in the early years of the first decade of the twentieth century. It was translated into English and published in 1913 by E. A. W. Budge (Keeper of the Egyptian and Assyrian Antiquities of the British Museum). Many of the recipes in this section are attributed to Dioscorides, Galen, and other Greek physicians of note, and they seem to have been translated from Greek into Syriac by a Syrian (probably Nestorian) physician, possibly at Edessa, during the first three or four centuries A.D. This text, therefore, represents the closest we can come to an exemplar of the proximate source of Greek influence on Islamic medicine. It gives us a window on the Syriac perception and practice of Greek medicine through which the Hellenic tradition ultimately was passed on into Arabic. Data for the present research were collected using Budge’s English translation of the Book of Medicines, as it is recognized as fairly literal and accurate and is the only translation of the text yet produced. Although the translator has been the subject of considerable derision in Egyptological circles, Middle Easternists and medical historians have found little fault with his work on this particular text. Most criticisms of Budge stem from his tenacity in advocating his own orthographic transliteration of the Egyptian hieroglyphic script, which others rejected in favor of a different system. Budge states that there are nearly one thousand recipes given in this section of the manuscript, although the researcher only counted approximately seven hundred that included multiple identifiable drug plants. Due to the large number of prescriptions, most of which were far lengthier (on average) than those found in any other source considered, a presumably representational sample of one hundred forty-six recipes was selected using a random numbers table. The author of the manuscript identifies around one hundred seventy plants and plant products in a list of “medicines mentioned in the book” (including those drugs used in the prescriptions grouped herein under the rubric Syriac β). One hundred twelve plants in the one hundred forty-six recipes used met the selection criteria outlined in the “Methods” chapter. Syriac β. The prescriptions making up the source referred to as “Syriac

β” are derived from the third part of Budge’s English translation of the

150 Syriac Book of Medicines (see Syriac α for details of the manuscript’s provenience). The third part of the Syriac manuscript, from which these recipes are drawn, “The Book of Native Medicines,” consists of a record of the local practices of traditional Mesopotamian folk healers, presumably collected by the same physician who composed the first section (Syriac α) of the manuscript. It consists of approximately four hundred prescriptions, one hundred three of which met the selection criteria and were used in the data analysis. Seventy-two drug plants used in these recipes met the selection criteria outlined in the previous chapter. P. Chassinat. The Chassinat Coptic Medical Papyrus is the lengthiest preserved text of its kind (Chassinat 1921). It consists of a collection of 237 prescriptions dated by paleographic means to sometime between the ninth and tenth centuries A.D. Approximately ninety plants can be equated to Linnean generic taxa with a fair level of certainty. The papyrus utilizes over a dozen Arabic plant names, indicating at least a minimal degree of diffusion of Arabic terminology, if not of cognitive categories. It uses almost twice as many Greek terms as Arabic. One hundred forty-four of the two hundred thirty-seven recipes and sixty of the eighty-nine identifiable plants met the selection criteria for, and were included in, the analysis. P. Ebers. This source is the longest hieroglyphic medical text (one hundred ten pages) known from ancient Egypt and is dated (based on internal evidence) to ca. 1534 B.C. While its original provenience is uncertain, it is likely that the papyrus came from the tomb of a physician in the Theban necropolis on the west bank of the Nile. An English interpretation of other- language translations (French and German) was produced by Bryan in 1930 and a direct English translation was published by Ebbel in 1937, but the extant English versions are notoriously unreliable (Nunn 1996). The present research utilizes the current “definitive” German translation of von Westendorf (1999), which preserves the hieroglyphic names for materia medica. The text includes eight hundred seventy-seven recipes, three hundred seventy- two of which met the selection criteria for this project. Approximately one hundred forty-two plants are used in the papyrus, although only thirty-nine could be identified with relative certainty (based on authoritative scholarship) as equivalent to plants occurring in other sources of Group II.

151 Results

The results of the study are presented in three topical divisions corresponding to the three research questions listed in the previous chapter. The first addresses the shape or appearance of Middle Eastern systems (ancient, medieval, and modern) of drug plant prescription taken as individual wholes. The second addresses the interrelationships among sources, implied by the results of a numerical taxonomy of Groups I and II, i.e., the relative similarity evaluations necessary for confirming or disconfirming the hypotheses under consideration. The third division is a brief evaluation of the efficacy and accuracy of the instruments and measures used in this study.

The Cognitive Structures of Islamic Medical Systems and Their Antecedents

The cognitive “shapes” of Middle Eastern traditions of drug-plant prescription have hitherto been obscured. This state of affairs results from our methodological inability to view the overall structure of these patterns apart from either generalized impressionistic portraits of ethnomedical systems or simplistic reporting of specific (and typically only the most unusual or exceptional) cures. Consequently, descriptions of Islamic ethnomedicine have historically taken a perspective that does not allow for any kind of quantification of similarity across societies. The figures presented in this subsection of the chapter provide a quantifiable mid-range perspective through the visual representation of the distributional similarities of drug plants for the sources considered in this study. They provided the basis for the cross-society comparisons made in reference to Research Hypotheses I and II. Since, for the purposes of this project, no particular hypotheses were made regarding the shape of their expected individual structures (only regarding their interrelationships determined through comparison across sources), the immediately following analysis will be limited to a simple graphic presentation of their overall forms. It is hoped that these portraits may spur botanists, pharmacologists and ethnobotanists to investigate the chemical and cultural properties of drug plants that the sources have clustered together. The goal is to determine the specific motivations (both denotative and connotative) underlying their similar use patterns in a given source. The drug plants appearing in the figures are labeled with their Linnean taxon gloss. For purposes of coding and data entry with SYSTAT 10, some

152 glosses needed to be abbreviated. Thus, for example, “ALLIUM1” is equivalent to Allium cepa, “ALLIUM2” is equivalent to Allium sativa, “ALLIUM3” is equivalent to Allium porum, “CASSIA1” is equivalent to Cassia acutifolia, “CASSIA2” is equivalent to Cassia acutifolia, etc. In addition, multiple sources consistently present more than one native language name for the same Linnean taxa. When multiple names for the same Linnean taxon are nearly identical across sources, they are coded as two separate entities, indicated by ancillary numbers after the Linnean identification. Thus, for Anethum graveolens, we have both “ANETHUMGRAV1” and “ANETHUMGRAV2,” glossing knds and krfs, respectively, in the Egyptian and Moroccan sources. In addition, any gloss ending with the designation -“COMBO” represents an agglomerated taxon where two or three native language names, often unique to a source, are applied to the same Linnean taxa, but are not consistent across sources. Occasionally, details of coding vary slightly (but not meaningfully) from figure to figure. Thus, Cassia acutifolia may be coded as “CASSIA1” in one figure and as “CASSIAACUTIF” in another. However, variations in representation are incidental to our concerns here and the Linnean genus designations we are most concerned with are easily recognizable in all of the figures. As discussed in the previous chapter, the last two “joins” on a cluster tree (in the following figures, the last two nodes to the right of a given tree, excluding the furthest node to the right that joins all member items into that tree) divide the items into three to four groups, or “piles.” The “piles” arrived at in this manner are indicated by roman numerals to the left of each figure. Each “pile” is separated from its nearest neighbor(s) by a thin horizontal line. Thus, in Figure 4.1 (representing Aleppo, Syria), we may observe that Cassia acutifolia, Foeniculum vulgare, Pimpinella anisum and Rosa sp. occur in the “pile” marked with roman numeral I. In Figure 4.2 (representing the distributional similarity for Cairo, Egypt), we may observe the same plants grouping together in the lower part of the “pile” labeled with a roman numeral III. By agreeing on grouping these four plants together into one of three to four piles, Aleppo and Cairo show a quantifiable modicum of agreement in their overall classification of drug plants. Group I, Modern Sources Only. The first seven figures that follow (4.1-4.7) present the distributional similarity of all drug plants meeting the selection criteria of chapter 3 for each source in Group I. It should again be noted that extending the analysis of a source to more than a handful

153 PIMPINELLA FOENICULUM CASSIAACUTIF ROSA TILIA JUNIPERUS I ZIZYPHUS ERYTHREA MATRICARIA TRIGONELLA CORIANDRUM ALLIUMCEPA ALPINIA ZINGIBER MYRISTICA1 ANACYCLUS PIPERN EUGENIA CINNAMOMUMC ELETTARIA MYRISTICA2 IRIS CINNAMOMUMZ PEGANUM NIGELLA LEPIDIUM PIPERCHABA ASTRAGALUS PISTACIAL NARDOSTACHYS PUNICAGRANAT ARECA QUERCUS II TERMINALIA1 COMMIPHORA PISTACIAT PINUS AMYGDALUS DAUCUS RAPHANUS AMMI CARUMCOPT ANETHUMGRAV1 CITRUSLIMON ASAFOETIDA GLYCYRHIZA CUMINUM PRUNUSM ARACHIS PAPAVER ANETHUMGRAV2 ALOE III CITRULLUS LUPINUSTERM RHEUM TERMINALIA2 LINUM IV LAVENDULA CROCUS BOSWELLIA

0.0 0.5Distances 1.0 1.5 Figure 4.1. Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions from Aleppo, Syria.

154 PEGANUMHARMA TRIGONELLAFO RAPHANUSSATI ERYTHRAEASAT CORIANDRUMSA CYPERUSLONGU ARECA I ORIGANUM LAVENDULA NARDOSTACHYS IRIS CINNAMOMUMZ LAWSONIAINER CARUMCOPTICU PUNICAGRANAT QUERCUSINFEC TERMINALIA2 RHEUMRAPONT TERMINALIA1 II MATRICARIA MENTHA CITRUSLIMON CUMINUM ZINGIBEROFFI ALPINIAGALAN EUGENIACAROP CINNAMOMUMC ANACYCLUS ELETTARIA ANETHUMGRAV1 MYRISTICA2 III MYRISTICA1 INULAHELENUM ZINGIBERZERA PISTACIATERE ANETHUMGRAV2 PIMPINELLAAN ROSAALBA FOENICULUMVU CASSIAACUTIF CYMBOPOGONCI AMMIVISNAGA GLYCYRHIZAGL TILIASYLVEST LINUMUSITATI PRUNUSMAHALE IV CINCHONASUCC PISTACIALENT COMMIPHORAMY ALOE BOSWELLIACAR NIGELLASATIV ACACIAARABIC ASTRAGALUS

0.0 0.5Distances 1.0 1.5 Figure 4.2. Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions from Cairo, Egypt.

155 CINCHONA TARAXACUM MENTHA FOENICULUM PLANTAGO CITRULLUS I CUCUMISMELO RAPHANUS ROSMARINUS ROSACOMBO ALTHEA MYRTUS SALVIACOMBO JUNIPERCOMBO AMYGDALUS VITIS TILIA IRIS DAUCUS ANETHUMGRAV1 ACHILLEA II CANNABIS INULA PRUNUSMAHALE VIOLA ZIZYPHUS PINUS ASTRAGALUS LAWSONIA SINAPSIS QUERCUSINFEC CINNAMOMUM III PISTACIAVERA CITRUSLIMON COFFEA MYRISTICA1 ELETTARIA ORIGANUM MATRICARIA PIMPINELLA FUMARIA IV PUNICA ARTEMIS1 MELISSA TAMARINDUS ZINGIBER

0.0 0.5Distances 1.0 1.5 Figure 4.3. Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions from Gaziantep, Turkey.

156 EMBLICA ELETTARIA CUCUMIS BAMBUSA TERMINALIA PUNICA CORIANDRUM FOENICULUM DAUCUS WITHANIA TRIGONELLA VITIS VIOLA PIPER COMMIPHORA VALERIANA I CROCUS SALVADORA ABROMA URGINEA CASSIAACUTIF ACORUS ASAFOETIDA MYRISTICA ZINGIBER ALLIUMCEPA ALLIUMSATIVA RAPHANUS ALPINIA AETHGRAV1 ANACYCLUS JUNIPERUS TAMARINDUS RHEUM RUTA NARDOSTACHYS MELIA BOSWELLIA SANTALUM FUMARIA BERBERIS OCIMUM CYPERUS PEGANUM DELPHINIUM CANNABIS CITRULLUS II CINNAMOMUM NIGELLA CARUMCOPTICU ARISTOLOCHIA AQUILARIA ARTEMISIA NEREUM TARAXACUM CUMINUM RAUWALFIA PAPAVER ACACIA ARECA CASSIAFISTUL ALOE LINUM ALTHAEA III PRUNUS SESAMUM ASPARAGUS GLYCYRHIZA PLANTAGO ZIZYPHUS

0.0 0.5Distances 1.0 1.5 Figure 4.4. Hierarchical clustering tree representing the distributional similarity of drug plants from India, as classified by . their Ayurvedic properties

157 ARTEMISIA ACHILLEA ABROMA RAUWOLFIA WITHANIA MELIA BERBERIS I CORIANDRUM BAMBUSA CINNZEYLANIC ELETTARIA EMBLICA SALVIA LAVENDULA SANTALUM CINNAMOMUMCA TERMINALIA PIPER CROCUS AQUILARIA PUNICA VALERIA II ZINGIBER PISTACIALENT PLANTAGO PISTACIAVERA ASPARAGUS PINUS MYRISTICA DELPHINIUM URGINEA OCIMUM ZIZYPHUS ALTHAEA III VIOLA GLYCYRHIZA PIMPINELLA RHEUM CASSIAFISTUL ROSACOMBO

0.0 0.5Distances 1.0 1.5 Figure 4.5. Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions from Karachi, Pakistan.

158 PIMPINELLA ANETHUMGRAV2 NIGELLA FOENICULUM DAUCUS ANETHUMGRAV1 LINUM LEPIDIUMSAT SESAMUM SINAPSIS I ANACYCLUS ELETTARIA PIPERCHABA MYRISTICA2 MYRISTICA1 PRUNUSAMYG CINNAMOMUMC ZINGIBER CYPERUS EUGENIA ALPINIA MYRTUS RAPHANUS II NERIUM IRIS ALOE MENTHA GLYCYRHIZA LAVENDULA CITRUSLIMON SALVIA PINUS RUBIA DELPHVITIS ASAFOETIDA CASSIAFISTUL III ALLIUMSATIV ALLIUMCEPA ROSA MELISSA CORIANDRUM PAPAVER RUTAGRAV THYMUS ARTEMISIA ROSMARINUS JUNIPERUS TRIGONELLA ARACHIS IV PEGANUM QUERCUS PUNICAGRANAT

0.0 0.5Distances 1.0 1.5 Figure 4.6. Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions from Marrakech, Morocco.

