Advances in generative structuralism: Structured ...

The Journal of Mathematical Sociology

ISSN: 0022-250X (Print) 1545-5874 (Online) Journal homepage: http://www.tandfonline.com/loi/gmas20

Advances in generative structuralism: Structured agency and multilevel dynamics Thomas J. Fararo & Carter T. Butts To cite this article: Thomas J. Fararo & Carter T. Butts (1999) Advances in generative structuralism: Structured agency and multilevel dynamics, The Journal of Mathematical Sociology, 24:1, 1-65, DOI: 10.1080/0022250X.1999.9990228 To link to this article: http://dx.doi.org/10.1080/0022250X.1999.9990228

Published online: 26 Aug 2010.

Submit your article to this journal

Article views: 104

View related articles

Citing articles: 10 View citing articles

Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=gmas20 Download by: [University Of Pittsburgh]

Date: 06 May 2016, At: 14:14

Downloaded by [University Of Pittsburgh] at 14:14 06 May 2016

Journal of Mathematical Sociology 1999, Vol. 24(1), pp. 1-65 Reprints available directly from the publisher Photocopying permitted by license only

© 1999 OPA (Overseas Publishers Association) N.V. Published by license under the Gordon and Breach Publishers imprint. Printed in Malaysia.

ADVANCES IN GENERATIVE STRUCTURALISM: STRUCTURED AGENCY AND MULTILEVEL DYNAMICS THOMAS J. FARAROa,* and CARTER T. BUTTSb a

Department of Sociology, University of Pittsburgh, Pittsburgh, PA 15260, USA; bDepartment of Social and Decision Sciences, Carnegie-Mellon University, Pittsburgh, PA 15213, USA (Received 30 April 1999; Revised 16 July 1999)

We set out an approach to theorizing termed generative structuralism. In two stages, respectively, we address two problems in social theory - agency-structure and micromacro linkage - in relation to a central problem in the discipline, namely, the sense of its fragmentation. The first theoretical problem is: How can we create generative models and methods that are responsive to the interpenetration of agency and structure in social practices? We show how Bourdieu's concept of habitus and Giddens's concept of structure can be articulated to generative model-building ideas and methods. The second theoretical problem is: To what extent can we create generative multilevel models of large-scale social system dynamics? We address this problem through a strategy illustrated by a model of emergent cultural stratification, including a study of its properties through simulation analysis. In undertaking these tasks, we both describe and illustrate the general character of generative structuralism. In doing so, we combat fragmentation by articulating connections among theoretical concepts, theoretical problems, and theoretical methods drawn from three sociological traditions, which we term, respectively, structured agency theory, network structuralism, and sociological social psychology. In a concluding metatheoretical discussion, we describe generative structuralism as an orienting strategy that can be implemented in a variety of theoretical research programs. KEY WORDS: Generative models, micro-macro linkage, agency-structure problem, multilevel dynamics, habitus, production systems, cultural stratification, expectation-states theory, metatheory.

* Corresponding author. E-mail: [email protected]. 1

2

T.J. FARARO AND C.T. BUTTS


INTRODUCTION It sometimes seems that the image of fragmentation in our discipline is all but overwhelming our ability to engage in necessary partial syntheses of work accomplished in different research traditions. Some would say that the proliferation of diverse and seemingly disconnected paradigms is but the most recent sign of what is, from their perspective, "the impossible science" (Turner and Turner, 1990). Other observers see our vast data-collection capabilities as having so far outrun our ability to direct and absorb findings through theoretical synthesis that what we have produced is a mere succession of efforts that leave data "on the shop floor" (Molotch, 1994) piling up, not accumulating. And so far as the organ of synthesis is concerned - theory - it has been proclaimed that we have entered a new epoch characterized by recognition of the failure of earlier hopes for the development of rigorous theoretical science (Seidman, 1994; Lemert, 1995). Yet, amidst the growing fragmentation of the field and the somewhat anomic condition it can generate for participants, promising developments have taken place. The diverse materials for partial theoretical syntheses are in place but the perspective on them that would encourage the actual production of such syntheses is lacking. The materials include the writings of some of our most influential recent social theorists. But they also include research programs that are often regarded as highly specialized and esoteric creatures. Among these materials are key developments in mathematical, formal and computational sociology involving concepts such as networks, grammars and nonlinear dynamics. In this paper, one aspect of our work is to assemble one group of such materials and show that they can be more than merely juxtaposed - put positively, that they can be articulated to constitute a coherent and historically important development in general theoretical sociology, namely advances in what we term "generative structuralism." This term has been used in some recent contexts (Fararo, 1989a; Fararo and Skvoretz, 1993). What it refers to is an example of what has been termed an "orienting strategy" (Berger and Zelditch, 1993). As will be discussed in more detail in our conclusion, an orienting strategy is a cognitive entity superordinate to particular research programs and theories in the sense of enabling diverse implementations.


GENERATIVE STRUCTURALISM

3

As such, it can exhibit a proliferation of paths of development. Yet, unless this proliferation is accompanied by integrative efforts, the overall picture of how social phenomena are to be grasped in a generative structuralist mode will be obscured by the sheer multiplicity of vocabularies and models. Thus, this paper proceeds from the premise that sociology can advance through recurrent episodes of partial integration amid the inevitable proliferation that characterizes any present-day scientific endeavor, employing what has been called "the spirit of unification" (Fararo, 1989a,b). Three Problems, Three Traditions

If such a commitment to recurrent syntheses is an important and also shared aspect of some strands of recent social theory (Ritzer, 1990), explication of a metatheory that encourages it will not be persuasive without the explicit treatment of recognizably important problems. In this paper, we focus on three problems of social theory. Two of the problems are general-theoretical. First, how can we create a theoretical strategy that enables us to construct models that fruitfully combine structure and agency? Let us call this "the agency-structure model-building problem." It is a version of the well-known agencystructure issue in social theory. Second, how can we account for dynamic equilibria in large complex social systems comprised of multiple levels of processes? Let us call this "the multilevel dynamics problem." It is a dynamic version of the problem of micro-macro linkage (Alexander et al., 1987). The third problem pertains to the state of the discipline and its theory. How can we proceed with respect to these two general-theoretical problems in such a way as to partially synthesize some of the relevant but diverse strands of theory characterizing sociology? Let us call this "the knowledge capital problem." In this paper, the general strategy of generative structuralism will be discussed and illustrated by reference to these problems. Three research traditions play a central role in this effort. Each has its own inner diversity and we do not attempt to do more than draw upon selected constructs and principles from particular programs within these traditions. However, one of our key purposes is to articulate certain conceptual connections among seemingly quite different traditions and styles of theorizing. In our terms, the three traditions


4


are "structured agency theory," "network structuralism," and "sociological social psychology." By structured agency theory we mean key ideas in the works of Pierre Bourdieu and Anthony Giddens. Both authors have responded to and considerably revised ideas originating in French structuralism, Bourdieu in close contact with empirical studies and Giddens in close contact with the classic and contemporary theory traditions. Concepts such as habitus in relation to fields (Bourdieu) and practical consciousness in relation to structure (Giddens) are intended to overcome dualisms, especially - as these authors see the problem - that between the objectivism of structuralist thought and the subjectivism of phenomenological and interactionist sociologies. While structuralism, in its French variant, studies objective social and cultural structures in terms of ideas such as binary opposition, the interactionist and phenomenological types of theorizing highlight agency in the generation and reproduction of structure. By contrast, Bourdieu and Giddens stress the notion of social practices, which have both a structural and an agency aspect. It should be noted that Bourdieu (1990b) has termed his approach "genetic structuralism," and his meaning is very close in spirit to our "generative structuralism." The major difference is our commitment to a more formal mode of theorizing, often treating abstract and idealized cases, while Bourdieu rather strongly prefers to develop his ideas in the context of particular empirical studies. By network structuralism — we prefer this term to the alternative "American structuralism" - we mean not only social network analysis but also a range of contributions in the domain of formal, mathematical, and computational sociology. This form of structuralism itself has a multi-parental history, including the influence of developments in anthropology (French structuralists and British social network analysts), in linguistics (Chomsky's generative and transformational grammars), and the wider field of mathematical model-building in social science (Mullins, 1973: Chapter 10). By the late 1970s, one part of this whole development achieved considerable penetration into sociology in the form of social network analysis. Here the focus has been on the formal and quantitative description of patterns of social relations among actors. Recent developments in social network analysis have emphasized the evolution of networks (Doreian and Stokman, 1996). Process representations have been an important part of mathematical



