Vous êtes sur la page 1sur 35

Foundations for a Multi-Agent Transactional Theory of Literacy John E. McEneaney, Ph.D.

Department of Reading and Language Arts Oakland University, Rocheseter, MI 48309-4494 Email: mceneane@oakland.edu; Voice: 248-370-4155 Transactional theory represents one of the most significant frameworks for conceptualizing reading and writing developed in the twentieth century and continues to have a powerful influence on literacy researchers and practitioners. At the core of transactional theory is the notion that meaning is produced in the transaction of a reader with a text. In contrast with other models that locate meaning within the text and conceive of reading as the extraction of that meaning, transactional theory defines meaning as an emergent property of a reading transaction, requiring simultaneous attention to both reader and text as the reading process unfolds. Transactional theory has, however, been motivated primarily by a concern with aesthetic interpretation of literary works, while research focusing on efferent interpretation (i.e. expository reading) has continued to be dominated by more traditional text-centric frameworks, often relying on text propositionalization as a basis for analysis (Kintsch & van Dijk, 1978; Graesser, 1985; Meyer, 1985). The purpose of this paper is to outline a general transactional theory of literacy that seeks to more effectively integrate aesthetic and efferent forms of reading, as well as providing a basis for the principled integration of formal methods, including those based on text propositionalization. Although the framework presented accounts for reader transactions in both online and offline reading environments, the impetus for this work is rooted in the inadequacy of existing theory in addressing special aspects of reading and writing online that have challenged traditional notions of author, reader, and text (McEneaney, 2001; Landow, 1992; Reinking, 1998). As developed, the theory is general enough to subsume transactions in traditional print as well as the more dynamic transactions characteristic of online reading. Two features of the proposed framework most clearly distinguish this work from predecessors. One feature is an explicit acknowledgment of roles for both human and non-human agents as participants in online literacy transactions, an element that responds to the genuinely interactive potential of online literacy transactions (Reinking, 1998, p. 1). A second feature is the adoption of formal methods, both to achieve greater generality through abstraction and to define a theoretical vocabulary that will serve to bridge the fields of interactive system design and literacy theory, both of which play crucial but, to our disadvantage, still largely incommensurate roles in defining online literacy theory and practice. The theory presented builds on prior work in the areas of multi-agent software systems (Weiss, 2001; Bradshaw, 1997; dInverno & Luck, 2001), structural theories of hypertext (dInverno & Priestley, 1995; Halasz & Schwartz, 1994; McEneaney, 2001, 2002), and transactional theory, as articulated by Rosenblatt (1978, 1994, 1995). The proposed work has three main objectives. The first objective is to develop a novel interpretation for an existing formal system (dInverno & Luck, 2001) originally defined as a theory of multi-agent systems that draws on an informal theoretical framework defined by Rosenblatt (1995, 1994), in effect, defining a formal transactional theory of literacy. A second objective is to extend the reinterpreted formal system by incorporating elements specific to hypertext with the goal of providing support for recently developed