159 DAUCUS ACACIA CYMBOPOGON BOSWELLIA TRIGONELLA PISTACIAL QUERCUS ORIGANUM ALLIUMSATIV ZINGIBERZ I ARECA ALLIUMCEPA COMMIPHORA SALVADORA RUBIA ZIZYPHUS PIPERN LEPIDIUM PEGANUM NIGELLA CITRUSLIMON SESAMUM PIMPINELLA CARUMCOPT FOENICULUM ROSA CASSIAFIST II CASSIAACUT TERMINALIA1 ALPINIA CROCUS IRIS ANACYCLUS EUGENIA COFFEA ZINGIBERO III CINNAMOMUMC ELETTARIA THYMUS DELPHVITIS MENTHA MATRICARIA

0.0 0.5Distances 1.0 1.5 Figure 4.7. Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions from Sanaa, Yemen.

160 of other drug plants not shared with other sources could lead to the production of significantly different hierarchical clustering trees. Figures 4.1-4.7 represent classifications based on only those plants shared with other sources in Group I that could be identified as equivalent to Linnean taxa. Since all sources are subjected to the same “imbalance” or “handicap,” it seems logical that the comparisons made using these trees should even out any discrepancies due to missing data. The hierarchical clustering trees for all sources in Group I utilized the complete linkage (MAX) method, based on the Pearson coefficient. Group II, Pre-Modern and Modern Core Sources. Figures 4.8-4.16 include the hierarchical clustering trees representing the distributional similarity for all sources in Group II. The pre-modern sources (all sources in Group II excluding modern Egypt and Syria) utilized a total of fifty-seven drug plants that were not used by more than one modern source. Subsequently, several of the taxonomic trees for Group II (N=3) included over 100 drug plants. Due to the high number of taxa being grouped in some sources, the most complex of the cluster trees, though legible, may appear somewhat crowded. This situation is regrettable, and technical support staff at SYSTAT confirmed that software limitations preclude any improvements using the current version of the program. It is hoped that future editions may be more accommodating to larger data sets. Other statistical software packages with a hierarchical clustering function are faced with the same difficulty and have not offered any significant improvements over the SYSTAT output. As with the sources in Group I, Group II data were clustered using the complete linkage (MAX) method. While the majority of trees were based on the Pearson correlation coefficient, the cluster trees for ibn Wafid and Pseudo- Avicenna were based on average Euclidian (dissimilarity) distance (see Chapter 3).

Relationships and Influences among Sources

The application of numerical taxonomy to the sources under consideration produced results that allowed for the testing of this project’s two main research hypotheses (based on the upper-level clusters revealed in their individual distributional similarities, see Figures 4.1-4.16, where the clusters are indicated by roman numerals to the left of each figure). General Research Hypothesis 1 posited that there would be a strong correlation between (a) the overall similarity of patterns of drug plant prescription among sources in Group I and (b) their proximity and shared history. The

161 RHEUM I CITRULLUS LUPINUSTERM PRUNUSM TERMINALIA2 LINUM II LAVENDULA BOSWELLIA CROCUS ALOE TERMINALIA1 NARDOSTACHYS PISTACIAL ASTRAGALUS CITRUSLIMON ASAFOETIDA GLYCYRHIZA CUMINUM ANETHUMGRAV1 AMMI CARUMCOPT AMYGDALUS PINUS PISTACIAT COMMIPHORA QUERCUS ARECA III PUNICAGRANAT ANETHUMGRAV2 NIGELLA PEGANUM MYRISTICA2 CINNAMOMUMZ IRIS ELETTARIA CINNAMOMUMC EUGENIA PIPERN ANACYCLUS MYRISTICA1 ZINGIBER RAPHANUS LEPIDIUM PIPERCHABA DAUCUS ALPINIA CORIANDRUM TRIGONELLA MATRICARIA TILIA IV JUNIPERUS ERYTHREA ZIZYPHUS PAPAVER PIMPINELLA FOENICULUM CASSIAACUTIF ROSA

0.0 0.5Distances 1.0 1.5 Figure 4.8. Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions from Modern Syria.

162 CUMINUM CITRUSLIMON I MENTHA MATRICARIA TERMINALIA1 RHEUMRAPONT TERMINALIA2 QUERCUSINFEC PUNICAGRANAT CARUMCOPTICU LAWSONIAINER PEGANUMHARMA TRIGONELLAFO RAPHANUSSATI II ERYTHRAEASAT CORIANDRUMSA CYPERUSLONGU NARDOSTACHYS LAVENDULA ARECA ORIGANUM IRIS CINNAMOMUMZ PRUNUSMAHALE LINUMUSITATI TILIASYLVEST AMMIVISNAGA GLYCYRHIZAGL ACACIAARABIC III ASTRAGALUS LUPINUSTIRMI PISTACIALENT ALOE COMMIPHORAMY BOSWELLIACAR NIGELLASATIV ZINGIBEROFFI ALPINIAGALAN EUGENIACAROP CINNAMOMUMC ANACYCLUS ELETTARIA ANETHUMGRAV1 MYRISTICA2 IV MYRISTICA1 INULAHELENUM ZINGIBERZERA PISTACIATERE ANETHUMGRAV2 PIMPINELLAAN ROSAALBA FOENICULUMVU CASSIAACUTIF

0. 0 0. 5 Di s t a n c e s 1. 0 1. 5 Figure 4 .9. Hierarchical clustering tree representing th e distributional similarity of drug plants used in prescriptions from Modern Egypt.

163 FOENICULUMVU ANETHUMGRAV1 PIMPINELLAAN PIPER2 ALECTORIA CAPPARISSPIN RHUSCORIARI PORTULACA VITISSP PINUSCOMBO ALLIUMSATIVA LENSESCULENT VIOLAODERATA IRISSP LAURISNOBILI RICINUSCOMMU CICERARIETI LUPINUSTERMI AMYGDALUS VICIAERVILA LEPIDMSATCOM LINUMUSITATI CYPERUS POLYPODIUM AMMICARMCOPT GLAUCIUM CARTHAMUSTIN HYSSOPUS URGINEA ANETHUMGRAV2 JASMINIUM OPOPONAX FERULAGALB TRIBULUS ASAFOETIDA CUCURBITA RUMEXCHENOPO ERYTHREASPIC ECBALLIUM INULA CASSIAACUTIF I CORIANDRUMSA PRUNUSARMEN AGARICUS LAWSONIALBA ALLIUMPORUM TRIGONELLAFO BRASSICAOLEA FICUSCARICA ALPINIA CINNAMOMUMC MYRISTICA1 NARDOSTACHYS PISTACIALENT ELETTARIA1 EUGENIACAROP CINNAMOMUMZ AQUILARIASP ZINGIBEROFFI JUGLANS MYRTUSCOMMUN RUTAGRAVCOMB ORIGANUM MATRICARIACA MELILOTUS ARTEMISIA MORUS THYMUS1 ZIZYPHUS THUYA SANTALUMALBA ARECASP CUPRESSUS DELPHINVITIS BALSAMODENDR BOSWELLIACAR PAPAVERSOMNI ACACIA ASTRAGALCOMB CYDONDAUCUS AGRIMONIA SESAMUMORIEN OCIMUMMINI CUMINUMCYMIN CURCUMALONG1 CROCUSSATIVA ANACYCLUS COMMIPHMCOMB ALOE FERULAM ACORUS DORONICUM RHEUMRAPONTI II HYOSCAMUS VITEXAGNUSCA EUPHORBIA CHICORIUMEND ZINGIBERZER NIGELLASATIV CITRULLUS FERULAP MANDRAGORA PEGANUMHARMA HORDEUM TERMINALIA1 III GLYCYRHIZAGL ARISTOLOCHIA PLANTAGOALB ROSASP PUNICAGRANAT MENTHAAQUATI VI TAMARINDUSIN TAMARIX PIPER1 QUERCUSINFEC

0.00.51Distances .01.5 Figure 4.10. Hierarchical clustering tree representing the distributional similarity of drug plants used in prescriptions by al-Kindi.

164 URGINIA RUTAGRAV PIPERCUB PEGANUM OPOPONAX MENTHA GENTIANA FERULAMARMAR EUGENIA DORONICUM CINNAMOMUM ASAFOETIDA ALPINIA ACORUS ALLIUM ANETHUMGRAV CARUMCOPTICU CUMINUM ERYTHREA FERULAGALBA FOENICULUM JUNIPERUS NIGELLA ORIGANUM I PIMPINELLA POLYPODIUM THYMUS RICINUS PISTACTEREB MYRISTICA IRIS DAUCUS CARTHAMUS ARTEMISIA ALOE AQUILARIA BOSWELLIA COMMIPHORA FICUS JASMINUM PISTACLENTIS RHEUMRAPON TRIGONELLA SOLANUM QUERCUS GLAUCIUM ASTRAGALUS BERBERIS II PLANTAGOMAJ RHUSCORIARI CANNABIS ARECA TAMARINDUS PUNICA TRIBULUS ROSA MORUS CYDONIA BALSAMODENDR ACACIA ALECTORIA CERATONIA LAWSONIA MYRTUS TERMINALIA ZIZYPHUS VITIS VICIAERVILA TAMARIX PINUS NARDOSTACHYS MATRICARIA LINUM LAURUS CUPRESSUS III CASSIA AGRIMONIA AGARICUS BRASSICA CORIANDRUM ELETTARIA LAVENDULA LUPINUS MELILOTUS OCIMUM PRUNUSMAHAL THUYA VITEXAGNUS SESAMUM CICER JUGLANS RUMEX GLYCYRHIZA CHICORIUM PRUNUSARM VIOLAODERATA CUCURBITA CITRULLUSV PORTULACA INULA CYPERUSESC ZINGIBER

012 567 34Distances Figure 4.11. Hierarchical clustering tree representing the distributional similarity of drug plants noted by ibn Wafid, as classified by their humoral propoerties.

165 BRASSICA ANACYCLUS VALERIANA PIPER MYRISTICA1 EUGENIA CROCUS CASSIAANGUS I ANETHUM ALLIUMSATIV ASAFOETIDA CINNAMOMUMZE CUMINUM LAVENDULA MYRISTICA2 TRIGONELLA ZINGIBER TAMARIX CYDONIA II ARTEMISIA CANNABIS JUNIPERUS PAPAVER MELILOTUS LAWSONIA CUCUMIS ALOE CITRULLUS HYSSOPUS LINUM SESAMUM MALVA VIOLA GENTIANA BERBERIS III ARECA SANTALUM RHUS PLANTAGO PEGANUM JASMINUM COMMIPHORA MYRTUS PISTACIALENT PUNICA ROSA CITRUS ZIZYPHUS TAMARINDUS PIMPINELLA NIGELLA JUGLANSREGIA FOENICULUMVU ELETTARIA CHICORIUM BOSWELIA ACACIA AQUILARIA IV CARTHAMUS CORIANDRUM FICUSCARICA GLYCYRHIZA MENTHA ORIGANUM RAPHANUS OCIMUM MATRICARIA PHOENIX VITIS

012 567 34Distances Figure 4.12. Hierarchical clustering tree representing the distributional similarity of drug plants noted by Chisti (Pseudo- Avicenna), as classified by their humoral properties..

166 RUBIA INULACOMBO RAPHANCOMBO LAVENDULA VALERIANA DAUCUSCOMBO CINNAMOM2 PIPERCOMBO ZINGIBCOMBO NARDCOMBO I ELETTARIA BALSAMODENDR CINNAMOM3 EUGENIA BOSWELCOMBO COMMPIHCOMBO CROCUS HYSSOPUS PISTACIALENT TERMINALIA CASSIACOMBO ROSA PAPAVERCOMBO HYOSCYAMUS LACTUTASATIV CARTHAMUS RHUS II ACACIACOMBO MYRTUS THUYA VITIS PUNICA MANDRAGORA PHOENIX ALOE RHEUM LEPIDIUM DORONICUM ORIGANCOMBO THYMUS CITRULLUS FERULAPERSIC AGARICUS CONVOLVOLUS POLYPODIUM SESAMUM CAPPARIS CYPERUS VIOLA LUPINUS MORUS BERBERIS ECBALLIUM LAURUS ACORUS FICUSCARICA CICER PRUNUSARMENI ASPARAGUS JASMINUM SANTALUM MALVA OCIMUM ARECA ALECTORIA AGRIMONIA ARISTOCOMBO ARTEMCOMBO VERATRUMCOMB PORTULACA III ERYTHREA JUNIPERUS NIGELLA IRISCOMBO BETA CINNAMOM1 CEDRUS PISTACIATER HORDEUM RUMEX CHICORIUM VICIA BRASSICACOMB TAMARIX TRIGONELLA LINUM MATRICARIA MELILOTUS CUPRESSUS ANETHUM ASTRAGALUS GLYCYRHIZA AMYGDALUS PINUS ALLIUMPORUM CORIANDRUM CUCURBITA PEUCEDANUM CARUMCOPTICM EUPHORBIUM ANACYCLUS FOENICCOMBO RUTACOMBO ALPINIA CUMINUM PIMPINELLA MENTHACOMBO QUERCUS FERULMCOMBO FERULAGALBAN OPOPONCOMBO GLAUCMCOMBO

0.0 0.5Distances 1.0 1.5 Figure 4.13. Hierarchical clustering tree representing the distributional similarity of drug plants used in c. 400-500 A.D. Greco-

Syriac prescriptions (Syriac α).

167 ZINGIBER CAPPARIS RUBIA BOSWELLIA SESAMUM COMMIPHORA CICER MORUS ANACYCLUS CINNAMOMUM ALOE CEDRUS NIGELLA LACTUTA CANNABIS BETA ARISTCOMBO PINUS CUPRESSUS I MATRICARIA ASPARAGUS BRASCOMBO IRIS JUGLANS PAPAVER CORIANDRUM CROCUS RUTA CARTHAMUS RAPHANUS TAMARIX ALLIUMPORUM ARTEMISIA HORDEUM ROSA MENTHA HYOSCYAMUS MYRTUS VITIS AMYGDALUS LINUM II PEUCEDANUM CUMINUM PIPER CYDONIA ALLIUMSATIVA ASAFOETIDA PORTULACA VERATRUM LAWSONIA PISTACIATER III EUGENIA HYSSOPUS ALLIUMCEPA PHOENIX FICUS GLYCYRHIZA CERATONIA VICIA MANDRAGORA PLANTAGO FOENICULUM ANETHUM CUCURBITA TRIGONELLA IV QUERCUS PIMPINELLA LENS PUNICA RHUS BERBERIS CHICORIUM

0. 0 0. 5 Di s t a n c e s 1. 0 1. 5 Figure 4 .14. Hierarchical clustering tree representing the distributional similarity of drug plants used in c. 200 - 300 A.D. “native” Syriac prescriptions (Syriac β).