5

social science throughout its history (Fararo, 1997). But, until recently, the models tended to be simpler than theorists would like to see, but new developments in computational sociology (illustrated later in this paper) enable complex systems to be studied from a theoretical point of view. Moreover, such computational models are not necessarily numerical. They use techniques developed in such areas as artificial social intelligence, cognitive science, and complexity theory, including neural networks, production systems, and genetic algorithms (Bainbridge et al., 1994; Carley, 1996). Mathematical ideas, new and old, pervade all of these developments, implying a strong overlap with mathematical social science in the older and still viable sense. Taken as a whole, then, these research traditions involving networks, grammars (or production systems, see below), process representations and computational model-building are what is meant here by "network structuralism." By sociological social psychology we mean two interconnected modes of thought that have in common a strong focus on the analysis of social interaction itself, rather than only its antecedents and consequences within a broader social and cultural environment. One set of research programs in this category is often called "interpretive sociology" or "micro-interactionism." It includes such familiar perspectives as symbolic interactionism, ethnomethodology, and reality construction theory (Berger and Luckmann, 1966). We do not regard such perspectives as anti-scientific merely because they have not stressed large-scale phenomena, quantification, or model-building. We consider their ideas to be a significant part of the knowledge capital of the discipline and have built on those ideas in developing our approach. The second set of research programs of sociological social psychology deal with "group processes" (Foschi and Lawler, 1994; Szmatka et al., 1997). Among the long-term research programs in this tradition are affect control theory (Heise and MacKinnon, 1993) and expectation states theory (Wagner and Berger, 1985; Cohen, 1989; Berger and Zelditch, 1993). These and other group process research programs have been largely, but not entirely, social psychological and experimental. They are keyed to the idea of doing research that is theory-driven and cumulative. Taken as whole, this branch of sociological social psychology is a strongly sociological form of social psychology because the emphasis is on how social structures shape face-to-face interaction or

6


how situational social structures emerge in the course of such interaction (Lawler et al., 1993).


Overview We will be presenting our ideas in two stages, corresponding to the following two major parts of the paper. In the first stage, the theme is "structured agency" model-building rooted in the above three sociological research traditions, components of which can be partially synthesized. The focus is conceptual and representational. We cite specific generative structuralist features as we go along, but we postpone a discussion of the most general and defining features of the approach until the conclusion of this part, so that we can refer to the materials just encountered as examples of what is meant. Then we turn, in the second stage of our work, to a treatment of the problem of multilevel dynamics. We construct and analyze a specific theoretical model dealing with the generation of cultural stratification to illustrate our general approach to the problem. In the last part of the paper, we will return to the themes of our opening remarks. We will summarize our treatment of the two general-theoretical problems of structured agency model-building and large-scale multilevel dynamics, respectively, and then return to the issue of fragmentation, discussing the place of generative structuralism in the present state of the discipline.

STRUCTURED AGENCY By structured agency we mean the idea that actors embody forms of competent social action that are intrinsically connected to or interpenetrate with social institutions. This competence is framed somewhat differently in different theories, however, so we treat a sequence of articulated developments drawn from the three families of research programs, linking them as we proceed. Our plan is as follows. We will open with a capsule summary of two key ideas of structured agency theory: habitus (Bourdieu) and structure (Giddens). Then we proceed to synthesize these ideas with certain contributions of the network structuralist approach relating to grammars, artificial intelligence and cognition. The key linkage is through the construction of generative


7


models. Finally, we turn to showing the linkage between a core idea coming out of sociological social psychology, as synthesized with a social network viewpoint, converges with and even explicates another aspect of structured agency in its habitus sense. At the end of this sequence of synthesizing constructive developments, we pause to summarize and characterize the basic features of generative structuralism.

Habitus The ideas of Bourdieu have captured the attention of sociologists over the past decade (Calhoun et al., 1993; DiMaggio, 1994). It is not possible to review the enormous corpus of his work. Most relevant for this paper are Bourdieu (1977, 1984, 1989, 1990a,b, 1993) and Bourdieu and Wacquant (1992). For present purposes, the following brief orientation will have to suffice. Bourdieu's image of social life is that it involves various "fields," in which people and things occupy positions more or less near each other in terms of their possession of various forms of capital, such as economic capital and cultural capital. Thus, this space of positions associated with a field is taken as a given, although subject to transformations. A key idea is that individuals are socialized at particular locations in the space and hence acquire generalized forms of disposition or "habitus." The idea is expressed by Bourdieu (1990a: p. 53) as follows: The conditionings associated with a particular class of conditions of existence produce habitus, systems of durable, transposable dispositions, structured structures predisposed to function as structuring structures, that is, as principles which generate and organize practices and representations that can be objectively adapted to their outcomes without presupposing a conscious aiming at ends or an express mastery of the operations necessary in order to attain them. In this somewhat enigmatic statement and in related uses of the concept, the following properties of habitus are specified: (1) Habitus is durable. It is "a present past that tends to perpetuate itself into the future by reactivation in similarly structured practices" (Bourdieu, 1990a: p. 54). (2) Habitus is transposable. It generates practices in diverse fields of activity, adjusted to specific situations.

8



(3) Habitus consists of structured structures. These incorporate the objective social conditions of their inculcation. As has been indicated, Bourdieu places inculcation in the context of a position in a social space, defined in terms of combinations of amounts of different forms of capital, so that actors whose initial positions are similar will tend to acquire a similar habitus. Put another way those in the same class will share the same habitus. For instance, being upper class, they will tend to have a similar "feel for the game" (Bourdieu, 1990a: p. 66) that differs considerably from those in the lower classes. Yet each will embody tacit abilities to recognize situations and to act appropriately in them, from the standpoint of their positions in social space. Further, they will embody standpoints toward the structure of social space itself, such standpoints themselves structured by position in the space. In short, both the practices and the representations (perceptions, images, theories) are shaped by position in social space. In regard to practices, it is clear that Bourdieu employs habitus in at least two different analytical contexts. In one context, the analytical focus is on how life-styles vary with position in space. Habitus here has several analytic roles, but one is to simply account for variation in life-style as a function of variation in position in space. Habitus, in this analytical role, is an unmeasured "intervening variable." It is an actor's dispositions as presumed to be adjusted to position in space, so that the resulting dispositions explain life-styles "choices." For instance, those with considerable wealth will have "a taste for luxury," while those who are impoverished will have "a taste for the necessary" (Bourdieu, 1984: p. 178). Thus, we have a sequence of states, each in the relevant space (Bourdieu, 1984: p. 126): position (in social space) —+ habitus state (in habitus space) —* tastes (in life-style space) The distinctions made in terms of tastes are modes of classification made by actors that then classify them. In turn, this contributes to the reproduction of the underlying patterning of unequal distribution of capital that forms the basis of the social space. It is the second and related analytical context of the use of the habitus construct that is more relevant at this point. Namely, the focus is



9

on situated behavior. For ifistance, some actors occupying positions characterized by high overall volumes of capital interact with some actors occupying positions of low overall capital. In such a situation, the habitus construct functions to bridge the distance from the social space exogenous to the situation to the actual production of action in the situation. For instance, the externally well-placed tend to dominate the situation of interaction. To account for this, Bourdieu uses the habitus again as an intervening variable and the logic of the analysis is a sequence in which the higher element is given at the time of the formation of the lower element: position (in social space) —> habitus state (in habitus space) —> relational acts (in situated action space) As in the first context, so here as well the outcome of the local situation is one among a set of such outcomes that tend to contribute to the reproduction of the structure of the social space that was given for that occasion. Structure

In Giddens's (1984) contribution to what we are calling structured agency theory, three interrelated ideas are defined relating to social structure: structural principles, structures, and social systems, the latter with structural properties. Associated with these distinctions is the principle that structures have a duality aspect: they are not only properties of social systems but also ingredient in the constitution of agents. So neither a purely subjective frame of reference, locating structure in the minds of individuals, nor a purely objective frame of reference, locating structure as external to individuals, is adequate. Structures, in Giddens's conceptual scheme, consist of rules and resources. Such a structure is said to be virtual, not actual. It is said to be "out of time and space." It exists only in the memories of individuals and as instantiated in their practices. Contrasted with structure is system. Social systems have structural features, implicated in direct and indirect social relations between actors. Structural principles are those structural properties implicated in the reproduction of societal totalities involving a multiple set of institutions.