Multi-Agent Transactional Theory - Page 2 of 35

navigational modeling techniques (McEneaney, 2001). The third objective is to explore the resulting theoretical system as a predictive and explanatory framework in support of new ways of thinking about literacy in electronic environments that include multiple interacting agents, transactions involving human and machine agents, and even transactions based on complexes of non-human (i.e., machine) agents. Although transactional theory has traditionally been framed as a pragmatic naturallanguage phenomenology of reading, there are at least two practical reasons for formal methods, in addition to the more obvious theoretical arguments that might be offered. Perhaps the most compelling of these reasons is the remarkable transformation of literacy practice since the early 1990s, with the widespread use of computer-based literacy tools and the Internet. These technologies challenge some of our most fundamental assumptions about readers and text and all are, to a greater or lesser extent, grounded in formal frameworks. Existing theory in literacy has, however, generally operated as an independent theoretical layer, floating above the theoretical substructures that support the new technologies of literacy that are having such a profound impact on literacy practice. While it would be both unwise and unproductive to demand that all of literacy theory as we know it be defined within a formal framework, it seems quite clear that there are points of contact where literacy theory and the formal frameworks established by the new technologies of reading and writing can and should be connected and that these connections will help us redefine literacy theory in ways that more adequately acknowledge the power and influence of technology in shaping literacy theory and practice. The case for a tighter integration of literacy theory and its technological substructures finds a second flip-side argument when viewed from an engineering perspective. Technology is always situated in a context that includes phenomenological (e.g., Anderson, 1997), anthropological (Marshall, 1997), sociological, and other dimensions of usability. Human factors engineering is predicated on the idea that effective technology integration requires an understanding of these human dimensions and that ignoring them may doom even the most technically elegant solutions. Consider, then, the explosion of literacy related technologies and the fact that nearly all are based on informal conceptualizations of reading and writing with origins in print technologies. The difficulties and frustrations readers experience in online reading environments may no longer seem surprising in such a context and the principled integration of literacy theory with technological research and development takes on significant urgency. It is simply not enough to build better systems. We need to frame theory that is large enough to provide a broader view that incorporates both systems and users. There are, of course, other more purely theoretical reasons for interest in formal methods as well. The use of rigorously defined formal abstractions as a basis for theory supports a more cumulative approach to theory development that allows theoretical work to develop in a systematic fashion. Formal theory also has the advantage of leading much more naturally to computer-based simulations (Just & Carpenter, 1992; McClelland & Rumelhart, 1988; Seidenberg, 1992), and discourse analysis (Kintsch & van Dijk, 1978; Meyer, 1985), which have played important roles in the development of theoretical models in reading. Moreover, although formal systems and simulations have been associated with specifically quantitative research programs, there is nothing in the formalists methodology that requires an a priori commitment to quantitative science.

Multi-Agent Transactional Theory - Page 3 of 35

Formal models can serve to clarify theoretical concepts and relationships, as well as provide a broader meta-theoretical framework to support structured comparisons of models, revealing common foundations and model-specific differences. In the present context, although a user interface designer, a sociologist, linguist, and a literacy educator would likely conceptualize a readers use of reading materials in very different terms, the formal approach to model building developed in this paper can be applied in ways that allow each one to develop more specialized views that illuminate specific aspects of the reading event, while still acknowledging a common core of ideas that bind all these conceptualizations together under the general heading of literacy. It is, therefore, the purpose of the high-level theory presented in this paper to formally define transactional theory to support and promote the principled integration of multidisciplinary perspectives on literacy, as well as promote greater integration of structural levels within literacy theory proper. An introduction to formal specification in Z The notational system adopted in this work is the Z specification language (Bowen, Dunne, Galloway, & King, 2000; Spivey, 1988, 1989), founded on set theory and firstorder predicate logic, and originally developed for application in the formal specification of software systems (Spivey, 1992; Singh, Rao, & Georgeff, 1999). Z has, however, also been applied in numerous other areas, resulting in a well-developed literature grounded in both theoretical and applied studies. The use of Z as a general specification language has also recently been enhanced by the appearance of a formal standards document for Z notation (Z Standards Panel, 2000). Most immediately relevant in the present context is the application of Z as a specification language for multi-agent systems (dInverno, Kinny, Luck, & Wooldridge, 1998; dInverno & Luck, 2001) and hypertext theory (dInverno & Priestley, 1995), as the transactional theory developed explicitly attempts to model the interaction of human and software systems, in effect treating problems in human-computer interaction as literacy events involving human and machine agents. One advantage noted for formal theory is the support it provides for subsequent theoretical extension and refinement. In the present case, the transactional theory developed is based on and extends work by dInverno and Luck (2001) in defining a formal theory of multi-agent systems, a well developed and rapidly growing subdiscipline within computer science that addresses the interaction of functionally independent and potentially intelligent programs designed to operate within a larger electronic environment. Multi-agent systems theory, also referred to as distributed artificial intelligence (Weiss, 2001; Ferber, 1999), has been productively applied in robotics, user interface design, computer supported cooperative work, and other areas. The intent of the present work is to apply this approach in the domain of literacy, both as traditionally understood to refer to a single human reader, and as extended to incorporate multiple interacting literacy agents. As a result, the proposed model can be considered a specialized application of agent systems theory into the domain of humancomputer interaction, as well as a formal specification for transactional theory. Given that the present work builds on fairly extensive prior development, the review of relevant background is limited to concepts and relations that are most crucial in subsequent development. Accordingly, some formal elements are introduced with only