168 CUMINUMCOMBO BRASSICA MATRICARIA I PISTACLENTIS CUCUMIS PEGANUM MYRTUS MELILOTUS ROSA BOSWELLIA CONVOLVOLUS LEPIDIUMCOMB PIPER TERMINALLIA CURCUMA ARTEMESCOMBO RAPHANUS SOLANUM SANTALUM ALLIUMCEPA ALLIUMPORUM PINUS ORIGANUM HYSSOPUS MALVA LAURUS II GENTIANA CHICORIUM ANETHUM ELETTARIA HORDEUM FICUSCARICA LINUM NIGELLA ZIZYPHUS PISTACTEREBI OPOPONCOMBO JUNIPERUS CARTHAMUS PUNICA FOENICULUM LENS PHOENIX TAMARIX VICIA ARISTOLCOMBO RICINUS CORIANDER NARDOCOMBO EUGENIA III ALPINIA ZINGIBER ALOECOMBO ASAFOETIDA INULA ACACIACOMBO IV PAPAVER COMMIPHCOMBO CROCUS ASTRAGCOMBO

0.0 0.5Distances 1.0 1.5 Figure 4.15. Hierarchical clustering tree representing the distributional similarity of drug plants used in c. 800-900 A.D.

Coptic prescriptions (P. Chassinat).

169 TRIGONELLA CASSIA MYRTUS HORDEUM ROSA INULACONYZA MENTHA I LACTUTASATIV PHOENIX RICINUS CHICORMCOMBO ASTRAGALUSGU ARTEMISIAABS VALERIANA ANACYCLUSPYR CORIANDRUM ANETHUM CUCUMISSATIV PIMPINELLA LEPIDIUM II FICUSCARICA VITIS CUMINUM CERATONIA ACACIACOMBO ALOE ASAFOETIDA COMMPIHORA BOSWELLIA III CINNAMOMUM BALSAMODENDR JUNIPERCOMBO VITEX SINAPSIS CYPERUSECOMB IV ZIZYPHUS CITRULLUS ALLIUMCEPA CANNABSSATIV

0. 0 0. 5 Di s t a n c e s 1. 0 1. 5 Figure 4 .16. Hierarchical clustering tree representing the distributional similarity of drug plants used in c. 1500 B.C. Egyptian

prescriptions (P. Ebers ) .

170 specific corollary of this hypothesis stated that a distributional similarity-based numerical taxonomy of contemporaneous sources would show sources from the Core region to be most similar to one another and those in more peripheral areas to be most similar to sources in the Core or to their immediate non-Core neighbors. General Research Hypothesis 2 posited that there would be a strong correlation between (a) the overall similarity of patterns of drug plant prescription by sources in Group II and (b) the relative strength of historical (antecedent-successor) relationships among those sources. Its corollary, Specific Research Hypothesis 2, stated that a numerical taxonomy of sources from three or more periods or streams of Middle Eastern medical tradition would show the earliest and latest sources to be more similar to the middle sources than to one another. Within each group, clusters of sources were identified that compare favorably with the known geographic, cultural, and historical relationships and influences that have shaped Middle Eastern medicine. The clusters and their relation to the facts of the culture area and its development will be discussed below under headings related to each of the two hypotheses. A more descriptive and inferentially historical scheme of interrelationships among sources, based on the rankings and clusters reported in this chapter, will be posited in the concluding chapter. Group I, modern sources only. Table 1 presents the similarity rankings for Aleppo, Cairo, Gaziantep, India, Karachi, Marrakech, and Sanaa, from the point of view of each source. The rankings are based on the arithmetic mean

(A), geometric mean (G), arithmetic mean of the usability quotient (UA), and geometric mean of the usability quotient (UG) of overall similarity in patterns of drug-plant prescription among the sources. The measurements given within the table are identical to those used in the data matrices from which the hierarchical clustering trees, multidimensional scaling plots, factor analyses, and k-means clustering results found in figures 4.17-4.20 and table 2 were generated. While the figures in table 1 are technically percentages, they are treated herein as arbitrary units that are relative, i.e., scaled within the bounds of each measure. This is due to the fact that the A and G measures are clearly operating on a different scale from the U measures, although, within each group, the relative rank orderings of sources by the various measures are generally in overall agreement regarding which sources are most similar and dissimilar to one another. It was found that the hierarchical clustering trees based on A and G were identical. Those based on UA and UG were also nearly identical. As a

171

Table 1 Similarity Rankings of Sources in Group I, in Arbitrary Units

Aleppo, Syria (Al) A Ca .598 Sa .486 Ma .484 In .417 Ga .358 Ka .347 G Ca .594 Ma .478 Sa .467 In .416 Ga .336 Ka .326 UA Ca .721 Ma .609 Sa .586 In .464 Ga .422 Ka .333 UG Ca .720 Ma .608 Sa .577 In .462 Ga .418 Ka .326 Cairo, Egypt (Ca) A Al .598 Ma .435 Sa .424 In .307 Ga .281 Ka .195 G Al .594 Ma .435 Sa .418 In .307 Ga .273 Ka .190 UA Al .721 Sa .571 Ma .547 In .442 Ga .362 Ka .239 UG Al .720 Sa .566 Ma .547 In .439 Ga .361 Ka .236 Gaziantep, Turkey (Ga) A Al .358 Ca .281 In .239 Ma .238 Ka .166 Sa .112 G Al .336 Ca .273 Ma .233 In .230 Ka .166 Sa .112 UA Al .422 Ca .362 In .359 Ma .307 Ka .257 Sa .181 UG Al .418 Ca .361 In .352 Ma .306 Ka .256 Sa .182 India (In) A Ka .442 Al .417 Ma .356 Sa .315 Ca .307 Ga .239 G Ka .426 Al .416 Ma .355 Sa .309 Ca .307 Ga .230 UA Ka .589 Al .464 Sa .476 Ca .442 Ma .435 Ga .359 UG Ka .566 Al .462 Sa .461 Ca .439 Ma .430 Ga .352 Karachi, Pakistan (Ka) A In .442 Al .347 Ma .314 Sa .238 Ca .195 Ga .166 G In .426 Al .326 Ma .307 Sa .238 Ca .190 Ga .166 UA In .589 Al .333 Ma .331 Sa .317 Ga .257 Ca .239 UG In .566 Ma .328 Al .326 Sa .317 Ga .256 Ca .236 Marrakech, Morocco (Ma) A Al .484 Ca .435 In .356 Sa .344 Ka .314 Ga .238 G Al .478 Ca .435 In .355 Sa .341 Ka .307 Ga .233 UA Al .609 Sa .559 Ca .547 In .435 Ka .331 Ga .307 UG Al .608 Sa .556 Ca .547 In .430 Ka .328 Ga .306 Sanaa, Yemen (Sa) A Al .486 Ca .424 Ma .344 In .315 Ka .238 Ga .112 G Al .467 Ca .418 Ma .341 Ka .238 In .309 Ga .112 UA Al .586 Ca .571 Ma .559 In .476 Ka .317 Ga .181 UG Al .577 Ca .566 Ma .556 In .461 Ka .317 Ga .182

172 Sanaa

Aleppo

Cairo

Marrakech

Karachi

India

Gaziantep

0.0 0.5 1.0 1.5 2.0 Distances

Figure 4.17. Complete Linkage Cluster Analysis of Sources in Group I Pearson Correlation Coefficient of A.

Karachi

India

Gaziantep

Cairo

Aleppo

Marrakech

Sanaa

0.0 0.5 1.0 1.5 2.0 Distances

Figure 4.18. Complete Linkage Cluster Analysis of Sources in Group I Pearson Correlation Coefficient of UA

173 2

1 Karachi India Sanaa

Marrakech 0 Aleppo Cairo

-1

Gaziantep

-2 -2 -1 0 1 2

Figure 4.19. Two-Dimensional MDS Solution for Sources in Group I, UA

1.0 Marrakech

Aleppo 0.5 Cairo

Sanaa

0.0 India Karachi

-0.5 Gaziantep

1.0 0.5 1.0 0.0 0.5 0.0 -0.5 -0.5 -1.0 -1.0

Figure 4.20. Plot of Principal Components of Similarity Data, A

174 Table 2 K-Means Clustering Results for Sources in Group I (A)

Cluster K=2 K=3 K=4 K=5

Cluster 1 Gaziantep Karachi Karachi Karachi Karachi India India India India

Cluster 2 Aleppo Aleppo Aleppo Aleppo Cairo Cairo Cairo Cairo Marrakech Marrakech Sanaa Sanaa

Cluster 3 -- Gaziantep Gaziantep Gaziantep

Cluster 4 -- -- Marrakech Marrakech Sanaa

Cluster 5 ------Sanaa

result, only the trees based on A and UA will be shown below. Multidimensional scaling plots and factor analyses were somewhat more variable, but not significantly so in terms of the actual clusters they suggest. Therefore, the figures presented below will be limited to those based on the measures that produced the clearest portrayal of tendencies evident in the majority of outputs generated using each instrument. As shown in the hierarchical clustering trees of figures 4.17 and 4.18, Aleppo, Cairo, Marrakech, and Sanaa are one clear cluster. Within this cluster, Aleppo-Cairo is a definite subcluster. Due to a contrast between the two figures (and the measures in table 1), it is unclear whether or not Sanaa or Marrakech is more closely related to Aleppo-Cairo, or if Marrakech and Sanaa are simply equidistantly “orbiting” around the subcluster. Karachi-India is another clear cluster. Gaziantep is only weakly linked to the Karachi-India cluster in figures 4.17 and 4.18 (at a Euclidian distance only slightly below 1.5). In table 1 it is always one of the two most weakly

175 related sources to any other source (including India and Pakistan). In addition, a clustering tree using a single linkage (MIN) algorithm (rather than complete linkage) showed the India-Pakistan cluster linking to the Aleppo-Cairo-Marrakech-Sanaa cluster before Gaziantep (which was the last to join). These facts suggest that Gaziantep should be seen as acting as its own group (an isolate) according to the results of the numerical taxonomy. Multidimensional scaling (MDS), factor analysis and K-means clustering results lend further support to all of the above contentions (see especially table 2, K=3-5). From a holistic point of view, the overall shape of the clusters seen in figures 4.17-4.20 and table 2 accords rather well with the culture area morphology described in the previous chapter. The reader should recall that Aleppo and Cairo are representatives of the Proximal Core, Marrakech and Sanaa of the Proximal Domain and Distal Core/Proximal Domain, respectively, and Karachi (in the so-called “Double-Periphery Zone”) is well within the orbit of the Inner Asian sphere (represented by the out group, India, which Pakistan borders). In addition, Gaziantep, Turkey, is part of the Distal Sphere of the culture area. The results of numerical taxonomy place Aleppo and Cairo at the center of a cluster (its Core). They are orbited by Marrakech and Sanaa (representing the Proximal Domain). India (a member of the Inner Asian Domain/Sphere) and Karachi (in the double periphery between Inner Asia and the Islamic Middle East, but geographically nearer to, and possibly within, the Inner Asian sphere) are in another cluster. Turkey (belonging to the Middle Eastern Sphere) is placed as an isolate. From a more particularistic point of view, Aleppo and Cairo (members of the Proximal Core) both rank each other as most similar to themselves; Gaziantep, Marrakech, and Sanaa (members of the Sphere and Proximal Domain) each rank Aleppo and Cairo as most similar to themselves, and both Karachi and India rank Aleppo as the most similar to themselves after the other member of their own cluster. Thus, the numerical taxonomy shows sources in the Core (i.e., Aleppo and Cairo) to be more similar to one another than to sources on the periphery. Peripheral sources (i.e., Gaziantep, Karachi, India, Marrakech, Sanaa) are more similar to sources in the Core or to their immediate neighbors (e.g., India-Pakistan). These results support Specific Hypothesis 1 in all details. Group II, pre-modern and modern Core sources. Table 3 presents the similarity rankings of Modern Syria (Aleppo), Modern Egypt (Cairo), al-Kindi, Ibn Wafid, Pseudo-Avicenna, Syriac α, Syriac β, P. Chassinat, and P. Ebers,

176 from the point of view of each source. The rankings are based only on the arithmetic mean of the usability function (UA) and geometric mean of the usability function (UG) of overall similarity in patterns of drug-plant prescription among the sources in Group II. As with the measures given in table 1, the measures given in table 3 are identical to those used in the data matrices from which the hierarchical clustering trees, multidimensional scaling plots, factor analyses, and k-means clustering results found in figures 4.21-4.24 and table 4 were generated. Again, as with the results for Group I, the units in table 3 are technically percentages, but are treated as arbitrary (following Yablokov 1986). It was found that the factor analysis plots and K-means clustering results based on UA and UG were identical to one another. As a result, only the factor analysis plot and K-means solutions based on UA will be shown below. Hierarchical clustering trees and multidimensional scaling plots were somewhat more variable, but not significantly so in terms of the requirements necessary for confirming or rejecting the relevant hypotheses. The similarity rankings in table 4, hierarchical clustering trees in figures 4.21 and 4.22, and MDS and factor analysis plots in figures 4.23-4.25 suggest two main clusters for Group II: (a) Egypt-Syria and (b) a cluster centered on al-Kindi and Ibn Wafid (the two most strongly related sources in the sample) extending to include Pseudo-Avicenna and Syriac α. Contrastingly,

Syriac β, P. Chassinat, and P. Ebers do not cluster well with other sources. In terms of higher-level groupings, it is unclear from the figures whether Egypt-Syria is closer to the al-Kindi-Ibn Wafid-Syriac α cluster, or whether

Syriac β, P. Chassinat and P. Ebers are. However, by averaging the similarity scores of Egypt and Syria in relation to al-Kindi, Ibn Wafid, and Syriac α (UA

.545 and UG .524) and likewise averaging the similarity scores of Syriac β,

P.Chassinat and P. Ebers in relation to the al-Kindi-Ibn Wafid-Syriac α cluster (UA .519 and UG .495), it becomes apparent that Egypt-Syria is slightly closer to the al-Kindi-Ibn Wafid-Syriac α cluster. On the whole, these patterns are in concord with what is known about the major historical relationships of descent and influence among the sources. Al-Kindi, ibn Wafid, and Avicenna are broadly contemporaneous