10



By "rule," Giddens means a generalized procedure, involving a mastery of some activity, an ability to do something as part of the knowledgeability of a human being. Such rules constitute or create meanings as well as control conduct in the sense of implying some degree of possible sanctioning. Types of rules can be distinguished in terms of a number of characteristics, put in the form of binary oppositions, including: (1) intensive vs. shallow (i.e., whether constantly invoked or not, e.g., rules of a natural language are intensive) (2) tacit vs. discursive (i.e., know-how vs. talk about) (3) informal vs. formal (4) weakly sanctioned vs. strongly sanctioned. One of Giddens's most frequently cited distinctions is between practical consciousness and discursive consciousness, clearly related to the second binary distinction in the above list. Tacit generalized procedures characterize most of routine social activity. Agency is involved here because to be an agent means to have the know-how to do things in the world, especially in relation of other human beings. So human beings are knowledgeable agents with practical consciousness employing tacit rules, some intensively, others in a more shallow mode, some strongly sanctioned, some not. This framework is meant to incorporate the insights of ethnomethodological and related studies of everyday social activity, with a particular emphasis on the largely unintended consequences of such local activity for the reproduction of institutionalized practices. The Production System Principle

Given these brief explications of two major concepts of structured agency theory, we turn to the linkage of these ideas to principles and models drawn from the other two traditions, network structuralism and sociological social psychology. We begin with the structure concept just reviewed. Giddens does not employ any formal representation that would enable the construction of structural models that satisfy the conception of structure he has set out. This is possible through the use of production system models (Axten and Fararo, 1977; Fararo and


11


Skvoretz, 1984; 1986b; Skvoretz and Fararo, 1996a; Heise and Durig, 1992). These types of models are among those employed in what has been termed "social artificial intelligence" (Bainbridge et al., 1994). The directly relevant idea has been stated by Fararo (1989a: p. 211) as follows: The Production System Principle. An institution is both a social space defined by constitutive rules and a set of normal forms of interaction within it that are generated by a distributed set of situated production subsystems linked to universal interpretive procedures.

The concept of institution presupposed here is based on Berger and Luckmann (1966), as well as Nadel (1951) whose "if, then" conception of institutionalized action is embodied in the idea of "situated productions." This mode of formalization is adapted from the work on human problem solving by Newell and Simon (1972) in which the theory of production systems was set out and applied. Only a necessarily brief statement of the ideas will be given here. A situated (or "situational") production takes the form: situation -> ACTION where the left side is a type of situation and right side a type of action. Newell and Simon constructed a number of systems of situational productions - each a tacit procedure used by an actor - to account for over-time problem-solving activity, e.g., solving of a cryptarithmetic puzzle. For instance, one situational production in one model — a system of situational productions — takes the form: if an assignment of a number to a letter is new then FIND and PROCESS the column containing the letter Here a state of knowledge is the contingency under which a linked set of cognitive operations are carried out. The problem-solver is thought of as testing his or her current knowledge state against the type of situation appearing in each of a series of productions, arranged in a priority order (inferred by the model-builder from behavior). For instance, for a certain puzzle, if the current state of knowledge includes a new assignment of a number to a letter, say 5 to the letter D, then


12


(assuming no higher-order production applies), the problem-solver finds a column (FIND) containing D and replaces that letter with 5 throughout (PROCESS). These latter two actions as well as such entities as "letter" are defined in the actor's particular "problem space," a particular implementation of the more general idea that cognition presupposes a symbolic space where symbols that designate objects and actions are defined. Newell and Simon reserve the term "production" for what we call the situational productions, whereas computer scientists and formal language theorists use the term more broadly to cover also these entity-declaring expressions in the symbolic space. We resolve the potential confusion by referring to the nonsituational type as "constitutive rules" (Fararo, 1989a: Chapter 3) so that in our context "production" and "situational production" coincide. It should be noted that this type of model-building is extremely rigorous as to description of processes involving symbols, using a special notation (Backus Normal Form) developed within the computer science community to describe formal languages. Although we have employed this notation in our own work, for the purposes of this paper, beyond representing types of action in capital letters, we will employ just one other feature of the notation. Namely, a symbol that can take a variety of substitution instances, defined in a symbolic space, is a variable and is placed in brackets, e.g., (letter) and (numeral). An analogous procedure in the use of production systems in linguistics is illustrated by (noun) with specified substitution instances (e.g., "boy" and "book"). In our work, for sociological purposes, as indicated by The Production System Principle, the focus is on models that are intended to represent institutionalized social action, involving typifications and relative uniform behaviors that can be represented in terms of a production system. We are dealing with models of actors as knowledgeable agents, where the first-order constructs of the actors (Schutz, 1973) are components of the models. This means that the symbolic space is a social and common or cultural space defining types of entities and operations or relations concerning them. The "constitutive rules" referred to in the Production System Principle define very general types of entities in the common world of the actors as well as more specific types of entities relevant to particular institutions. In either case, the Schutzian notion of "typification" conveys the sociological intuition.



13

Numerous variables in particular models are such social typifications with particular persons or groups as substitution instances. For instance, we have modeled the familiar restaurant situation (Axten and Fararo, 1977; Skvoretz and Fararo, 1996a) in such a way that we have several types of actors, in particular, {waitperson) and (customer). When the interaction system is analyzed, it is particular people who interpret each other as instances of these types. For instance, a particular instance of (waitperson) in a particular instance of (restaurant) classifies people around her as instances of (customer) whom she will serve. She does so in the context of having activated WAITPERSON, a production system which contains a production calling for SERVE as contingent on the full array of states of current instances of (customer) which she has in mind as parts of a knowledge state referent to the dynamic situation. When that situation, as known to the actor, matches the type of situation which calls for SERVE (in respect to a particular customer) and no higher order production is satisfied with respect to that knowledge state, then in fact SERVE occurs. The model generates dynamic situated interaction, involving actors interpreting each other in respect to type (in the relevant symbolic-cultural space) and in respect to state of action (in the relevant typification as a type of actor, e.g., customer). Once SERVE is produced, it alters the state of the situation of relevant actors and thereby their knowledge state, and thereby their production of further action in accordance with their subsystem of productions. Thus, unlike the Newell-Simon models, in the social-institutional context a production system is distributed among actors, each having a subsystem of productions that generate actions that change the situation of self and others. The change in situation is recorded as a knowledge state of one or more others and thereby presents a contingency that is responded to by one of their productions or presents a cognitive problem of a nonroutine sort, in which case a Newell-Simon problem-solver can be invoked to account for the action. Because of the distributed character of this sort of "grammar of interaction," the integrative problem of coordination among different parts of a single system is an intrinsic problem dealt with in the models. In cognitive psychology, production system model-building employs a distinction between two forms of knowledge: declarative and procedural (Anderson, 1993). The constitutive rules defining a space