Multi-Agent Transactional Theory - Page 4 of 35

limited explanation, although in these cases the reader is referred to work that defines these elements in full formal detail. Moreover, in every case where formal elements are introduced, the narrative always provides an informal explanation so that the reader is not required to consult references to follow the informal outlines of the argument presented. This interleaving of formal explication and informal commentary is an essential aspect of formal specification in Z (Spivey, 1992, p. 2), with formal elements clearly demarcated by typographic and layout conventions. Logical foundations of agents and multi-agent systems The purpose of this section is to present a selective overview of the framework defined by dInverno & Luck (2001) as a formal foundation for a general theory of agency. Taking the elements of the Z language as a given (see appendix for an overview of Z), dInverno and Luck begin by defining primitive concepts (attributes, actions, and motivations) and an entity hierarchy defining a progressively more specific ontology (see Figure 1). Informally, an entity is anything that can be defined by a set of attributes. Objects are entities that have the capability to alter the state of their environment. Agents are objects that have the added capacity to be assigned or adopt goals that dispose them to act on their environments in support of goals. Finally, autonomous agents are agents that have a capacity to create new goals based on motivations that are not subject to the direct control of other agents. All of these elements are presumed to operate within an environment, defined simply a collection of entities Interpreting this ontological hierarchy in terms of traditional literacy environments, an entity is simply a generic way of referring to anything in a literacy environment, without saying anything whatever about its role in that environment. Objects, on the other hand, are entities that have a capability of effecting change on their environments, but this capability is not directed by an internal goal or motivation. In transactional theory, for example, while a book has a capacity to influence a literacy environment, this capacity cannot be realized until a reader engages the book in a literacy transaction (i.e., a book is just Entities paper and ink (Rosenblatt, 1994, p. ??) until a Objects reader with a goal picks it up. According to a Agents transactional view, a book that was formerly simply an object in the environment is Autonomous transformed into a literacy agent when a reader Agents (i.e., an autonomous agent) applies it (i.e., she reads it) in support of a goal or objective that motivates a literacy transaction. A transactional literacy system, therefore, is based on a fourtiered ontological hierarchy with an entity Figure 1. Entity hierarchy (adapted from serving as the most general category that dInverno & Luck, 2001.) serves as a template for three more specific category types. Objects are defined as literacy artifacts (i.e., paper and ink) that are not presently (but may in the future) be engaged in literacy transactions, while agents are objects that participate in literacy transactions, and autonomous agents are those agents that have motivations that allow them to initiate literacy transactions.

Multi-Agent Transactional Theory - Page 5 of 35

Primitive elements, definitions, and schemas for entities, objects, agents, and autonomous agents are presented in Section Z1 below. Primitive elements are defined as given sets without indicating how they are represented in any specific application. Informally, attributes are features of the world, actions are events that can alter the world, and motivations are dispositions to act in support of adopted goals. Given sets are indicated with the use of square brackets (e.g., [Attribute] is the set of all possible attributes). Definitions of basic concepts employ the symbol == with the term being defined on the left and defining characteristics to the right. Schemas define more complex concepts and may include an upper declarative part that identifies entity variables representing elements that make up the entity defined by the schema and a lower predicate part that specifies required properties of the entity variables in the declarative part of the schema. Schema variables are defined using an a : B format (e.g. attributes : Attribute) where a represents the defined variable and B specifies the variable type. The four schemas presented in the following section define the four progressively more specific concepts described earlier (entity, object, agent, and autonomous agent). Each schema consists of a labeled and partitioned frame. The label at the top of the frame is the name of the schema. The top part of the schema frame, referred to as the declaration, identifies components that serve as elements of the schema, with each element assigned a variable name and a data type. The bottom of the schema frame, referred to as its predicate part, specifies properties that are required of the entity variables in the declarative part of the schema. The Z specification language presumes certain general mathematical data types (e.g., integers, natural numbers, sets, sequences, etc.) and operators from predicate logic and set theory (e.g., , , , etc.), but also supports new types on the basis of given sets and definitions. In the case of the Entity schema, for example, the declaration identifies four components that include attributes (representing permanent features of the entity that are not determined by its situation in an environment), capabilities (actions the entity can take), goals (defining a desired environmental state), and motivations (defined in terms of the Motivation given set.) Note, however, that the Entity schema, also includes a predicate section (i.e., the bottom part of the schema) requiring that an entity have at least one attribute (formally, that its set of attributes not equal the empty set), although this requirement is not specified for its other components (i.e., an entity may not have capabilities, goals, or motivations.) In effect, the one requirement for an entity is that it must have at least one attribute, and while entities can have capabilities, goals, and motivations, the presence or absence of these elements are not critical in defining an entity. As indicated in the following schemas, however, in order for an entity to be considered an object it must have attributes and capabilities, agents must have attributes, capabilities, and goals, and an autonomous agent must have all these and motivations. This conceptual hierarchy thus has a nested quality like that displayed in Figure 1 and formally expressed by the relation: autonomous agents agents objects entities. Primitives [Attribute] the set of all possible perceivable features [Action] the set of all possible actions that can change the state of an environment [Motivation] the set of all possible motivations that can lead to the adoption of a goal