177

Table 3 Similarity Rankings of Sources in Group II, in Arbitrary Units

Modern Syria (Sy) UA Eg .703 Ki .557 Wa .553 Sα .523 Av .507 Ch .440 Sβ .389 Eb .214 U Eg .703 Wa .533 Ki .526 Ch .440 Eb .210 G Av .505 Sα .496 Sβ .386 Modern Egypt (Eg) UA Sy .703 Sα .625 Wa .616 Ki .550 Av .433 Ch .373 Sβ .343 Eb .222 U Sy .703 Wa .584 Ki .512 Av .429 Ch .372 Eb .219 G Sα .584 Sβ .339 Al-Kindi (Ki) UA Wa .822 Sα .775 Av .625 Sβ .587 Sy .557 Eg .550 Eb .480 Ch .443 U Wa .821 Av .606 Sy .526 Eg .512 Ch .421 Eb .418 G Sα .775 Sβ .571 Ibn Wafid (Wa) UA Ki .822 Sα .777 Av .728 Ch .635 Eg .616 Sβ .556 Sy .553 Eb .513 U Ki .821 Av .715 Ch .613 Eg .584 Sy .533 Eb .459 G Sα .776 Sβ .548 Pseudo-Avicenna (Av) UA Wa .728 Ki .625 Sα .608 Sβ .553 Sy .507 Ch .451 Eg .433 Eb .381 U Wa .715 Ki .606 Sy .505 Ch .450 Eg .429 Eb .366 G Sα .591 Sβ .553 Syriac α (Sα) Ch .549 Sy .523 UA Wa .777 Ki .775 Sβ .672 Eg .625 Eb .625 Av .608 Wa .776 Ki .775 Av .591 Eg .584 Ch .524 Sy .496 Eb .363 UG Sβ .657 Syriac β (Sβ) UA Sα .672 Ki .587 Wa .556 Av .553 Ch .504 Eb .414 Sy .389 Eg .343 Ki .571 Av .553 Wa .548 Ch .502 Eb .396 Sy .386 Eg .339 UG Sα .657 P. Chassinat (Ch) UA Wa .635 Sα .549 Sβ .504 Av .451 Ki .443 Sy .440 Eg .373 Eb .373 U Wa .613 Av .450 Sy .440 Ki .421 Eg .372 Eb .351 G Sα .524 Sβ .502 P. Ebers (Eb) UA Wa .513 Ki .480 Sα .414 Sβ .414 Av .381 Ch .359 Eg .222 Sy .214 U Wa .459 Ki .418 Ch .351 Eg .219 Sy .210 G Sβ .396 Av .366 Sα .363

178 Syria

Egypt

Syriac α

Al-Kindi

Ibn Wafid

Pseudo-Avicenna

P. Chassinat

Syriac β

P. Ebers

0. 0 0. 5 1. 0 1.5 2. 0 Di s t a n c e s

Figure 4.21. Complete Linkage Cluster Analysis of Sources in Group I Pearson Correlation Coefficient of UA

Egypt

Syria

Pseudo-Avicenna

Ibn Wafid

Al Kindi

Syriac α

Syriac β

P. Chassinat

P. Ebers

0. 0 0. 5 1. 0 1.5 2. 0 Di s t a n c e s

Figure 4.22. Complete Linkage Cluster Analysis of Sources in Group I Pearson Correlation Coefficient of U G

179 2

1 Egypt P. Ebers

Al Kindi Syria 0 Syriac α Ibn Wafid

Pseudo-Avicenna Syriac β

-1 P. Chassinat

-2 -2 -1 0 1 2

Figure 4.23. Two-Dimensional MDS Solution for Sources in Group II, UA

2

1 Egypt P. Ebers Al Kindi Syria Pseudo-Avicenna 0 Ibn Wafid Syriac α

Syriac β -1 P. Chassinat

-2 -2 -1 0 1 2

Figure 4.24. Two-Dimensional MDS Solution for Sources in Group II, UG

180

1.0 P. Chassianat

0.5 Syriac α Syriac β Ibn Wafid SYRIA Pseudo-Avicenna

0.0 EGYPT Al Kindi

-0.5

P. Ebers 1.0 0.5 1.0 0.0 0.5 0.0 -0.5 -0.5 -1.0 -1.0

Figure 4.25. Plot of Principal Components of Similarity Data, UA

181 Table 4 K-Means Clustering Results for Sources in Group II (UA/G)

Cluster K=2 K=3 K=4 K=5 K=6 K=7 K=8

Cluster 1 Syria Syria Al-Kindi Al-Kindi Al-Kindi Al-Kindi Al-Kindi Egypt Egypt Ibn Wafid Ibn Wafid Ibn Wafid Ibn Wafid Ibn Wafid Al-Kindi Al-Kindi P-Avicenna Syriac α Syriac α Syriac α Ibn Wafid Ibn Wafid Syriac α P-Avicenna Syriac α Syriac α Cluster 2 Syriac β P-Avicenna Syriac β P. Chassinat P. Chassinat P. Chassinat P. Chassinat P. Chassinat Syriac β P. Chassinat P. Ebers P. Chassinat

Cluster 3 P. Ebers P. Ebers P. Ebers P. Ebers P. Ebers P. Ebers

Cluster 4 Egypt Egypt Egypt Syria Syria Syria Syria Syria

Cluster 5 P-Avicenna Syriac β Syriac β Syriac β Syriac β

182 Table 4 – Continued K-Means Clustering Results for Sources in Group II (UA/G)

Cluster K=2 K=3 K=4 K=5 K=6 K=7 K=8

Cluster 6 P-Avicenna P-Avicenna P-Avicenna

Cluster 7 Egypt Egypt

Cluster 8 Syriac α

183 (“medieval,” i.e., c. 500-1500) representatives of Greek humoral medicine in an Islamic framework and thus should be expected to cluster together. Given the sources under investigation, Syriac α ought to cluster with them as a near antecedent (it will be recalled that the Syriac tradition of Alexandrian medicine is the conduit through which Galenic theory and practice came to the Muslim world). Egypt and Syria are contemporaries to one another and should therefore cluster together as well. “Native,” non-Galenic traditions (Syriac β, P. Chassinat, and P. Ebers) likely have deep historical roots in their own territories. Even though two of them (Syriac β and P. Chassinat) come from time depths where contact with Greeks and some intellectual cross-pollination may have been possible (see chapter 1), the general scholarly consensus is that they over-ridingly represent local non-Greek systems. The modern sources (Egypt and Syria) are more similar to the medieval cluster than they are to these “local”/”native” pre-Greek systems. Similarly, Syriac β, P. Chassinat, and P. Ebers are more similar to Al-Kindi, ibn Wafid, and Avicenna than they are to modern Egypt or Syria. For Group II, a particularistic perspective does not provide many more enlightening insights than a holistic one, as it does for Group I. However, some observations of import may still be made. After Al-Kindi and ibn Wafid, Syriac α is most similar to Syriac β. Syriac β is most similar to Syriac α.

This special kinship between Syriac α (Alexandrian medicine as practiced by a third-fourth century Syrian) and Syriac β (traditional Mesopotamian cures from the same period) is what we would predict of two sources from the same temporal and geographic provenience, even if they claim to adhere to different ethnomedical traditions (much like the modern, near-neighbors Karachi, Pakistan, and India in Group I). Pseudo-Avicenna, a modern summary of a nineteenth-century commentary of an abridgement of a medieval work, though most similar to the “true” medieval Islamic medical texts, is still the most peripheral member of its cluster. This is what should be expected in a situation where translation and abridgement (without significant modification) prevail. Finally, P. Ebers, our most ancient text (at a remove of at least 1,834 years from its nearest chronological neighbors, Syriac α and Syriac β) is the most consistently dissimilar source in relation to all others in the sample.

184 It is only ever ranked higher than the least similar source by the two Syriac sources, which are the nearest to it in time. In placing (a) the medieval sources together in a cluster, (b) the Alexandrian Galenic source along with them, and (c) the cluster of modern sources descended from the medieval Islamic tradition nearest to the medieval cluster, a numerical taxonomy of the sources in Group II shows similar interrelationships to those that are known from the historical record of Islamic pharmacological practice. These results support the second pair of general and specific research hypotheses in all respects.

The Efficacy of Various Instruments and Measures

The third research question under consideration focused on discovering what instruments and measures might best facilitate the application of numerical taxonomy procedures to G2 cognitive categories. Each of the measurements and instruments used in the course of this investigation will be briefly discussed below. Instruments. The four instruments used in this study are agglomerative hierarchical clustering, multidimensional scaling, factor analysis and K- means clustering. They each have their individual pros and cons. Taken as a whole, however, they serve as checks on one another, alleviating the most extreme misinterpretations that would likely arise if procedures were limited to a single-instrument approach. Hierarchical clustering procedures produced results visually similar to those with which historical linguists and evolutionary biologists are most familiar (i.e., the Stammbaum). The trees made the strongest clusters readily apparent with only a cursory inspection. Occasionally, however, the necessity of clustering all entities with one or more other entities can mislead a naïve researcher into assuming relationships where there are none (as in the case of clustering Gaziantep with India and Karachi in Group I in

Figs. 4.17 and 4.18, or the discrepancy between the Group II trees for UA and

UG in Figs. 4.21 and 4.22). Double checking results based on complete linkage (MAX) clustering by using a single linkage (MIN) algorithm alleviated some of this problem. Multidimensional scaling (MDS) does not force items into clearly demarcated clusters and thereby circumvents the major pitfall of hierarchical clustering. Rather, it allows entities to gravitate toward one another without necessitating the drawing of hard and fast boundary lines between groups or of making connecting lines like the nodes of a tree. In addition,

185 MDS seems to provide a fairly accurate representation of the numerical interrelationships of sources considered in this project. For the two groups under consideration, the “stress” (how well actual similarities fit the distances in an MDS plot) was between .11 and .12 for the MDS plots shown in Figures 4.19, 4.23 and 4.24, a relatively “good” result (see Wilkinson et al. 1996:667). In addition, the Shepard diagrams for the plots produced multi- step outputs where the variance of the residuals was equal to or larger than the variance of the fitted curve, also a good result (see Wilkinson et al. 1996:668). One drawback to MDS is that it can occasionally place fairly dissimilar items into proximities that are visually suggestive of relationships not actually present. For example, in Figure 4.23, Syriac β and Pseudo-Avicenna appear to be as similar to one another as modern Syria and Egypt are to each other, a patently false conclusion in the face of the numerical evidence in table 3. Factor analysis generates factor plots that seem to accentuate and disambiguate the similarities and dissimilarities evident in MDS clusters. The added dimensionality of a multi-factor plot, like the one found in figure 4.25, precludes some of the misinterpretation that could result from MDS solutions like the ones shown in Figures 4.23 and 4.24. However, also like MDS (but unlike hierarchical clustering), factor analysis does not allow for the recognition of higher levels of grouping. Nevertheless, the factor analysis results fit the data at hand reasonably well. The three factors of figures 4.20 and 4.25 accounted for approximately 71 percent and 76 percent of the total variation present, respectively. For the present sample, K-means clustering analyses with K set between three and five seemed to produce the most revealing clusters. By running multiple cluster analyses with K set at different levels, it was possible to test the clusterings suggested by the other instruments and confirm or disconfirm them. Unlike MDS and factor analysis, a multiply-iterated K-means clustering procedure does allow higher level grouping. However, it also shares a disadvantage with hierarchical clustering: it must join all variables into however many groups the program has been instructed to make, sometimes lumping dissimilar items together for want of anything else to do with them. In the final analysis, none of the instruments is superior in all respects to the others. Taken together, they are highly complementary, with each providing a corrective to the excesses or short fallings of the others.

186 Measures. Like the instruments, the various measures served to constrain and complement one another. Both the arithmetic and the geometric means seem to provide roughly the same overall results. Any disagreements between them (such as the discrepant rankings of the relationships prevailing between Gaziantep and Marrakech, Karachi and Sanaa, and Gaziantep and India in Table 1) are rather minor issues of sub-ranking within a series of rankings that, relative to the rankings of other pairs, is generally established as a continuous “run” of interrelationships by both measures. Both A and G offer a fairly “fine-grained” measurement because they are based on numerous implicationally ordered characters. The U measures (not based on implicationally ordered characters but on the “usability” of variables as discovered through paired comparison of sources) are inherently less nuanced than either the A or G measure, but still provide results that generally comparable to them. In addition, they provide a distinct advantage over the standard A and G measures in their ability to measure similarity across a larger pool of sources while demanding what is only a minimally more labor intensive process with the addition of each additional source

Summary

This chapter consisted primarily of a description of each source in the sample, the results of testing the research hypotheses, and an evaluation of the instruments and measures used to achieve the results. The results confirmed both the general and specific research hypotheses, demonstrating that a multi-instrumental, multi-measure numerical taxonomy approach accurately groups cognitive dialects of Islamic medicine and representatives of its antecedents into clusters that would be expected given the known facts of history and geography in the region. In the final chapter, we will review the study as a whole, as well as how the results of the present chapter relate to the purposes of the project and to the studies reviewed in Chapters 1 and 2. In addition, statements the findings of the project will be made and implications drawn regarding both the study of Middle Eastern ethnomedicine and future prospects for the development of cognitive dialectometry as a method. Finally, suggestions for further research using the techniques of cognitive dialectometry will be proposed.

187 SUMMARY DISCUSSION: THE EFFICACY OF A PHENETIC APPROACH TO COGNITION

This chapter consists of three primary sections: (a) an overview of the study as a whole, (b) an interpretive review of the results reached and some conclusions that maybe drawn from them, (c) a discussion of the implications of the results and conclusions of this study for theory and method, and (d) recommendations for “next steps” in applying the lessons learned along the way.

Overview of the Study

This section begins with a brief discussion of the “problem of history” in cognitive studies in general and how it was addressed in the previous chapters. This is followed by a summary of the procedures undertaken in the present attempt to address this issue. Finally, the research hypotheses considered are reviewed.