14



correspond to shared declarative knowledge, the situated productions to procedural knowledge. For instance, knowledge that a piano has a keyboard with black and white keys, no matter how detailed and accurate, is far different from competence to play the piano. The latter is procedural knowledge (know-how), while the former is declarative knowledge (know-that). Productions have four key aspects (Anderson, 1993: p. 31): (1) Modularity. Each is a unit of procedural knowledge. (2) Abstraction. They have generality because they contain variables. (3) Goal structuring. They link together to form hierarchies or meansends chains. (4) Condition-action asymmetry. Action is conditional on situation. Let us illustrate these points with respect to the gender institution in its contemporary societal form and reiterate some important points made earlier in the context of this new example. Point (1) emphasizes the practical action aspect. For instance, the socially knowledgeable actor has procedural knowledge for "doing gender," as in Butler's (1990) discussion of gender as performance. The variables of point (2) are the categories which actors use to classify each other, such as (man) and (woman), defined through constitutive rules - in this case that any person is a man or a woman and not both (in the conventional gender institution). As noted, in sociological applications the variables are typifications in the sense of Schutz (1973). Point (3) gives depth to structures of productions. Given a goal which is part of a knowledge condition of a production, the corresponding action itself may be interpreted as a more micro-production system. For instance, suppose that in a certain situation, a system of productions, call it P, has been activated. Let it include a production that, to sustain a female identity in a situation, invokes the action APPLY-MAKEUP (to one's face). Within P, there is no inner structure to this action. But, in illustration of point (3), it could be represented by a system of productions in its own right, one that is in micro—macro relation to the system P. This means that the larger action, in a certain sense, is P itself. For instance, let P = WOMAN. In conformity to point (1), WOMAN is a subsystem of productions that performs a variety of actions, each contingent on the type of situation, and in relation to other actors, notably others as classified as (woman) or (man). Note that (woman)



15

designates a type of object, while WOMAN is a part of production of action, a subsystem of the whole gender institution. Just as in speaking we do not consult the relevant grammatical rule before talking, so in the case of the gender institution, one does not suppose that situated productions are consulted before acting as a woman or a man. This is embodied knowledge. We may note, then, that this interpretation of a situated production system closely matches Bourdieu's habitus concept, a point to be taken up again later. A system of distributed productions, associated with a symbolic cultural space, defines a generative model. This model has a grammatical aspect in that the cultural space and the productions impose a kind of syntax on social interaction. In principle, any "normal form of interaction" is a kind of interaction sentence in an institution language (Skvoretz and Fararo, 1980; Skvoretz, 1984). One aspect of the empirical adequacy of a such a generative model is that it does not generate institutional nonsense, such as "The couple were married by the quarterback." Also, the generative model has a "dynamical" aspect, involving social events. Because action is contingent on situation conditions encoded in knowledge states and because actors' states are coupled in a system of interaction, there is a sequential unfolding of the state of the system and the behaviors that depend upon it. In other words, the temporality of events is captured in this aspect of production systems (Heise and Durig, 1997). This feature is a departure from the standard image of generative grammar in linguistics, which does not generate talk (situated interactive speech) but rather a totality of legitimate utterances in abstraction from social situations, a kind of purely cultural system. French structuralism followed the linguistics model, with no sense of dynamics, whereas our approach involves generative structuralism: this means a conception of generativity that is "hybrid," i.e., involves both a grammatical and a dynamical aspect, capturing both the syntax of social life and its eventfulness. (There exists a correspondence between types of language and types of "automata," which are systems in the dynamic sense and in the second part of this paper, we employ an automaton representation.) In linking these ideas to Giddens, we note that it is obvious that language is both internal and external. It is internal in a "virtual" form, as encoded in "memory traces" and external in the form of actualization, i.e., talk. This applies to institutions generally, taking language


16


as a special case. It is useful to introduce the concept of a "unitinstitution," (Fararo and Skvoretz, 1984): a single scheme of typifications involving actors, acts and situations in a system of social interaction, somewhat analogous to a game. Each unit-institution has a generator which is a distributed production system, hence a form of procedural knowledge. Such a unit-institution generator is generative of a specific totality of normal forms of interaction, analogous to all the possible plays of a game. The frequency with which the various forms of interaction arise is not determined by the generator so much as the elements of action external to that generator, because specific choices are not given by the generator in the same sense that the choice of the article "the" or "a" is not determined by the grammar of English. Each generator thereby leaves space for the factor of agency. In Giddens's terms, a unit-institution generator is internal, but what it generates is external and has social reality for each actor because each "sees" the situation in terms of the categories intrinsic to it. Institutions are complex systems of unit-institutions and so have the same duality. The confusion about what Giddens means by arguing that structure is "virtual" is resolved by the principle that such a structure consists of productions. Each institution corresponds to a complex of generators, distributed production systems which are embodied knowledge. As incessantly activated, such production systems generate the normal forms of interaction comprising the institutional aspect of a social system. The multiplicity of interlinked unit-institutions of a social system constitutes the "structural properties" of the system, in Giddens's terms. Put another way, these structural properties constitute "structured agency." It should be noted that an institutional structure, as embodied in production system terms, can be treated as "multiply embodied" in numerous collectivities, such as organizations (Jepperson, 1991). The use of production system models clarifies two key principles of structuration theory. First, it clarifies Giddens's (1984: p. 25) "duality of structure" principle: "Structure is not 'external' to individuals: as memory traces, and as instantiated in social practices, it is in a certain sense more 'internal' than exterior to their activities in a Durkheimian sense. Structure is not to be equated with constraint but is always both constraining and enabling." Productions are internal immanent



17

generators of action. As such, they are stored in memory. In particular occasions, given their contingent form, they are triggered or activated, producing action that changes the state of the situation from the perspective of one or more other actors and thereby triggering their productions in turn. In this way, observable structured interaction episodes take place in "the external world." Duality is transparent. Second, production models clarify the "reconstructed problem of order" (Giddens, 1984: p. 35): "to explicate how the limitations of individual 'presence' are transcended by the 'stretching' of social relations across space and time." The recent contribution by Heise and Durig (1997), without reference to Giddens, nevertheless goes some distance in solving this type of order problem in terms of production system models. They use "cooperative productions" to show how "macroactions" are generated. The author of a macro-action, in this sense, is an individual, but one who can control a variety of elements of action in order to realize events of wide scope embracing numerous other actors in their routine performances. Having connected Giddens's structure concept to the conception of a distributed production system that generates institutionalized interaction sequences, we turn to a discussion of conceptual problems relating to "rules" in social theory, drawing upon a contribution by Bourdieu to create a clarification of the issues. Rules, Norms, Productions and Habitus

A perplexing problem in treating social practices involves the concept of rule. One could say, for instance, that walkers are rule-governed in the way that they recognize and produce instances of walking together (Ryave and Schenkein, 1974). Then the production system would be said to formalize these "rules" in a model. But the use of the term "rule" can be the occasion of considerable ambiguity. Bourdieu (1977: p. 27) has discussed rules in an effort to distinguish among three senses of the term: rule as norm, rule as element of theoretical model, and rule as scheme or principle immanent in practice. These distinctions help us to clarify the interpretation of the generative model-building suggested here for thinking about structure in the institutional sense. To make use of Bourdieu's distinctions for this purpose, let us draw upon some research of Willis (1977) as discussed by Giddens