Toward the Measurement of Cognitive Dialects of Islamic Ethnomedical Practice

The attempt to quantify similarity in overall pattern among data points is tantamount to a “cognitive dialectometry” of Islamic ethnopharmacology and its precursors, and is a first step in the development of a comparative historical approach to cognition. This research project investigated whether the degree of similarity between various Islamic and pre-Islamic Middle Eastern societies’ overall patterns of drug-plant prescription, as calculated using principles of numerical taxonomy, would correlate with the known facts of culture area morphology and succession of intellectual traditions in the Greater Middle East region. The project was specifically concerned with practices of drug-plant prescription in Islamic societies, but of paramount concern was the notion of

188 measuring cognitive variation and similarity in general, and the consideration of how the results of such measurements might be applied to the investigation of historical questions in particular. Until the present project, only a “first-generation” perspective on cognition had been applied to the investigation of intra-society historical relationship, descent, and borrowing. Such a perspective can only see the mental categories of folk classifications in terms of minimal contrastive semantic features, arranged in relationships of inclusion or contrast on successive levels. The first-generation emphasis on minimal defining features (or criterial attributes) and clear, either-or relationships of category inclusion or exclusion has allowed cognitive anthropologists interested in history to posit reconstructions of proto-cultures’ semantic structures and taxonomic trees. Their techniques are modeled on procedures for internal and comparative reconstruction in traditional historical linguistics. However, such procedures are limited in their application to cognitive systems, in that their focus on the linguistic sign (i.e., the lexeme) does not generally allow for cross-linguistic comparison of structure above a fairly rudimentary level. First-generation studies have generally only focused on reconstructing the denotative features of categories on a single level of contrast in taxonomic hierarchies (e.g., Friederich 1970) or on assigning cognitive systems to places in very general typologies of organization appearing among a group of societies (e.g., Bright and Bright 1965). However, no first- generation study seems to have ever posited quantifiable degrees of affinity among classificatory systems taken as wholes. This project represents an attempt to apply principles from one small sub-branch of historical linguistics, to a “second-generation” perspective on cognition. As noted in the introductory chapter, a second-generation perspective is one where mental categories are seen as gradient, prototype- centric gestalts sharing both denotative and connotative attributes and having “fuzzy” boundaries (see D’Andrade 1995; Kronenfeld et al. 1985; Lakoff and Johnson 1999, and Rosch 1973 and 1978). One way of observing the tendencies of category representatives to cluster together around best case exemplars involves paying attention to their distributional similarities, or their tendency to co-occur in the same environments (see D’Andrade et al. 1972 and Stefflre 1972). Incidences of co-occurrence are easily quantified, and therefore amenable to comparison. The problem, then, was where to find

189 an accurate method for comparing these tendencies toward grouping mental and material phenomena in the same way. One possibility for comparing whole patterns of category co-occurrence was numerical taxonomy, called “dialectometry” in its linguistic applications or “phenetic classification” in its biological and archaeological incarnations. Numerical taxonomy, in its most basic form, involves the quantification of character correlations or “family resemblances” using external markers of similarity among a set of entities, whether they be biological, linguistic, or cultural (see Embleton 1991, Goebl 1991, Heywood and McNeill 1964, Hill and Crane 1982, Nerbonne and Kretzschmar 2003, O’Brien and Lyman 2000 and 2003, Sneath and Sokal 1973, Sokal 1986, and Yablokov 1986). As in second-generation cognitive studies, pheneticists of all stripes see categories as gradient aggregations whose members share constellations of attributes with one another. The more attributes they share in common, the more similar they are reckoned to be. These phenotype- based classifications do not necessitate a distinction between genetic (i.e., inherited) characters and borrowed or independently evolved characters. However, pheneticists argue that affinities resulting from common ancestry underlie the bulk of surface-level similarities (Sokal and Sneath 1963:220) and that the quantification of perceptually similar categories can serve as an initial starting point for researching their genetic relationships. Some theorists have proposed that surface-level patterns of practice likely reflect the influence of deep-level idea units, or “memes,” that are inexactly replicated through imitation and are transmitted both vertically, from generation to generation, and horizontally, through “thought contagion” from neighboring groups (see Aunger 2002, Blackmore 1999, Cartwright 2000, Dawkins 1976, 1989, and 1999, and Lynch 1996). While the memetic model has the advantage of accommodating the possibility of more than one path or kind of idea transmission, memeticists have so far been unable to develop a method for directly measuring the sharing of memetic categories or structures across groups. By combining the basic idea of memetic transmission through both vertical and horizontal pathways with dialectometric means of quantifying overall family resemblances (regardless of their means of transmission), this project is a first step toward the measurement of cognitive dialects in a geo-historical perspective. In order to validate the method of “cognitive dialectometry” developed in reaction to (and as a synthesis of) second-generation cognitive studies,

190 numerical taxonomy, and memetic theory, it was necessary to apply the resulting procedures to a test case. Islamic medical ethnobotany served as an example for this initial trial. The Islamic ethnomedical system may be characterized as traditionally transmitted patterns of drug-plant prescription found from Morocco to Pakistan (where Islam is both a dominant symbol and the primary ideology) and claiming a common intellectual heritage supposed to derive ultimately from the humoral medical system of the ancient Greeks. The “Islamic commonwealth,” Like other culture areas, is divisible into a continuum of Core, Domain, Sphere, and “double periphery” territories that realize the socio-cultural and linguistic effects of both geographic circumstances and the historical interactions of people and ideas. It provided a nearly ideal scenario for an initial assay at cognitive dialectometry: a well-known culture area which that readily sub-dividable into a Core (Egypt and Syria), Domain (Morocco and Yemen) and Sphere (Pakistan and Turkey, which could also be considered double periphery areas depending upon which out-groups are considered). In addition to a relatively uncomplicated culture area morphology, Islamic medicine and its antecedents have a long-recorded history, which allowed for diachronic as well as synchronic comparisons. A number of the pronouncements made by Middle Easternists regarding the history of medical influences in the area have merely been subjective inferences based on piecemeal and incidental bits of data or vague “impressions” (e.g., Budge 1913, Breasted 1930, Crum 1950, and Nunn 1996). An objective quantification of similarity through dialectometric methods promised to bring clarification to several ambiguous and contested relationships (such as the relation of Coptic medicine to the ancient Egyptian medical system, and whether Greek humoral categories were simply “lifeless theory” to Medieval Arab physicians or something more practical). Ultimately, this project addressed two problems. The first was the need for an objective means to measure similarity between cognitive systems that are likely to be primarily based on tendencies toward similarity in practice rather than on hard and fast categories of explicit “logical” classification. The second was the necessity for a clearer understanding of the inter-relationships among modern and pre-modern Middle Eastern ethnopharmacological systems.

191 Procedures for Measuring Cognitive Dialects

In order to apply the principles of dialectometry to cognitive structure in a second-generation perspective, it was necessary to develop a set of procedures tuned to the special problem of deriving quantifiable and comparable “attributes” from tendency-based distributional similarity judgments. This study considered a total of fourteen data points (i.e., sets of prescriptions or “native” descriptions of medicinal activities/properties of drug-plants). One component of the sample consisted of seven modern (1974 to present) sources from Aleppo, Syria; Cairo, Egypt; Gazintep, Turkey; India; Marrakech, Morocco; and Sanaa, Yemen. The other component of the sample was made up of seven pre-modern (1534 B.C. to 1075 A.D.) sources including one ancient Egyptian papyrus (P. Ebers), a Coptic papyrus (P. Chassinat), two portions of a Syriac medical manuscript (containing both “Alexandrian” and “native” sections), two medieval Arabic sources (al-Kindi and ibn Wafid), and a modern handbook of an eighteenth-century abridgement of an eleventh-century medical classic (the Qānūn of Avicenna). The procedure applied in this project involved a number of multi- component steps, which may be summarized briefly as follows: 1. Selection of Sources: A so-called “purposeful sample” was established, where each source selected provided unique information that could not be obtained elsewhere. In addition, plant terms in each source were identified with terms found in other sources (or with Linnean taxa). 2. Selection of Variables: This step involved the selection of plants that could be identified as appearing in more than one source, the selection of recipes that included two or more identifiable drug-plants, and the derivation of groupings of drug-plants from their distributional similarity in a given source. Groupings of drug-plants in each source were derived by applying a hierarchical clustering program to a data matrix reflecting the drug-plant prescription/attribute correlations of each source. 3. Comparison of Sources: For each individual source, the last two or three “joins” of the distributional similarity cluster tree were treated as “pile sort” results. Shared groupings of two or more plants across sources were tallied in a “source by source” data matrix and treated as shared characters. Various means and functions of similarity (the arithmetic mean, the geometric mean, and the arithmetic and geometric 192 means of the “usability function”) were calculated based on the sharing of groupings among sources. These degrees of overall similarity among sources were then modeled using hierarchical clustering, multidimensional scaling, K-means clustering and factor analysis techniques. 4. Evaluation of Dialectometric Results: Evaluation of the results included a holistic appraisal of how well the various models of overall similarity among sources matched the culture area morphology of the modern sources and the presumed relationships of descent and influence among modern Core and pre-modern sources. It also entailed particularistic evaluations of how well the similarity scores reflected known and presumed primary influences and interrelationships among sources.

Research Hypotheses Considered

The two specific research hypotheses considered in this project were as follows: Specific Research Hypothesis 1. Sources in the Core of the culture area will be shown, through numerical taxonomy, to be more similar to one another than to sources on the periphery. Peripherally located sources will be shown to be more similar to Core sources and immediate neighbor sources than they are to non-neighbor peripheral sources. Exceptions will generally conform to known influences not subsumed by the culture area model (e.g., the influence of trading relationships, or membership in a “double periphery zone,”). Specific Research Hypothesis 2. Where there are sources from three or more successive periods of an intellectual tradition, numerical taxonomy will show that the earliest and the latest sources are more similar to the middle sources than they are to one another.

Results and Conclusions

This section of the chapter includes an overview of the results of testing the research hypotheses by means of the procedures described above, posits some conclusions stemming from the results and offers some preliminary interpretations of these conclusions.

193 Major Findings

Both the general research hypotheses and their specific corollaries were confirmed. A numerical taxonomy approach to Islamic medicine showed a clear relationship between the proximity and shared history of contemporaneous localities and their overall degree of similarity in drug- plant prescription practices. Sources representing the Core of the culture area morphology were more similar to one another than they were to peripheral sources, and peripheral sources were more similar to Core sources or immediate neighbors than they were to non-neighbor peripheral sources. In addition, numerical taxonomy showed the degree of overall similarity between sources from different time periods to correlate with the relative strength of their presumed relationships of descent and influence. The earliest and latest sources were shown to be more like the middle sources than they were to those on the opposite end of the temporal continuum. Specifics of the rankings of interrelationships among sources were reviewed in detail in the previous chapter, which should be consulted for clarification regarding the following conclusions.

Conclusions and Interpretation

Aleppo, in Syria and Cairo, in Egypt, are the most strongly interrelated modern locales. The informed reader will not be surprised that these members of the Core region should be so similar. They likely share a common genetic heritage closer than that of any other pair of sources considered in this sample and have had continuing high-level interactions in their roles as major centers of Islamic civilization. It should be recalled that the maristan ‘hospital’ of Cairo was built on a Damescene (Syrian) model, and that these cities were major points on the same trading networks and both were supreme centers of Islamic medical learning for at least five hundred years. Also, it should be noted that the Arabic Koine of the Islamic Imperium was developed in Aleppo and Damascus before spreading to Egypt and beyond. Based on their overall similarity ratings, Marrakech, in Morocco, and Sanaa, in Yemen, closely orbit the Core of the culture area (represented by Aleppo and Cairo in this sample). These sites represent the Proximal Domain/Distal Core of the region. Together with the Core, they create a closed circuit circumscribing the heart of the Islamic culture area. Again, this comes as no surprise. All four of the locales are Arabic-speaking, and 194 Marrakech and Sanaa are most strongly linked to the rest of the Islamic world via trade and pilgrimage routes centered on Egypt and Syria (see Figures 2.3- 2.7). Karachi, in Pakistan, and India are also strongly linked locales. Pakistan had its genesis in the Sind, Baluchistan, Northwest Frontier, Punjab, and Bengal provinces of India (Esposito 2002:644). Its primary language, Urdu, is mutually intelligible with Hindi (Ethnologue 2005). It comes as no surprise that the two sources should show such strong commonalities. Pakistan is also influenced, though slightly more weakly, by Syria. The double-periphery membership of Pakistan is clearly displayed in its relations with both India and Syria. Although the author of the Pakistani recipes used in this sample claims to be practicing Tibb (Arab Islamic medicine), his practice is more strongly influenced by Ayurvedic patterns than by traditional Islamic ones. This suggests a synthesis of the two, a kind of cognitive creolization or blending of the two traditions. Turkey is only weakly related to the other sources under consideration and should therefore be considered something of a cognitive isolate (by analogy with linguistic isolates). It is ranked last in similarity from the perspective of all sources except Aleppo and Cairo (for both of which it ranks second to last ahead of Karachi). For historians who find this result objectionable on the grounds that it seems to contradict the long-term importance of the Turkish-based Ottoman Empire, it should be recognized that much of Ottoman Islamic civilization involved the adoption and promulgation of Arabic cultural forms from the Core. In addition, although Turkish scholars used both Persian and Arabic as literary languages, the Turkish language (genetically unrelated to either Arabic or Persian) was always the primary means of communication among the Ottomans and later Turks. Objection might also be raised in regard to the relegation of Yemen to the Domain rather than the Core of the culture area. However, the data considered in this study came from North Yemen, a region geographically isolated from the rest of the Arab-Islamic world for much of its history, where tribal structures and religious sectarianism severely limited its contact with outsiders (Spencer 2000:161). The out-group, India, was most strongly linked to Aleppo, Syria. Syria is the meeting place of the Asian overland trade route and the Persian Gulf sea route from India (see figures 2.3-2.5), so it is only logical that it should be the next in similarity to India after Karachi. In addition, both

195 Syria and India have been historically dominated by the Hanafi school of Muslim law (see figure 2.8) and are thereby linked in intellectual spirit. The fact that Syria’s relationship with Pakistan is weaker than its relationship with India proper implies that the direct Arabian Sea route terminating in the Persian Gulf has been a more dominant influence than the overland route. Figure 5.1 shows a descriptive and inferentially historical scheme of interrelationships among sources, in nearly their true geographic positions (some distances have been slightly elongated or truncated for ease of representation), based on the rankings and clusters reported in the previous chapter. Locales are connected by arrows according to the leading interregional influences indicated by the numerical values obtained. The strongest relationship is shown by a heavily weighted arrow. Reciprocal relationship is represented with a double arrow. Less bold arrows represent relations of secondary strength, while broken-line arrows indicate relations of tertiary strength. The direction of arrows is primarily determined by the level of significance the other member of a pair has in relation to the first member. Influences are not necessarily entirely in the directions indicated. Rather, the prevailing trend is represented by the direction of the arrow. In sum, the figure represents the presumable historical derivations of localized traditions, as revealed in the correlations of the numerical taxonomy for the “modern” sources considered in this project. The conclusions that may be drawn from the results of comparing the pre-modern (medieval and ancient) sources with modern Cairo and Aleppo resolve some ongoing debates in the study of the history of medicine in the Middle Eastern ecumene. Based on the data considered and the results obtained in chapter 4, it seems that any survivals of Ancient Egyptian, Coptic, or native Syriac medical systems are so picayune that they have little or no influence on the practices of modern practitioners in the territories where these traditions once held sway, relative to the influence of the Galenic system. This flatly contradicts Crum’s assertions to the contrary regarding Coptic (1950). In fact, these pre-Arabic systems, as represented by the pre-modern sources, are shown to be the least like the modern Egyptian and Syrian sources. Contrary to the estimation of Waugh (1995:417) and others (e.g., Nunn 1996:209), Coptic patterns of prescription do not reflect the survival of older Egyptian tendencies to any significant degree. Instead, the P.

196 Gaziantep, Turkey SPHERE

Aleppo, DOMAIN Syria Karachi, Pakistan

CORE Marrakech, Cairo, Morocco Egypt INDIA

Sanaa, Yemen

Figure 5.1. Principal ethnopharmacological influences as reconstructed from values in Table 1.