18


(1984: pp. 289-304) dealing with the behavior of pupils in certain British schools. These pupils, called "lads," are nonconformist. Giddens emphasizes how the unintended consequence of their behavior in school is to condemn them to take up the positions as the lowest, unskilled level of the work force. Their very rebellion against "the system" has the ironic consequence, thereby, of helping to reproduce that system. Here we focus on the local behavior in the classroom to illustrate the distinction between norm and production. First, a norm in the classroom situation is that pupils sit still in class. Second, this norm is a component of the institution with its two relevant typifications: pupil and teacher, along with the corresponding generators labeled PUPIL and TEACHER. However, what we want here is the nonconforming variant of the former, labelled PUPIL*. In constructing the model, the norm is an item of declarative knowledge, not a production. Third, the productions can be construed as rules immanent in practice in the observed classrooms. A partial sketch of the distributed generator that embodies such tacit rules is as follows (see Figure 1 and recall that we are minimizing notational rigor, calling attention only to certain key features of the formalism): PUPIL*: PI: If (teacher) is present, then MOVE AROUND (the room). P2: If (teacher) reminds one of (norm), then GIVE ACCOUNT (with sarcasm). TEACHER: P3: If (pupil) is violating (norm), then REMIND (the pupil of the norm). The cultural space for this model includes not only the typified actors but also the particular norm that pupils are expected to sit still in class. Note that in production P2, (teacher) is a type of social object, whereas TEACHER is a component of an agent. Bourdieu's point that a norm and a rule immanent in a situation are distinct entities is represented by the model "rules" (productions). Namely, the immanent rule that these "lads" move-around (in defiance of the norm) is represented by the model rule PI and the status of the norm (framed in the space) as in fact normative is shown by the P3 response by the authoritative actor, the teacher, whose immanent rule tries to bring the behavior back in line by reminding the violator to sit still in place. But this reminder is merely the occasion, via P2, for a lad


Knowledge State


Production Subsystem

Agent 1

19

Knowledge State

Production Subsystem

Agent 2 (a) Abstract Two-Agent System.

pupil-2 moving around

TEACHER

Agent 1

Agent 2

(b) Classroom Example. Agent 1 is the teacher, Agent 2 is a pupil. FIGURE 1 Distributed production system model. (In part (b) only a part of the knowledge state of each agent is shown.)

to express contempt for such authority and its norms, along with the giving of an account which treats the norm as a rhetorical resource: the account is given in a sarcastic fashion. To sum up, a norm is a rule that is an item of shared declarative knowledge in the situation of action. But practice exhibits an altogether different rule of conduct - a production - that presupposes this declarative knowledge and acts in relation to it. Hence, the distinction between a norm and a rule immanent in practice is formally represented in the model, capturing what Bourdieu calls "the logic of practice." What the example also suggests is the hypothesis that at least some of the content of habitus is made up of productions. This

20



correspondence appears plausible, too, when we juxtapose two statements related to the embodied and dispositional character of the habitus. First, a production system viewpoint: Intuitively, declarative knowledge is factual knowledge that people can report or describe, whereas procedural knowledge is knowledge people can only manifest in their performance... The same abstract knowledge can have both procedural and declarative embodiments. Thus, declaratively we might have memorized the layout of the typewriter keyboard, and procedurally we may know the keyboard as part of our typing skill (Anderson, 1993: p. 18). Next, Bourdieu: The schemes of the habitus... owe their specific efficacy to the fact that they function below the level of consciousness and language, beyond the reach of introspective scrutiny or control by the will. Orienting practices practically, they embed what some would mistakenly call values in the most automatic gestures or the apparently most insignificant techniques of the body - ways of walking or blowing one's nose, ways of eating or talking... (Bourdieu, 1984: p. 466).

Note the parallel concreteness of physical action in the two cases. What this suggests is that habitus, in one aspect, may be modeled by structures of production rules. The advantages of doing so are twofold: first, the obvious way in which productions are generative clarifies this aspect of habitus; and, second, structured agency as tacit embodied "know-how" is formally represented. Resources and Transformations

For Giddens, structure consist of "rules and resources." The explication of the generative logic of structure through the Production System Principle has been confined to the first of these two elements. Giddens's inclusion of resources under the rubric of structure has given rise to conceptual difficulties because it does not seem plausible or useful to regard resources as "virtual." For example, Sewell (1992) points out that it is actual money, actual material factories and the like that are resources and the treatment of these as akin to "rules" is confused and confusing. In a production system context, the point can be made as follows. Actions are of various sorts, and instrumental actions depend upon control over certain facilities if they are to count as instances of an action of a given institutional type. One can write an


21

instrumental action in the form:


situation -> ACTION (factors) (-> products) In this form, an activated production has a product or output, resulting from combining one or more factors or inputs. Any entity that functions as a factor of production, in this generalized sense, is a resource. The cultural and cognitive aspect of a resource is indicated by the element of typification in a production system context. Such typifications are variables in a model. For instance, (dollar), (factory), and (name of Jane Austin novel) are variables. The factor of production, hence the resource, is not the variable but the instance recognized by actors: not (dollar) but particular items that count as money in the system of interaction. In Sewell's axioms treating the transformation of structures, rules ("schemas" in his avoidance of the ambiguous "rule" term) are virtual and resources are actual. Resources are read as texts to recover the schemas they instantiate. The particular building with its operations is recognized as a factory. But this "reading" is true of things generally a key theme in symbolic interactionism, one might note. For instance, at the level of practical consciousness, the teacher in our earlier example is "reading" a person in the classroom as an instance of (pupil). But this does not in itself make the pupil a resource, since the reading may not be embedded in the context of an instrumental action with the pupil as a factor of production. A school, on the other hand, has a succession of cohorts of pupils as a factor of production with an output of prospective credentialed workers to the economy. This example also illustrates the idea that concrete agents can function as resources, a point noted by Ruef (1997b), whose paper should be consulted for a further formalizing discussion of Sewell's argument. The treatment of transformation of structures is closely related to the problem of reproduction framed within structured agency theory. Both are problems of multilevel social dynamics. Essentially, Bourdieu's central thematic concern is "the question of how stratified social systems of hierarchy and domination persist and reproduce intergenerationally without powerful resistance and without the conscious recognition of their members" (Swartz, 1997: p. 6). Later in this paper's second major part, we will set out a generative model that

22


treats this question, using it as an example of a generative structuralist approach to the problem of multilevel social dynamics.


Habitus and Actor Representations

The above analysis showed that the conception of habitus in the context of the generation of practice corresponds to the formal idea of a production system, a development in formal generative thought embedded in network structuralism. In treating actor knowledge, we have distinguished between procedural knowledge embodied in productions and another form of knowledge, called declarative above. The conception of habitus includes a declarative knowledge component, namely the actor's perspective on the social space in which that actor occupies some position. A classic empirical case is the system of images of class stratification constructed by the authors of Deep South (Davis et al., 1941). According to the empirical analysts (p. 63) "the very way in which people conceive of the class distinctions varies with their social position." They add that "the greater the social distance from the other classes the less clearly are fine distinctions made" (p. 71). A generative approach to images of stratification was taken in Fararo (1973: Chapterl2) and by Coxon and Davies (1986). We concentrate here on the process-oriented former approach in which the analytical objective is to trace out the over-time development of such "images" of social structures (see also Kosaka and Fararo, 1991; Hummon and Fararo, 1995). A formal generative theory that enables such model-building explains the sorts of empirical generalizations reported by the authors of Deep South. Based on a set of axioms describing a sequential process, the theory shows how stable images emerge in the social interaction of actors located within the class system. To illustrate a step in the generative process, consider a model with a class system defined by a composite ordering of the form HH, HM, HL, MH, MM, ML, LH, LM, LL. This model is based on two dimensions of ordering - two forms of capital in Bourdieu's terms - with three levels in each dimension. It is postulated that the actor is motivated to place the other actor and will seek information on the first (most salient) dimension to make a distinction. If no distinction can be made on the first dimension, the actor seeks information on the second dimension. With only two dimensions, if no distinction can be made at this point, the process



23

terminates with alter as "the same" as the actor. Thus, not all possible information about alter is sought out and the image that arises is a byproduct of the more local problem of defining the situation in respect to class placement of self and alter. For example, consider an HM actor with declarative knowledge in the form of the image [HM] of the relevant stratified network, an initial knowledge-state in which "everyone is like me." In the model, when this actor interacts with a HH actor, the second dimensional information is employed and the image, as a by-product, transforms to [HH HM]. Similarly, suppose later an alter is HL. Then a transition occurs from [HH HM] to [HH HM HL]. If the actor next interacts with a MH, MM or ML actor, there is no need to go to the second dimension to make a distinction, and the image, as a by-product, becomes [HH HM HL M]. Note that three classes in the actual composite ordering - MH, MM, ML - are merged. Eventually, assuming that the actor interacts with a lower-class actor, the image will become [HH HM HL M L]. This will be the stable image in the sense that all subsequent interactions will maintain it (under given conditions, including no social mobility by the actor). When all nine of the possible class locations of the focal actor are considered, each such actor arrives at a stable image that depends upon that location and that merges other classes (see Figure 2). Each generated equilibrium class-based perspective on the class system is a H.