197 Chassinat is very like the Greek-inspired sources of the Abbasid period. It is undeniable that indigenous and Greco-Arabic systems mixed. The relatively late P. London and Leiden, though primarily in Demotic Egyptian and formatted on the model of older Egyptian medical texts, includes extensive borrowings from Greek language and medicine. A few hundred years later, the P. Chassinat includes numerous Greek and Arabic loanwords. According to Mühlhäusler, mixed languages engaging in excessive borrowing are only one step away from “structural collapse” (1985:76-77). It is likely that cognitive structures undergo the same type of catastrophic change and that both the Egyptian and native Syriac medical systems collapsed under the weight of an overpowering influx of Greco-Arabic ethnopharmacological memes. Based on the dissimilar relationships prevailing between the P. Ebers and all other sources, we can conclude that contact with Egyptian medicine does not even remotely begin to explain the impetus behind the flowering of Greek medicine at Alexandria (contrary to Budge 1913:cxxx, Breasted 1930:16, Estes 1996:93, and Ghalioungui 1973:16 ff., among others). Although there might be a few borrowed elements, Egyptian memeplexes, if they ever had any real influence on Greek thought, seem to have been completely overwhelmed and obscured by a much stronger non-Egyptian signal. According to Ullman, the foremost historian of Islamic medicine, Galenic humoral criteria were simply “lifeless theory” to the Arabs (1978:106). The results of this study, however, suggest something else. The summary of ibn Wafid given in al-Sulamī’s examination for physicians, a medieval source, lists the humoral qualities of over 150 plants. The results of a hierarchical clustering based on these properties most closely resemble the results obtained by clustering plants prescribed in some 137 recipes by al-Kindi, a near contemporary of ibn Wafid. This demonstrates that al- Kindi’s patterns of prescription must follow a logic nearly identical to that of ibn Wafid’s humoral classification. The two are the most strongly related sources of all those considered in this project, both ancient and modern. In addition, the similarity of modern sources (i.e., Cairo and Aleppo) to these two medieval treatises implies that residues of the no-longer-explicit moist- dry distinction may be exerting something of an implicit substratal influence on the modern sources. Despite the fact that the modern sources are most closely related to the medieval Arabic ones, Cairo and Aleppo are nearly as dissimilar in relation to the medieval tradition as the ancient Egyptian, Coptic, and native Syriac sources are to the medieval Arabic sources. This seems to

198 confirm Bürgel’s assertion that the high tradition of early Arabic medicine (inherited from the Greeks via Syriac translations made at Gondeshapur) went through successive periods of steady decline and decay, ultimately terminating in a long, asthenic “senility” (1976:53, 60). Modern Cairo and Aleppo have what might be characterized as a distant relationship with Galenic medicine, but the genetic relationship is still clearly present in the fact that the two modern sources share their strongest relationships with Greek-influenced sources rather than with extinct native traditions from the same locale. In tandem with the observation that areal influences (like that prevailing between Karachi and India) are also evident in the data, we are led to conclude that cognitive mixing and creolization may be at work in the development of cognitive systems and their expression in practice.

Implications

The conclusions regarding Islamic medical ethnobotany that have been presented above have several ramifications for theory and methodology. The theoretical implications identified center on our understanding of the nature of cognitive structures situated in time and space and on aspects of Middle Eastern medical history. The methodological implications to be discussed focus on the utility of linking dialectometry with cognitive analysis.

Theory

From a theoretical perspective, this project makes three main contributions. First and foremost, it substantiates the existence and measurability of “cognitive dialects,” on analogy with linguistic lects. Cognitive dialects have been demonstrated to show tendencies in variation, interaction, and transmission that are similar to those present in linguistic data. Second, the reality of cognitive dialects that influence, mix with, and transform one another (via both vertical and horizontal memetic transmission) implies the possibility of cognitive creolization. “Cognitive creoles” are measurable syncretic mixtures of multiple cognitive systems. At this stage, we have only just become aware of the existence (or at least, the measurability) of these newly described entities. Theory regarding cognitive dialects and their creolization may initially be modeled on linguistic theory, but will require a great deal of further elaboration that can only be founded on additional work in this area. Third, in regard to the history of

199 pharmacy in the Middle East, it will be necessary to reconsider some of our long-cherished romantic hopes for discovering modern-day intellectual relics from antiquity. Perhaps the time has finally come to mourn their passing and to realize that they will not have died in vain if we can learn about the how, why, and when of their demise and apply what we learn from them to the preservation of still-living cognitive systems in contact.

Methodology

This project has two main implications in terms of research methodology. First, cognitive dialectometry is an effective means of measuring interrelationships between cognitive systems viewed in the second- generation paradigm. However, its apparatus will surely require a great deal of refinement as others take up the technique. Perhaps the most urgent issue would be the development of automated software specifically designed to process larger corpuses of similarity data in a numerical taxonomy format. A program that could automatically identify shared groupings across multiple sources or subjects (perhaps by automatically placing their contents into “sort tables” a la Agar 1996:190-195) would go a long way toward advancing the program of cognitive dialectometry. The ideal computer program would then proceed to calculate the degree of similarity between sources based on the sharing of these groups. Second, it will be necessary to devise new techniques constructed on the framework provided herein. At present, our ability to access historical evidence for cognitive relationships and their development is essentially at the same level where historical linguistics was in 1786, when Sir William Jones observed that Greek, Latin and Sanskrit must have “sprung from some common source.” The ability to identify phenetic similarity by inspection is still a far cry from phylogeny and reconstruction. However, it is a first step.

Recommendations for Future Research

There are several areas where further study might extend or clarify the results and conclusions of this initial trial of cognitive dialectometry. Both broader and narrower studies could be enlightening. For example, it would be helpful to expand the present data set by gathering recipes from other major centers in the Islamic world which are not represented in already published materials (e.g., Algiers, Baghdad, Kabul, Khartoum, Teheran, Tripoli, and Tunis). Another option might be a more fine-grained

200 investigation of cognitive dialects, within a tightly constrained area and based on multiple data points (for example, an analysis of patterns of drug- plant prescription among herbalists and nonherbalists in various provinces of Egypt). Such a project might reveal interesting isoglosses and lines of influence and contrast on a more localized scale. A study along these lines could be conducted by teams of fieldworkers sent out to collect pile sort results from multiple consultants in a large number of geographically contiguous cities, towns, and villages. On the opposite end of the scale, it would be interesting to test a sample comparable in size to that used in this study, but from a different culture area, using the same measures and instruments employed in this study. An analysis of interrelationships prevailing in the Inner Asian sphere would be one possibility, which specialists in that area’s languages and history might productively undertake. Such an endeavor would further confirm the efficacy of the methods used herein. Other cultural domains besides medicinal herbalism might be considered through an analysis of folklore or narratives. Domains for consideration might include animals, natural phenomena (weather, terrain), supernatural beings, etc. For example, those named supernatural beings (gods, spirits), that often appear together in episodes of story or myth, in incantations, or in segments of ritual, would be reckoned as having a higher distributional similarity (e.g., Artemis and Apollo in Greek myth). If neighboring groups can identify the names that identify “the same” personage in a neighboring group (for example, Zeus with Jupiter in Greek and Roman mythology, or the Aztec Quetzalcoatl with the Mayan Kukulkan), then it would be possible to use this method to undertake an exploration of cognitive dialects of myth.

Retrospect

This project represents a first step in the development of a method for cognitive dialectometry and its application to historical questions of cultural continuity and change. Such a method provides us with a new tool for arriving at a deeper, more nuanced understanding of the intricacies of culture change, syncretism, and the retention of traditional knowledge in a rapidly globalizing world. The results of applying the principles of numerical taxonomy to patterns of drug-plant prescription in Islamic societies have helped to elucidate important aspects of the transmission of

201 ancient and medieval medicinal plant knowledge over the long-term and into the present. The history of any memeplex, such as Islamic ethnomedicine, is the product of its coadapted ideas acting on the minds of people whose lives are embedded in webs of meaning, each of which in its turn impinges upon the networks of other thought worlds. These intersecting, interacting worldviews must ultimately accommodate, dominate, or submit to one another. In the final analysis, the similarities and interrelationships observed in this study are not merely the accidents of history, they are the stuff of which medical history is made.

202 BIBLIOGRAPHY

Agar, Michael H. 1996 The Professional Stranger. San Diego: Academic Press.

Ahmed, M. Salah, Gisho Honda, and Wataru Miki 1979 Herb Drugs and Herbalists in the Middle East. Tokyo: Institute for the Study of Languages and Cultures of Asia and Africa.

Antes, Peter 1989 “Medicine and the Living Tradition of Islam,” in Healing and Restoring Health: Health and Medicine in the World’s Traditions, ed. by Lawrence E. Sullivan. Pp. 173-202. New York: Macmillan.

Aunger, Robert 2002 The Electric Meme: A New Theory of How We Think. New York: Free Press.

Auerbach, Lisa S. 1982 “Childbirth in Tunisia: Implications of a Decision-Making Model,” Social Science & Medicine 16:1499-1506.

Bailey, Charles-James N. 1996 Essays on Time-Based Linguistics. Oxford: Clarendon Press.

_____ 1978 Variation and Linguistic Theory. Arlington: Center for Applied Lingusitics.

Bakker, Jogian 1992 “The Rise of Female Healers in the Middle Atlas, Morocco” Social Science & Medicine 35:819-29. bar-Sela, Ariel, Hebbel E. Hoff, and Elias Faris 1964 Moses Maimonides’ Two treatises on the Regimen of Health. Philadelphia: American Philosophical Society.

Başer, K.H.C., Gisho Honda, and Wataru Miki 1986 Herb Drugs and Herbalists in Turkey. Tokyo: Institute for the Study of Languages and Cultures of Asia and Africa.

Bates, Daniel G., and Amal Rassam 2001 Peoples and Cultures of the Middle East, second edition. Upper Saddle River: Prentice Hall.

Bateson, Mary Catherine 1967 Arabic Language Handbook. Washington, D.C.: Center for Applied Lingusitics.

203 Beals, Ralph L., and Harry Hoijer 1958 An Introduction to Anthropology. New York: The MacMillan Company.

Bellakhdar, Jamal, Gisho Honda, and Wataru Miki 1982 Herb Drugs and Herbalists in the Maghrib. Tokyo: Institute for the Study of Languages and Cultures of Asia and Africa.

Berlin, Brent 1978 “Ethnobiological Classification,” in Cognition and Categorization, ed. by E. Rosch and B. Lloyd. Pp. 11-27. Hillsdale: Lawrence Erlbaum Associates, Ltd..

_____ 1992 Ethnobiological Classification: Principles of Categorization of Plants and Animals in Traditional Societies. Princeton: Princeton University Press.

Berlin, Eloise A., and Brent Berlin 1996 Medical Ethnobiology of the Highland Maya of Chiapas, Mexico: the Gastrointestinal Diseases. Princeton: Princeton University Press.

Berlin, Brent, Dennis E. Breedlove, and Peter H. Raven 1968 “Covert Categories and Folk Taxonomies” American Anthropologist 70:290-299.

_____ 1974 Principles of Tzeltal Plant Classification. New York: Academic Press.

Berlin, Brent, and Paul Kay 1969 Basic Color Terms. Berkeley: University of California Press.

Blackmore, Susan 1999 The Meme Machine. Oxford: Oxford University Press.

Blackwelder, R.E. 1964 “Phyletic and Phenetic versus Omnispective Classification” in Phenetic and Phylogenetic Classification, edited by Vernon Heywood and J. McNeill. Pp. 17-28. London: The Systematics Association.

Bloch, Maurice 1990 How We Think They Think: Anthropological Approaches to Cognition, Memory, and Literacy. Boulder: Westview Press.

Bloomfield, Leonard 1933 Language. New York: Henry Holt and Company.

Boas, Franz 1988 (1896) “The Limitations of the Comparative Method of Anthropology,” in High Points in Anthropology, edited by Paul Bohannan and Mark Glazer. Pp. 85-92. New York: McGraw-Hill, Inc.

Boulos, Loutfy 1983 Medicinal Plants of North Africa. Algonac: Reference Publications.

204 Bourdieu, Pierre 1990 The Logic of Practice. Cambridge: Polity Press.

Breasted, James H. 1930 The Edwin Smith Surgical Papyrus. Chicago: The University of Chicago Press.

Bright, Jane O., and William Bright 1965 “Semantic Structures in Northwestern California and the Sapir- Whorf Hypothesis,” in American Anthropologist 67:249-258. Reprinted in Tyler 1969, pp. 66-77.

Browne, E.G. 1962 Arabian Medicine. Cambridge: Cambridge University Press.

Budge, E.A. Wallis 1913 Syrian Anatomy Pathology and Therapeutics or ‘The Book of Medicines”: The Syriac Text, Edited from a Rare Manuscript, with an English Translation, etc. London: Oxford University Press.

_____ 1996 (1928) The Divine Origin of the Craft of the Herbalist. New York: Dover Publications, Inc.

Bürgel, J. Christoph 1976 “Secular and Religious Features of Medieval Islamic Medicine” in Asian Medical Systems, edited by Charles Leslie. Pp. 44-62. Berkeley: University of California Press.

Burton, Michael 1972 “Semantic Dimensions of Occupation Names,” in Multi-Dimensional Scaling, Volume II., edited by A.K. Romney, R. Shepard, and S.B. Nerlove. Pp. 55-72. New York: Seminar Press.

Campbell, Lyle 1999 Historical Linguistics: An Introduction. Cambridge: The MIT Press.

Canfield, Robert L., ed. 1991 “Preface,” in Turko-Persia in Historical Perspective. Pp. i-ix. Cambridge: Cambridge University Press.

Cartwright, John 2000 Evolution and Human Behavior. Cambridge, The MIT Press.

Castleman, Michael 1995 The Healing Herbs. New York: Bantam Books.

Central Intelligence Agency 1990 World Factbook. Washington, D.C.: Central Intelligence Agency.

_____ 2005 World Factbook. Washington, D.C.: Central Intelligence Agency.

205 Chassinat, Emile Gaston 1921 Un papyrus médical copte. Cairo: Impr. de l’Institute française d’archéologie orientale. Chisti, G.M. 1991 The Traditional Healer’s Handbook: A Classic Guide to the Medicine of Avicenna. Rochester: Healing Arts Press.

Conklin, Harold 1962 “Lexicographical Treatment of Folk Categories,” International Journal of American Linguistics 28(2):119-141, reprinted in Tyler 1969:41-59.

Constantinides, Pamela I. 1977 “Ill at Ease and Sick at Heart: Symbolic Behavior in a Sudanese Healing Cult” in Symbols and Sentiments: Cross Cultural Studies in Symbolism, ed. by Ioan Lewis. Pp. 61-84. London: Academic Press.

Constantinides, Pamela I. 1985 “Women Heal Women: Spiritual Possession and Sexual Segregation in a Muslim Society,” Social Science and Medicine 21:685-92.

Croft, William 2000 Explaining Language Change: An Evolutionary Approach. Harlow: Pearson Education.

Crowley, Terry 1996 An Introduction to Historical Linguistics. Aukland: Oxford University Press.

Crum, Walter E. 1950 A Coptic Dictionary. Oxford: Clarendon Press.

D’Andrade, Roy 1995 The Development of Cognitive Anthropology. Cambridge: Cambridge University Press.

D’Andrade, Roy, Naomi R. Quinn, Sara Beth Nerlove, and A. Kimball Romney 1972 “Categories of Disease in American-English and Mexican-Spanish,” Multidimensional Scaling, vol. II, edited by A.K. Romney and S.B. Nerlove. New York: Seminar Press.

Darwin, Charles 1958 (1859) Origin of Species. New York: New American Library, Mentor.

Davidson, Basil 1991 Africa in History. New York: Simon and Schuster.

Dawkins, Richard 1976 The Selfish Gene. Oxford: Oxford University Press.

_____ 1989 The Selfish Gene, 2nd edition. Oxford: Oxford University Press.

_____ 1999 The Extended Phenotype: The Long Reach of the Gene. Oxford: Oxford University Press.

206 Dawood, N.J., editor (Franz Rosenthal, translator; Ibn Khaldûn) 1989 The Muqaddimah: An Introduction to History. Princeton: Princeton University Press.

Dixon R.M.W. 1997 The Rise and Fall of Languages. Cambridge: Cambridge University Press.

Douglas, Mary 1973 Natural Symbols. New York: Vintage Books.

Driver, H.E., and A. L. Kroeber 1932 “Quantitative Expression of Cultural Relationships,” University of California Publications in American Archaeology and Ethnology 31(4):211-256.

Dunn, Ross E. 1989 The Adventures of Ibn Battuta: A Muslim Traveler of the 14th Century. Berkeley: University of California Press.

Dyen, Isidore 1975 Linguistic Subgrouping and Lexicostatistics. The Hague: Mouton.

Early, Evelyn 1982 “The Logic of Well Being: Therapeutic Narratives in Cairo, Egypt,” Social Science & Medicine 16:1491-97.

_____ 1988 “The Baladi Curative System of Cairo, Egypt,” Culture, Medicine, and Psychiatry 12:65-83.

_____ 1993 “Fertility and Fate: Medical Practices Among Baladi Women of Cairo” in Everyday Life in the Muslim Middle East, ed. by Donna L. Bown and Evelyn A. Early. Pp. 103-108. Bloomington: Indiana University Press.

Eickelman, Dale F. 2002 The Middle East and Central Asia: An Anthropological Approach, fourth edition. Upper Saddle River: Prentice Hall.

Elliot, Dale, Stanley Legum, and Sandra A. Thompson 1969 “Syntactic Variation as Linguistic Data,” Papers from the Fifth Regional Meeting, ed. by R. I. Binnick, A. Davidson, G.M. Green, and J.L. Morgan. Pp. 52-59. Chicago: Chicago Linguistic Society.

Embleton, Sheila 1993 “Multidimensional Scaling as a Dialectometrical Technique: Outline of a Research Project,” Contributions to Quantitative Linguistics, edited by Reinhard Köhler and Brughard B. Rieger. Pp. 267-276. Dordrecht: Klewer Academic Publishers.

Esposito, John L. 1999 The Oxford History of Islam. Oxford: Oxford University Press.

207 Estes, J. Worth 1989 The Medical Skills of Ancient Egypt. Canton: Science History Publications.

Fabrega, H., Jr. 1970 “On the Specificity Folk Illnesses,” Southwestern Journal of Anthropology 26:305-314.

Fakhouri, Hani 1968 “The Zar Cult in an Egyptian Village,” Anthropological Quarterly 41:49-56.

Federal Research Division of the Library of Congress 1988-1998 Country Studies/Area Handbook. Washington: GPO for the Library of Congress.

Fillenbaum, S., and A. Rapoport 1971 Structures in the Subjective Lexicon. New York: Academic Press.

Finkler, Kaja 1984 “The Nonsharing of Medical Knowledge among Spiritualist Healers and Their Patients: A Contribution to the Study of Intra-Cultural Diversity and Practitioner-Patient Relationship,” Medical Anthropology Summer:195-209.

Foley, William A. 1997 Anthropological Linguistics: An Introduction. Malden: Blackwell Publishers.

Frake, Charles O. 1961 “The Diagnosis of Disease among the Subanun of Mindanao,” American Anthropologist 63:113-132. Reprinted in Hymes 1964, pp. 192-211.

_____ 1962 “The Ethnographic Study of Cognitive Systems” in Anthropology and Human Behavior, edited by T. Gladwin and W. Sturtevant. Pp. 73- 85 Washington, D.C.: Anthropological Society of Washington. Reprinted in Tyler 1969, pp. 28-41.

_____ 1964 “Notes on Queries in Ethnography,” American Anthropologist 66(3):132-145.

Frazer, James G. 1922 The Golden Bough. New York: Macmillan Publishing Company.

Friedberg, Claudine 1979 “Socially Significant Plant Species and Their Taxonomic Position among the Bunaq of Central Timor,” in Classifications in Their Social Contexts, edited by Roy F. Ellen and R. David Rason. Pp. 81-100. Academic Press: New York.

Friedrich, Paul 1970 Proto-Indo-European Trees. Chicago: University of Chicago Press.

208 Frisch, Jack A., and Noel W. Schutz, Jr. 1967 “Componential Analysis and Semantic Reconstructions: the Proto- Central Yuman Kinship System,” Ethnology 6(3):272-293.

Gardiner, Alan 1957 Egyptian Grammar. London: Oxford University Press.

Garro, Linda C. 1986 “Intracultural Variation in Folk Medical Knowledge: A Comparison Between Curers and Noncurers” in American Anthropologist 88:351- 370.

Gee, Henry 1996 In Search of Deep Time: Beyond the Fossil Record to a New History of Life. New York: Free Press.

Ghalioungui, P. 1973 The House of Life. Per Ankh: Magic and Medical Science in Ancient Egypt. Amsterdam: B. M. Israël.

Ghazanfar, Shahina A. 1994 Handbook of Arabian Medicinal Plants. Boca Raton: CRC Press.

Glinert, Lewis 1993 Hebrew in Ashkenaz: A Language in Exile. New York: Oxford University Press.

Goebl, Hans 1991 “Dialectometry: A Short Overview of the Principles and Practice of Quantitative Classification of Linguistic Data,” in Contributions to Quantitative Linguistics, edited by Reinhard Köhler and Brughard B. Rieger. Pp. 277-318. Dordrecht: Klewer Academic Publishers.

Good, Byron J. 1977 “The Heart of What’s the Matter: The Semantics of Illness in Iran,” Culture, Medicine, & Psychiatry 1:25-58.

_____ 1994 Medicine, Rationality, and Experience: An Anthropological Perspective. Cambridge: Cambridge University Press.

Good, Mary-Jo DelVecchio 1980 “Of Blood and Babies: The Relationship of Popular Islamic Thought to Fertility,” Social Science and Medicine 14B:147-156.

Gordon, Raymond G., Jr., editor 2005 Ethnologue: Languages of the World, Fifteenth Edition. Dallas: SIL International.

Gran, Peter 1979 “Medical Pluralism in Arab and Egyptian History,” Social Science and Medicine 13B:339-348.

209 Greenberg, Joseph H. 1974 “The Languages of Africa” International Journal of American Linguistics 29(1):1-177.

Greenwood, Bernard 1981 “Perceiving Systems: Cold or Spirits? Choice and Ambiguity in Morocco’s Pluralistic Medical System” Social Science and Medicine 15B:219-35.

Gunther, Robert T. 1959 The Greek Herbal of Dioscorides. New York: Haffner Publishing Co.

Hall, Robert A., Jr. 1966a “How Pidgin English Has Evolved,” New Scientist 9:413-415.

_____ 1966b Pidgin and Creole Languages, Ithaca: Cornell University.

Harris, Marvin 1968 The Rise of Anthropological Theory. New York: Harper and Row, Publishers, Inc.

Heider, Eleanor (Rosch) 1971 “Focal Color Areas and the Development of Color Names,” Developmental Psychology 4:447-455.

Held, Colbert C. 2006 Middle east Patterns: Places, Peoples, and Politics. Boulder: Westview Press.

Hennig, W. 1965 “Phylogenetic Systematics,” Annual Review of Entomology 10:97- 116.

1966 Phylogenetic Systematics. Urbana: University of Illinois Press.

Heywood, Vernon H., and J. McNeill 1964 “Preface” in Phenetic and Phylogenetic Classification, edited by Vernon H. Heywood and J. McNeill. Pp. 1-4. London: The Systematics Association.

Hill, Christopher, and Peter R. Crane 1982 “Evolutionary Cladistics and the Origin of Angiosperms,” Problems of Phylogenetic Reconstruction, edited by K. A. Joysey and A.E. Friday. Pp. 269-362. London: Academic Press.

Hockett, Charles F. 1964 “The Proto Central Algonquian Kinship System,” in Explorations in Cultural Anthropology, edited by Ward. H. Goodenough. Pp. ^^-^^. New York: McGraw-Hill Book Company.

Hoeningswald, Henry M., and Linda F. Weiner 1987 Biological Metaphor and Cladistic Classification: An Interdisciplinary Perspective. Philadelphia: University of Pennsylvania Press.

210 Honda, Gisho, Miki Wataru and Matsuko Saito 1990 Herb Drugs and Herbalists in Syria and North Yemen. Tokyo: Institute for the Study of Languages and Cultures of Asia and Africa.

Hopkins, Nicholas A. 1978 “Amuzgo and general Mesoamerican Ethnobotanical Nomenclature.” Paper presented to the XVIIth Conference on American Indian Languages: Otomanguean and Tupian, American Anthropological Association, 77th Annual Meeting, Los Angeles, California, November 14-19, 1978.

_____ 1980 “Amuzgo Ethnobotanical Structure and Terminology”. Presented to the symposium on “The Lexicon,” American Anthropological Association, 79th Annual Meeting, Washington, D.C., November, 1980.

Hunn, Eugene 1985 “The Utilitarian Factor in Folk Biological Classification,” in Directions in Cognitive Anthropology, edited by Janet W.D. Dougherty. Pp. 117-140. Urbana: University of Illinois Press.

Hunn, Eugene, and David H. French 2000 “Alternatives to Taxonomic Hierarchy: The Sahaptin Case,” in Ethnobotany, edited by Paul Minnis. Pp. 118-142. Norman: University of Oklahoma Press.

Hymes, Dell, editor 1964 Language in Culture and Society. New York: Harper & Row, Publishers.

Inhorn, Marcia Claire 1994 Quest for Conception: Gender, Infertility, and Egyptian Medical Traditions. Philadelphia: University of Pennsylvania Press.

Jacobsen, Frode F. 1998 Theories of Sickness among the Hadandowa Beja of Sudan. London: Kegan and Paul International.

Jankowsky, Kurt R. 1973 The Neogrammarians: A Re-Evaluation of their Place in the Development of Linguistic Science. The Hague: Mouton. al-Jawziyya, ibn Qayyim 1998 Medicine of the Prophet. Penelope Johnstone, translator. Cambridge: the Islamic Texts Society.

Jing-Feng, Cai 1996 “Traditional Chinese Medicine Today,” in Oriental Medicine, edited by Jan Van Alphen and Anthony Aris. Pp. 197-202. Boston: Shambhala Publications.

Kangas, Beth 1994 “Introduction to Ethnomedicine: Examples from Yemen,” Yemen Update 35:22-27. www.al-bab.com/yemen/soc/soc.htm

211 Kapoor, L.D. 1990 CRC Handbook of Ayurvedic Medicinal Plants. Boca Raton: CRC Press.

Kay, Paul 1966 Comment on “Ethnographic Semantics: A Preliminary Survey,” Current Anthropology 7(1):20-23. Reprinted as “Comments on Colby” in Cognitive Anthropology, edited by Stephen Tyler, 78-90.

Kerlinger, Fred N. 1977 Foundations of Behavioral Research.

Kessler, Brett 1995 “Computational Dialectology in Irish Gaelic,” in Proceedings of the Seventh European ACL. Pp. 60-67. Dublin: EACL.

King, Helen 2001 Greek and Roman Medicine. London: Bristol Classical Press.

Kirk, Lorraine, and Michael Burton 1977 “Meaning and Context: A Study of Contextual Shifts in Meaning of Maasai Personality Descriptors, American Ethnologist 4:734-761.

Kitching, Ian J., Peter L. Forey, Christopher J. Humphries, and David M. Williamset 1998 Cladistics: The Theory and Practice of Parsimony Analysis, second edition. Oxford: Oxford University Press.

Kronenfeld, David B., James D. Armstrong, and Stan Wilmoth 1985 “Exploring the Internal Structure of Linguistic Categories: An Extensionist Semantic View,” in Directions in Cognitive Anthropology, edited by Janet W. Dougherty. Pp. 91-110. Urbana: University of Illinois Press, 91-110.

Kuhn, Thomas S. 1970 The Structure of Scientific Revolutions, second edition. Chicago: University of Chicago Press.

Lakoff, George 1986 Women, Fire, and Dangerous Things: What Categories Reveal about the Mind. Chicago: University of Chicago Press.

Lakoff, George, and Mark Johnson 1999 Philosophy in the Flesh: The Embodied Mind and Its Challenge to Western Thought. New York: Basic Books.

Lane, Sandra D. and Marcia I. Millar 1987 “The Hierarchy of Resort Reexamined: Status and Class Differentials as Determinants of Therapy for Eye Disease in the Egyptian Delta,” Urban Anthropology 16:151-182.

Lapidus, Ira M. 2002 A History of Islamic Societies. Cambridge: Cambridge University Press.

212 Lass, Roger 1997 Historical Linguistics and Language Change. Cambridge: Cambridge University Press.

Leiser, Gary, and Noury al-Khaledy, editors and translators 2004 Questions and Answers for Physicians: A Medieval Arabic Study Manual by ‘Abd al- ’Aziz al-Sulamī. Leiden: Brill.

Levi-Strauss, Claude 1966 The Savage Mind. Chicago: University of Chicago Press.

Levey, Martin 1966 The Medical Formulary or Aqrābādhīn of al-Kindī. Madison: University of Wisconsin Press.

Levey, Martin 1967 The Medical Formulary of Al-Samarqandī and the Relation of Early Arabic Simples to Those Found in the Indigenous Medicine of the Near East and India. Philadelphia: University of Pennsylvania Press.