M.

HH HM HL

H

H

M

MH MM ML

M

L

L

LH LM LL

FIGURE 2 A generated structure of images of stratification (H. is HH, HM or HL, M. is MH, MM, or ML, L. is LH, LM, LL).


24


homomorphic reduction of it. The formal model thereby provides a generative account of the phenomenon pointed out by Bourdieu (1990b: p. 110), namely, "By internalizing his or her position in social space, each agent internalizes both his or her definite position - high, intermediate or low - and the structure within which this position is defined..." This sort of work can be carried further. For instance, Kosaka and Fararo (1991) treat two other problems, one empirical problem dealing explaining the tendency to middle-class self-classification in survey data and the other a conceptual problem dealing with reflexivity. We discuss only the latter effort here. In the present context, the reflexive problem may be put thus: What is distinctive about the sociological image? Is its claim to depict an "objective structure" really just one more image among the variety of perspectival images? Are we not just actors ourselves, socially located? There is no view from nowhere, so how can any view be abstracted from its locational "bias," i.e., its special distortions or reductions? In Fararo (1973: p. 366) an answer to this question was given in terms of the idea of a "universal object" in mathematics. Given a space of objects such as a class of functions, an object in the space is said to be universal if every other object in that space can be constructed in terms of it. By analogy, every number in the space of integers can be constructed from the number 1 by using repeated addition. So that object is both "just another object" and yet distinct in its universality. In using this idea in regard to images, the idea is that the sociological image is a universal object in the space of images of stratification. The axioms of the generative theory show how diverse images can be constructed, given the sociological image. All other images are homomorphic reductions of it and can be regarded as special perspectives on the structure depicted or represented by the sociological image. The linkage to Bourdieu is, however, only partial with respect to his elaborate discussions of representations. For example, the present state of formal generative modelbuilding does not treat symbolic struggles (Bourdieu, 1989). When social mobility is considered, sequential paths of images are generated in a moving equilibrium mode. The earlier acquired stable image becomes embedded within the new stable image, generated in the new circumstances of interaction. For instance, in the example given earlier, the HM actor acquired the stable image given by [HH



25

HM HL M L]. If the HM actor is then downwardly mobile to MM say, the process has the above image as its initial state but it is no longer a stable state: interactions with actors sharing the M location but differing in the secondary dimension of the overall ordering yield the new equilibrium [HH HM HL MH MM ML L] in which previously merged positions are distinguished in the image. Although this theoretical problem has not yet been fully studied, it appears that most paths of social mobility, in the cases analyzed, do not generate the sociological image. The actor has to "get around," in the social sense, and this is inhibited by the very class structure itself. Hence, the theory suggests that an image created by a researcher will be a sociological image to the extent that the investigator undergoes an extensive "virtual" mobility by interacting with actors in highly diverse locations in social space so as to acquire a perspective on the system from highly diverse locations - a process not ordinarily undertaken in the course of ordinary social interaction or "ordinary" social mobility. Of course, this idea is relevant to current discourse about social research in relation to the standpoints of diverse actors (e.g., Smith, 1990). Although this argument and the supporting mathematical developments in the cited papers are abstract and not as rich in the intuitive content of "reflexivity" that we desire as social theorists, it is surely an indication of a convergent development of a part of network structuralism and structured agency theory. Reflexive sociology, in Bourdieu's sense (Bourdieu and Wacquant, 1992; Meisenhelder, 1997), includes the sociology of sociology. And one aspect of such a sociology is related to the idea that all representation is perspectival. This is what the present discussion has tried to elucidate in terms of one development in network structuralism. Expectation States and Habitus

A particularly important research program in the group processes branch of sociological social psychology is Expectation States Theory. At its metatheory level, the fundamental commitment is to the realization of the idea of a theoretical research program as the unit of knowledge construction and cumulation and of theory growth (Wagner and Berger, 1985; Cohen, 1989; Berger and Zelditch, 1993). Within the set of core ideas of expectation states theory itself, the key notion is


26


"state-organizing process." What is meant by this phrase will be explicated below in the context of demonstrating an articulation of it with the habitus concept. The general point of this effort is to resolve another problem in the specificity of the habitus concept. Namely, missing from Bourdieu's treatment of the logic of habitus in relation to situated action is the actual demonstration of just how the habitus functions to produce the relational acts that comprise, say, the situational dominance of one class of actors over another. Without a careful explication of the relevant processes the theory is charged with the loss of voluntarism (Alexander, 1995). In particular, one difficulty is that despite the effort to retain agency in the structuralist approach, Bourdieu's habitus construct does not explain how situational inversions might arise in the context of externally given structures of dominance. For example, while men may dominate women in a macro-social sense, this does not mean that in all situations, women will defer to men. What we are suggesting is that this gap in structured agency theory can be overcome by linking Bourdieu's ideas to those of expectation states theory, because the formulation of detailed mechanisms of social interaction that function in the analysis of the dynamics of situated action is central to its research program. The relevant principle is that expectation states function in a specific mode called "state-organizing." Immediate behavior depends upon two things: the actor's expectation state and information from the immediate situation, "inputs" to the actor. In the course of acting, the state itself can change. Hence, there is a true dynamics of action. In equilibrium, the expectation state generates behavior that tends to reproduce each actor's state: the situation has been "state-organized" (Berger et al., 1977). To illustrate, consider a situation involving a panel of citizens whose task is to decide which of several candidates should receive a certain civic award. Let the panel members be strangers, but also have knowledge of each other in terms of institutionalized identities. In particular, female panel member A is known to be the CEO of a major corporation and female panel member B is known to be a homemaker. These identities, called diffuse status characteristics in expectation states theory, have generalized expectation states associated with them. Yet, because this is a fresh situation with concrete actors, members A and B



27

have no particular performance expectations for each concerning the task at hand. The theory addresses issues such as: Will the generalized expectation states be activated? Under what conditions? What will be the consequences for the over-time development of the performance expectations and for the ultimate situated influence the actors have upon each other and hence upon the group decision? In terms of the concrete example, under certain specified conditions, the CEO will come to dominate the situation of interaction with the homemaker. That is, specific performance expectations will be "derived" (by the actors) from their generalized expectation states (associated with their given identities) and govern their behavior toward each other. In terms of the properties of Bourdieu's habitus concept stated earlier, certain features of the generalized expectation states are to be noted (Berger et ah, 1985): (1) Generalized expectation states are durable. Further, they are a "present past that tends to perpetuate itself by reactivation in similarly structured practices," such as the practice of collective decision-making in discussion groups. (2) Generalized expectation states are transposable. When and if they are activated, the theory shows how they enter into the generation of performance expectations in diverse situations, and are adjusted to these situations in the same way that an abstract concept is instantiated in different concrete terms. (3) Generalized expectation states consist of structured structures. They are associated with institutional identities - thus, what passes for objectivity in social life (Berger and Luckmann, 1966). It follows that particular actors, through prior socialization in an institutional environment, will internalize identities that have associated expectations that reflect or mirror the objective (institutional) conditions of their inculcation. In short, considering each of the major meaning postulates for the concept of habitus, as drawn from Bourdieu's work, we find that "generalized expectation states" can be regarded as isomorphic to at least some component of habitus. This conceptual correspondence encourages the idea that the actual generative model-building undertaken by reference to expectation states theory is close in content as well as spirit to the idea of habitus as "generative." From such a generative

28


standpoint, the process skeleton or template of expectation state processes can be put in abstract form:


(1) E(t+\)=f[E(t);s(t)}, (2) b(t) = g[E(t);s(t)]. Expression (1) says that expectation states (hereafter also called ZT-states) at / + 1 depend upon the actor's prior Instates at time (or occasion) / and on information constructed from the situation s(t). Expression (2) says that the behavior in occasion t, denoted b(t), depends upon the current £-states and the situation knowledge. The two expressions should be thought of as generating the over-time history of behavior, and the governing £-states, in the situation. Some of the ^-states may be generalized and treated as fixed parameters, while others - such as the performance expectations - are evolving in the situation. Under certain conditions, a situational equilibrium is generated:

(3) E*=f[E';s(t)]. In terms of expectation states theory, drawing upon the interactionist tradition, the emergent definition of the situation (E*) comes to control future conduct in that situation. What we call "£-state structuralism" (Fararo and Skvoretz, 1986a) merges the social network concept with the concept of expectation states as organizing interaction. The " £ " then refers not to the expectations of a single actor in the situation, but to the entire set of such states of all actors, taken simultaneously. It is this total complex of states and the corresponding interactions that are evolving in the situation. These models are applied to analyze ongoing interactive dynamics, in which the states are not identifiable by either the observers or the actors. Nevertheless, formal methods are used to derive predictions about observable behavior, thereby testing particular theoretical models making explicit assumptions about how these states emerge and control behavior. For instance, in the analysis of a fluid interactive situation of people sitting around a table, discussing an issue in an effort to arrive at a collective approach to it, dynamic instate models generate paths of the states over the course of discussion, along with corresponding observable behaviors (Skvoretz and Fararo, 1996b).



29

The point is that the conception of actors as each acquiring and/or applying ^-states to structure the practical situation of interaction across a range of distinct contents or fields of activity, corresponds to the practical mode of the operation of habitus that Bourdieu has discussed. The three aspects of habitus stressed by Bourdieu are satisfied by generalized expectation states. Moreover, through the use of the mathematical techniques of is-state structuralism, a truly generative type of model-building helps to implement these habitus-type ideas in the context of a network of interacting agents whose various states are co-evolving in the situation. In terms of the problem of knowledge capital, note that this £-state structuralism is a formulation of a theoretical method based on merging ideas from expectation states theory and social network analysis (Fararo and Skvoretz, 1986a). Hence, it is in the zone of overlap of two traditions, sociological social psychology and network structuralism. In addition, there is a further overlap: the seemingly huge distance between structured agency theory and expectation states theory is reduced by the articulation of two key ideas from their programs, namely habitus and £-state-organizing process, respectively. In short, habitus is a state-organizing process and expectation states are components of habitus. Summary and Characterization of Generative Structuralism

The interconnected ideas we have presented represent a first phase of our generative structuralist approach. Let us summarize, beginning with the last contribution. First, habitus and expectation states - key constructs of two research programs in different traditions - are convergent in certain respects. Expectation states satisfy the meaning postulates for habitus and formal methods such as those of .E-state structuralism link expectation state processes to network thinking. It is not one actor in a situation whose states are changing over time, but the entire set of interrelated, interacting agents. Second, in the formal treatment of the sociology of representations, there is panoply of generated images, arising through interaction in a structured field or space such that each such equilibrium image mirrors the space from a perspective given by position in that space. The formal generative theory does not simply assert these latter generalizations, they are derived


30


through a generative process. While the expectation states can be regarded as part of the practice component of habitus, the image states can be regarded as part of the representational component of habitus. In both instances, there is a type of model-building involving both generativity and structure. Third, this key feature of generative structuralism was also implicated in our treatment of unit-institutions, each a generative entity comprised of a distributed production system in the sense of the Production System Principle. Here structure and agency are combined in that the components of unit-institutions are packages of tacit procedural knowledge that comprise both part of institutional structure and part of the actor. Each unit-institution is a grammar of institutionalized social interaction. It can be "multiply embodied" in the sense of being distributed over distinct sets of actors, either through time or space. This conception of institutions was linked to Giddens's concept of structure. Thus, the discussion linked a development within network structuralism - in this case, derived by merging Chomsky-type grammatical thought with Newell-Simon cognitive process thought to a key concept of structured agency theory. To add what we have already considered, it is important to explicate what these ideas drawn from structured agency theory, network structuralism and sociological social psychology have in common. Generative structuralism reveals a deeper structure of shared modes of thought in social theory by isolating and synthesizing these elements. First, each exemplifies a commitment to a concept of generativity in social theory. In the formal approach developed within network structuralism, this generativity is implemented in formal models that exhibit the recursive character of process. This means that the state of the system at present, coupled with inputs, is changing into a new state. Generative theory traces out such sequences of states. The phenomena to be explained are regarded as outcomes of such a process, with their variable properties dependent upon the parameters of the recursive process. In the expectation states theory program developed within sociological social psychology, generativity is implemented by the construction of interpersonal process theories. These set out general postulates involving theoretical constructs, such as generalized expectation states that satisfy the postulates as to their activation in abstractly described



31

social situations within the scope of the theory. Empirical generalizations are explained as outcomes of such interpersonal processes. In structured agency theory, generativity is implemented by the proposal of concepts intended to overcome historical dualisms. The emphasis is on social practices involving both agency and structure. Generativity is understood in Chomsky-like terms, namely the proposed concepts appeal to a conception of some underlying structure of immanent rules that accounts for large-scale social order. So, in the their variant ways, each of the traditions has research programs that emphasize generativity. In generative structuralism, the intellectual distance between these variants is narrowed, offering the possibility of subsequent further episodes of constructive theoretical and empirical work. Secondly, in each of the three traditions, generativity is closely linked to social reproduction. As understood within the formal approach of network structuralism, a nonlinear dynamical system can have various types of "attractors," as they are sometimes called, in which certain state descriptions are reproduced over time. For instance, there may be a stable cycle of behavior. In expectation states theory, once expectations stabilize they tend to produce action that reproduces the properties of the situation that gave rise to those expectations. In structured agency theory the acquisition and operation of habitus (in the Bourdieu contribution to this theory) accounts for the reproduction of structures, while Giddens stresses how unacknowledged social conditions lead to actions that have unintended consequences for the reproduction of large-scale social systems. Again, generative structuralism closes the intellectual distance between these three variant implementations of theoretical interest in reproduction. In short, considering all three traditions, there is a shared orientation to generativity and reproduction. And social reproduction is linked to structure. Generative structuralism is a generalized approach that articulates and connects these common features. Most abstractly, the key presuppositional principle of generative structuralism is the notion of a nexus of actualities such that transformations involve changes in patterning both within units and among units of the nexus. For example, George Herbert Mead's idea that human evolution is essentially a co-evolution of mind, self, society and


32


symbolism, is in the spirit of generative structuralism because it regards not only society (institutions) and symbols (culture) as generated in interaction but also it is in interaction that the fundamental unit-properties of mind (thinking) and self (reflexive consciousness) are generated. Structures of self, mind, institutions and symbols are dynamic equilibria relative to appropriate state descriptions of each of them. Mead's fundamental self-other frame of reference is perspectival (Strauss, 1977): the actor looks out on the co-created world and finds particular and generalized others, where the latter are particular configurations of social objects as known from that actor's perspective. The social whole is a real togetherness of all the perspectival standpoints in the interactive nexus. The analyst strives to represent the social generativity that accounts for the various features of the social nexus. Such a "foundational" characterization may seem remote from real research, yet a process worldview plays a definite role in shaping our orienting strategy. At the level of specific theoretical models, generative structuralism implies a focus on a dynamic network of interconnected units such that key properties of units are emergent features of their relational patterning to other units. What has been demonstrated here is that in several ways we can make progress in regard to the problem of knowledge capital by building on three traditions - structured agency theory, network structuralism, and sociological social psychology - thereby advancing our ability to create models embodying structured agency. The perspective characterizing this synthetic effort is generative structuralism. It amounts to drawing together, at a highly generalized level, common elements of three traditions, as indicated above. At the same time, however, these common elements do not imply that the complete synthesis of these traditions is involved in this effort. Structured agency theory, particularly in its Bourdieu wing, is strongly linked to critical social theory (Calhoun, 1995) and thereby to a diverse set of sociological interests connected with modern culture, politics and identity. Network structuralism, particularly in its social networks wing, is strongly linked to a wide range of empirical research programs that benefit from its operational approach to structure. Sociological social psychology is strong on empirical methods of research, whether experimental (in the group processes wing) or ethnographic (in the symbolic interactionist wing). Generative structuralism, therefore, is


33

not claimed to embrace these traditions as a whole: it is an evolving mode of synthesis, always incomplete, always in process.