Lounsbury, Floyd G. 1969 “The Structural Analysis of Kinship Semantics,” Cognitive Anthropology, edited by Stephen A. Tyler. Pp. 193-211. New York: Holt, Rinehart and Winston, Inc.

Lynch, Aaron 1996 Thought Contagion: How Beliefs Spread through Society. New York: Basic Books.

Mace, Ruth, and Mark Pagel 1994 “The Comparative Method in Anthropology,” Current Anthropology 35(5):549-564.

Meinig, D. W. 1969 “A Macrogeography of Western Imperialism: Some Morphologies of Moving Frontiers of Political Control,” in Settlement and Encounter, edited by Fay Gale and Graham H. Lawton. Pp. 213-240. Melbourne: Oxford University Press.

Majno, Guido 1975 The Healing Hand: Man and Wound in the Ancient World. Cambridge: Harvard University Press.

Manniche, Lise 1989 An Ancient Egyptian Herbal. Austin: University of Texas Press.

Martin, Gary 1995 Ethnobotany: A Methods Manual. London: Chapman and Hall.

Maxwell, Joseph 1996 Qualitative Research Designs: An Interactive Approach. Thousand Oaks, Sage Publications.

Merrifield, William R. 1981 Proto Otomanguean Kinship. Dallas: International Museum of Cultures.

213

Millar, Marcia Inhorn, and Sandra D. Lane 1988 "Ethno-Opthamology in the Egyptian Delta: An Historical Systems Approach to Ethnomedicine in the Middle East," Social Science & Medicine 26: 651-57.

Metzger, Duane G., and Gerald E. Williams 1966 “Some Procedures and Results in the Study of Native Categories: Tzeltal ‘Firewood,’” in American Anthropologist 68:389-407.

Millar, Marcia I. and Sandra Lane 1988 “Ethno-ophthalmology in the Egyptian delta: an historical systems approach to ethnomedicine in the Middle East,” in Social Science and Medicine 26:651-657.

Morgan, Lewis H. 1877 Ancient Society. New York: Holt.

Morris, Brian 2000 “The Pragmatics of Folk Classification” in Ethnobotany, edited by Paul E. Minnis. Pp. 69-87. University of Norman: Oklahoma Press.

Morsy, Soheir A. 1980a “Body Concepts and Health Care: Illustrations from an Egyptian Village,” Human Organization 39:92-96.

_____ 1980b “Health and Illness as Symbols of Social Differentiation in an Egyptian Village,” Anthropological Quarterly 53:153-161.

_____ 1981 “Childbirth in an Egyptian Village” in Anthropology of Human Birth, edited by Maragarita A. Kay. Pp. Philadelphia: F. A. Davis.

_____ 1993 Gender, Sickness, and Healing in Rural Egypt. Boulder: Westview Press, Inc.

Mossa, Jaber S. 1987 Medicinal Plants of Saudi Arabia. Riyadh: King Saud University Libraries.

Mühlhäusler, Peter 1985 “Patterns of Contact, Mixture, Creation and Nativization: Their Contribution to a General Theory of Language,” in Developmental Mechanisms of Language, edited by Charles-James N. Bailey and Roy Harris. Pp. 51-88. Oxford: Pergamon Press.

Mufwene, Salikoko 2001 The Ecology of Language Evolution. Cambridge: Cambridge University Press.

214 Myntti, Cynthia 1985 “Social, Economic and Cultural Context of Women’s Health and Fertility in Rural Northern Yemen,” in Micro Approaches to Demographic Research, ed. by V. Hill and A. Hill. Pp. 165-171. London: Routledge Paul.

Naylor, Larry L. 1996 Culture and Change: An Introduction. Westport: Bergin and Garvey.

Nerbonne, John, and William Kretzschmar 2003 “Introducing Computational Models in Dialectometry,” in Computational Methods in Dialectometry, edited by John Nerbonne and William Kretzschmar. Special Issue of Computers and the Humanities 37(3):245-255.

Newman, Isadore and Carolyn R. Benz 1998 Qualitative-Quantitative Research Methodology: Exploring the Interactive Continuum. Carbondale: Southern Illinois University Press.

Nunn, John F. 1996 Ancient Egyptian Medicine. Norman: University of Oklahoma Press.

O’Brien, Michael J., and R. Lee Lyman 2000 Applying Evolutionary Archaeology: A Systematic Approach. New York: Kluwer Academic/Plenum Publishers.

_____ 2003 Cladistics and Archaeology. Salt Lake City: The University of Utah Press.

Ochsenwald, William, and Sydney Nettleton Fisher 2004 The Middle East: A History. Boston: McGraw Hill.

Osgood, Charles 1964 “Semantic Differential Technique in the Comparative Study of Cultures,” American Anthropologist 66(3), Part 2:171-200.

Osgood, Charles E., G.J. Suci, and P.H. Tannenbaum 1957 The Measurement of Meaning. Urbana: University of Illinois Press.

Pittle, Kevin D. 2002 Epigraphy and Cognitive Anthropology: Reconstructing Elements of Ancient Egyptian Ethnobotany from Hieroglyphic Text. Poster presented to the session on “Language, Cognition, and Prehistory,” American Anthropological Association, 101st Annual Meeting, Chicago, Illinois, November 2002.

Rahman, A. 2002 “General Features and Areas of Interaction,” in India’s Interaction with China, Central and West Asia, edited by A. Rahman. Pp. 19-44. Oxford: Oxford University Press. pp. 19-44.

215 Randall, Robert 1976 “How Tall is a Taxonomic Tree? Some Evidence of Dwarfism” in American Ethnologist 3(3):545-546.

Randall, Robert, and Eugene S. Hunn 1984 “Do Life-Forms Evolve or Do Uses for Life? Some Doubts about Brown’s Universals Hypothesis,” American Ethnologist, 11(2):329-349.

Reisner, George A. 1905 The Hearst Medical Papyrus. Leipzig: J.C. Hinrichs.

Romney, A. K., and Roy Goodwin D’Andrade 1964 “Cognitive Aspects of English Kin Terms,” American Anthropologist 66(3), Part 2:146-170.

Rosch, Eleanor (see Heider) 1973 “Natural Categories,” Cognitive Psychology 4:328-350.

_____ 1978 “Principles of Categorization,” in Cognition and Categorization, edited by Eleanor Rosch and Barbara Lloyd. Pp. 28-49. Hillsdale: Lawrence Erlbaum Associates, Ltd.

Rosenberg, Seymour, and Andrea Sedlak 1972 “Structural Representations of Perceived Personality Trait Relationships,” in Multidimensional Scaling, Volume II, edited by A.K. Romney, R. Shepard, and S.B. Nerlove. Pp. 134-162. New York: Seminar Press.

Rosner, Fred, editor 1979 Moses Maimonides’ Glossary of Drug Names. Philadelphia: the American Philosophical society.

Rudestam, Kjell Erik and Rae R. Newton 1992 Surviving Your Dissertation: A Comprehensive Guide to Content and Process. Newbury Park: Sage Publications.

Ruhlen, Merritt 1987 A Guide to the World’s Languages, Volume I.: Classification. Stanford: Stanford University Press.

1994 On the Origin of Languages: Studies in Linguistic Taxonomy. Stanford: Stanford University Press.

Sa’id, Hakim Mohammad 1998 “Forward” in Medicine of the Prophet, translated by Penelope Johnstone. Pp. xiii-xv. Cambridge: the Islamic Texts Society.

Séguy, J. 1971 “La Relation entre la Distance Spatiale et la Distance Lexicale,” in Revue de Linguistique Romane 35:335-357.

Shankardass, Mala Kapur 2002 “India’s Interactions in Medical Knowledge and Practice,” in India’s Interaction with China, Central and West Asia, edited by A. Rahman. Pp. 274-296. Oxford: Oxford University Press.

216 Shaul, David L., and N. Louanna Furbee 1998 Language and Culture. Prospect Heights: Waveland Press.

Shiloh, Ailon 1953 “The System of Medicine in the Middle East Culture,” Middle East Journal, 15:277-278.

_____ 1968 “The Interaction between the Middle Eastern and Western Systems of Medicine,” Social Science and Medicine 2:235-248.

Sholkamy, Hania 1998 “Being Sickly or Eating Well: The Conception of Health and Ill- Health in an Upper Egyptian Village,” in Directions of Change in Rural Egypt, edited by Nicholas S. Hopkins and Kristen Westergard. Pp. 203-218. Cairo: The American University in Cairo Press.

Shore, Bradd 1996 Culture in Mind: Cognition, Culture, and the Problem of Meaning. New York: Oxford University Press.

Simpson, G.G. 1945 “The Principles of Classification and a Classification of Mammals,” Bulletin of the American Museum of Natural History 85:1-350.

Sneath, P.H.A. 1964 “Numerical Taxonomy: Introduction” in Phenetic and Phylogenetic Classification, edited by V. H. Heywood and J. McNeill. Pp. 43- 46. London: The Systematics Association.

Sneath, Peter H.A., and Robert R. Sokal 1973 Numerical Taxonomy: The Principles and Practice of Numerical Classification. San Francisco: W.H. Freeman and Company.

Sokal, Robert R. 1986 “Phenetic Taxonomy: Theory and Methods,” Annual Review of Ecology and Systematics 17:423-442.

Sokal, Robert R., and P.H.A. Sneath 1963 Principles of Numerical Taxonomy. San Francisco: W.H. Freeman and Company.

Sonbol, Amira el Azhary 1991 The Creation of a Medical Profession in Egypt, 1800-1922. Syracuse: Syracuse University Press.

Spencer, Herbert 1876 Principles of Sociology. New York: D. Appleton.

Spencer, William 2000 The Middle East. Sluice Dock: Dushkin/McGraw-Hill.

217 Stefflre, Volney 1972 “Some Applications of Multi-dimensional Scaling to Social Science Problems,” in Multidimensional Scaling: Theory and Applications in the Behavioral Sciences, edited by A. K. Romney. Pp. 211-243. New York: Academic Press.

Sturtevant, William C. 1964 “Studies in Ethnoscience,” American Anthropologist 66(3):99- 131.

Thomason, Sarah G. 2001 Language Contact: An Introduction. Washington, D.C.: Georgetown University Press.

Thomason, Sarah G., and Terrance Kauffman 1988 Language Contact, Creolization, and Genetic Linguistics. Berkeley: University of California Press.

Thompson, R. Campbell 1923 Assyrian Medical Texts from the Originals in the British Museum. London: Garrison-Morton.

Tyler, Stephen A. 1969 Cognitive Anthropology. New York: Holt, Rinehart and Winston, Inc.

Tylor, Edward B. 1889 “On a Method of Investigating the Development of Institutions; Applied to Laws of Marriage and Descent,” The Journal of the Royal Anthropological Institute of Great Britain and Ireland 18:245-272.

_____ 1924 (1871) Primitive Culture. New York: Brentano’s Publishers.

Trask, R.L. 1996 Historical Linguistics. London: Arnold.

Ullman, Manfred 1978 Islamic Medicine. Edinburgh: Edinburgh University Press.

Ushmanghani, Khan, Gisho Honda, and Wataru Miki 1986 Herb Drugs and Herbalists in Pakistan. Tokyo: Institute for the Study of Languages and Cultures of Asia and Africa.

Versteegh, Kees 1984 Pidginization and Creolization: The Case of Arabic. Amsterdam: John Benjamins Publishing Company.

_____ 1997 The Arabic Language. New York: Columbia University Press.

Von Deines, H. and H. Grapow 1959 Wörterbuch der ägyptischen Drogennamen. Berlin: Grundriss der Medizin der Alten Agypter.

218 Von Staden, H. 1989 Herophilus: The Art of Medicine in Early Alexandria. Cambridge: Cambridge University Press.

Yablokov, A. V. 1986 Phenetics: Evolution, Population, Trait. New York: Columbia University Press.

Wall, L. Lewis 1988 Hausa Medicine: Illness and Well-Being in a West African Culture. Durham: Duke University Press.

Wallace, Anthony, F. C., and John Atkins 1960 “The Meaning of Kinship Terms,” American Anthropologist 62:58- 80. Reprinted in Congitive Anthropology, edited by Stephen Tyler, 345-369.

Waugh, Richey L. 1995 The Eye and Man in Ancient Egypt. Ostende: JP Wayenborgh.

Weiss, R. S. 1994 Learning from Strangers: The Art and Method of Qualitative Interview Studies. New York: Free Press.

Weller, S.C., A.K. Romney, and D.P. Orr 1987 “The Myth of a Sub-Culture of Corporal Punishment,” Human Organization 46:39-47.

Weller, Susan C., and A.K. Romney 1988 Systematic Data Collection. Newbury Park: Sage Publications.

Von Westendorf, Wolfhart 1999 “Papyrus Ebers,” in Handbuch der Altägyptische Medizin, 2 Band. Pp. 547-710. Leiden: Brill.

White, Geoffrey M. 1985 “‘Bad Ways’ and ‘Bad Talk’: Interpretations of Interpersonal Conflict in a Melanesian Society,” in Directions in Cognitive Anthropology, ed. by Janet W. D. Dougherty. Pp. 345-370. Urbana: University of Illinois Press.

Wilkinson, Leland, Grant Blank, and Christian Gruber 1996 Desktop Data Analysis with Systat.® Upper Saddle River: Prentice Hall.

Williams, Colin H., and J.E. Ambrose 1988 “On Measuring Language Border Areas,” in Language in Geographic Context, edited by Colin H. Williams. Pp. 93-135. Clevedon: Multilingual Matters.

Wolfe, Michael 1997 1001 Roads to Mecca. New York: Grove Press.

Yablokov, A.V. 1986 Phenetics: Evolution, Population, Trait. New York: Columbia University Press.

219 Young, J.C. 1978 “Illness Categories and Action Strategies in a Tarascan Town,” American Ethnologist. 5(1):81-97.

Zelinsky, W. 1973 The Cultural Geography of the U.S. Englewood Cliffs: Prentice Hall.

Zipes, Jack 1991 Arabian Nights: The Marvels and Wonders of the Thousand and One Nights. New York: Penguin Putnam, Inc.

220 BIOGRAPHICAL SKETCH

Kevin Daniel Pittle was born in Morristown New Jersey on May 16, 1972, the son of Marshall D. Pittle and Eileen D. Pittle. After completing his work at Coral Springs High School, Coral Springs, Florida in 1991, he went on to receive the degree of Bachelor of Arts from Florida State University in August of 1995, and the Master of Science degree from Florida State University in August of 1998. In August of 2000 he entered the doctoral program of the Department of Anthropology at Florida State University. He has taught courses in cultural and linguistic anthropology at Florida State University, the Oregon Summer Institute of Linguistics, and Biola University. He currently resides in La Mirada California with his wife Natasha and their children.

221