MULTILEVEL SOCIAL DYNAMICS In our introduction to this paper, we cited three problems that prompted our work: the knowledge capital problem, the structureagency problem, and the large-scale multilevel social dynamics problem. Then we went on to show how our approach involves building on the current state of social theory and social knowledge, thereby addressing the knowledge capital and structured agency modelbuilding problems. We now embark on our second major task. Namely, we address the problem of large-scale multilevel dynamics. What has been presented in the prior part constitutes an initial phase of formulating the approach of generative structuralism, a kind of initial synthesis. A phase shift is now required to apply the approach to the large-scale multilevel dynamics problem. The Generative Structuralist Approach to the Problem

Our aim is to extend the generative structuralist approach to treat large-scale macro-social multilevel dynamics. We must implement the two foci of generativity and social reproduction. In particular, largescale dynamic equilibria have to be accounted for. As adapted to macro-dynamics, we anticipate variations in the mode of formal representation from theory to theory, but with the invariant feature of an explicit treatment of interactions, relations, and institutions. To realize structured agency in these theories, a knowledge-based model of the actor is to be used in order to produce a series of social interactions. These interactions in turn will be linked concretely to a mesostructural model which may be based on a social network, population sorting, or other formalism. These models, in turn, are expected to produce concrete effects at the actor level, as well as to give rise to macro-level system behaviors. Those behaviors, in turn, are treated as ensemble properties or as more complex aggregations of mesostructures. Thus, while some theories may treat more levels than others, a micro-macro model must operate on at least two levels simultaneously with explicit linkages between them (Markovsky, 1997). This

34



is in agreement with Coleman (1990) but the generative structuralist approach more strongly emphasizes dynamic linkages, as we shall see. Generating Cultural Stratification We shall now construct and analyze a theoretical model that operates within the guidelines of generative structuralism and which directly attacks the problem of multilevel complex dynamics with sociocultural content. This content is introduced by first considering a passage by Pierre Bourdieu (1968: p. 610): If such be the function of culture and if it be love of art which really determines the choice which separates, as by an invisible and insuperable barrier, those to whom it is given from those who have not received this grace, it can be seen that.. .their true function [is] to strengthen the feeling of belonging in some and the feeling of exclusion in others. In the above passage, Bourdieu speaks of artistic perception and, in particular, the way in which institutions such as art museums act to perpetuate a division between those who (by virtue of their education and social class) are trained to "decode" artistic works and those who are not. This statement, however, might as easily have been applied to rules of etiquette, occupational skills and credentials, or standards of formal research. Cultural knowledge - both in declarative and procedural form - serves to bind its possessors together (Durkheim, 1964; Collins, 1988: Chapt. 6; Carley, 1991) by creating the basis for social interaction, thereby also forming an "invisible and insuperable barrier" that prevents certain interactions from taking place. Such barriers, in turn, serve to perpetuate cultural differentiation itself, and hence reproduce a system of stratification which overrules - and outlasts - the efforts of any individuals passing through it. In the following section, a simple theory of cultural stratification will be presented which draws on these ideas, and results from a computational model will be shown. These results will serve to verify some of the intuitions regarding cultural stratification presented by Bourdieu (1968) and others (e.g., Collins, 1975), and will suggest some general principles governing the emergence of stratification in social systems. Because this model is founded at the micro level, it also relies heavily on insights from micro-interactionism (Blumer, 1969; Skvoretz and Fararo, 1996a). It is hoped that work such as this will demonstrate the



35

rich possibilities of generative structuralism for establishing a linkage between micro-level symbolic models and macro-level cultural theory. In the following multilevel dynamic model, in a sense we are constructing what amounts to an artificial society (Epstein and Axtell, 1996). We are employing a development located in this paper within network structuralism: computational sociology. This new development stresses the construction and analysis of complex process models through the technique of computer simulation. When process models are constructed with a view to analysis of them by simulation rather than by deductive analysis (which is prohibitive given the model complexity), the theoretical models are termed "computational." Simulation is best thought of as a method of nongeneral but logical derivation: consequences are generated, strictly from assumptions employed in constructing the model, but they are not derived theorems. Still, general patterns of consequences can be inferred from repeated simulation runs of the model. We will illustrate these general ideas in the context of the following model. We posit the nature of our agents and have them interact in a structured setting such that certain outcomes are generated that are of particular interest from a social theoretical standpoint, as just indicated. The complete multilevel model has three component models, one per level - in our case, three - each with certain stipulated properties: (1) an actor model that, for instance, endows actors with communicative capacities; (2) an interaction model that, for instance, treats the problem of coordination in substantive activity; and (3) a societal model that, for instance, treats flows of actors through organizations, each of which is of a certain institutional type and is also a network of "coupled" actors (a context for communication and coordination). We now proceed to formulate our model by stating how each of these levels is formally represented and how they dynamically interrelate. The Actor Model Following our earlier discussion, we seek to model individual actors as agents embodying immanent rules - who nevertheless may change their behavior over time. We will distinguish communicative and substantive aspects of behavior in our actor model, satisfying four conditions: (1) actors must be capable of learning via an enculturation or influence process; (2) they must be capable of communicating with


36


each other; (3) they must be able to engage in substantive behavior; and (4) actor communication must be able to influence behavior. Individually, these requirements are fairly trivial to satisfy. Influence models (e.g., Friedkin and Cook, 1991; Anderson, 1959) have been developed in numerous contexts for quite a number of years. Multiagent models in which actors exchange information directly have been developed (e.g., Carley et ah, 1992). Whole fields of study have grown up around the notion of actors as engaging in substantive behavior; and signaling in multiplayer games is these days considered mundane enough to qualify as textbook science (Binmore, 1992). To combine all of the above in a single model, however, is more difficult. Work by Carley et al. (1992) has included communicative actors engaging in substantive behavior, drawing upon the plural-SOAR agent model, itself initiated by the production system theorist Newell (1990). Other work in the distributed AI community, as well, has considered this sort of problem, but the frameworks in which these models have been applied have generally been very limited, containing only a few agents and relatively little opportunity for socialization. In order to construct a model of culture, then, we require a framework which satisfies the above requirements, but which is also simple enough to permit implementation at the population level. Our solution to this problem is as follows. Instead of working directly with production systems, we model each actor in terms of a finite-state automaton, an abstract "machine" that (as noted earlier) corresponds to a simple type of production system. The automaton possesses some number of states and makes transitions among states, each of which is tied to a communicative or substantive behavior through a rule. State transitions are contingent on the current state and the situational "input" to the actor. Thus, we have a dynamic network of agents, each capable of communicating, acting, and learning. This representation has the added advantage of being fairly compact, so that implementing the hundreds of agents required for a simulation of emergent cultural stratification is feasible. To see how this works, consider a simple five-state automaton as shown in Figure 3. Each state of the automaton (represented by a circle) implies a certain action (indicated within the circle). Since actions may be communicative or substantive, both types are included. The first type of behavior involves sending a communicative token


37

Receive 1

\ ^ Send \ —•-~^Re«xivel ^ 1 T o k e n 1 H" \ Receive