Multi-Agent Transactional Theory - Page 6 of 35

Definitions Environment == 1 Attribute (a non-empty set of attributes.) Goal == 1 Attribute (a non-empty set of attributes defining a desired environment.) Schemas Entity attributes : Attribute capabilities : Action goals : Goal motivations : Motivation attributes { } Object Entity capabilities { } Agent Object goals { } AutonomousAgent Agent motivations { } Reviewing, the schemas above define an entity as any thing that has an attribute or feature, while an object refers to something more specific that can act on the world (e.g., a physical entity such as a stone or book). Agents are objects that have a goal that has either been assigned by another agent (e.g., a paperweight I appropriate to keep my documents from blowing away or a recipe book I read to cook a meal) or have the capacity to generate goals (e.g., a human reader). Human beings would usually be characterized as autonomous agents since these entities have the capacity to generate goals on the basis of motivations (defined as a given set of primitives). It is, however, important to return to an idea mentioned earlier but deserving special emphasis given its significance in a transactional theory of literacy: the status of agency is transient. An object serving as an agent in support of a goal (e.g., my office paperweight or my recipe book) will cease to function as an agent if I no longer use it to achieve my goals. In effect, an otherwise neutral object can be promoted to the status of a special kind of agent (i.e., a server-agent, which will be formally defined at a later point) if it is appropriated for use by an autonomous agent. Moreover, it can subsequently be returned to the status of a neutral object if the goal that led to the creation of a server-agent is dropped. In effect, agency, particularly the special form of agency that traditional print assumes in a literacy system is an event in time marked by boundaries that functions in the same temporal sense that transactional theory speaks of the poem as an event in time (Rosenblatt, 1994, p. 12). Thus far, the mapping of this specification onto transactional theory is straightforward. Entities serve the same general-purpose role they do in the multi-agent framework. Objects have the same kind of status in transactional theory, as things unconnected

Multi-Agent Transactional Theory - Page 7 of 35

with goals. But reading and writing (as well as any other meaningful activity) necessarily involve goals and therefore when we refer to literacy objects (e.g., books and writing implements) used by readers and writers who engage in literacy-related activities, a transactional theory stipulates that these objects are transformed by our appropriation of them into agents by virtue of our goal(s) in using them. A book is an object by virtue of its physical nature, since falling from a high shelf it may itself be damaged or damage something else and thus alter an environment. However, when a book is used as a book, a reader engages it in a literacy act (i.e., a transaction) that is directed by a goal and this transforms the book-as-object into a book-as-agent. The goal-directed character of literacy ensures that anything appropriated in support of a literacy transaction becomes an agent and, in this sense, even a traditional printoriented perspective on literacy views a literacy transaction as a form of multi-agent system. Moreover, as will be examined in greater detail later, multi-agent transactional theory highlights the agency of text to an even greater extent in electronic reading environments where reading materials may incorporate background processing (i.e., interactive programming) to support the generation and adoption of goals by the text, thus potentially elevating text material to the level of an autonomous agent that enjoys (relatively speaking) the same ontological status as a human reader. There are, therefore, two radical ideas behind the multi-agent literacy theory presented in this paper. One idea (adopted directly from Rosenblatt) is that a book-as-object has no intrinsic meaning apart from when it is appropriated by another agent in a literacy transaction. The second radical idea (a consequence of recently developed literacy technologies) is that text may itself operate as an autonomous agent in a literacy transaction. Neutral objects, server agents, and multi-agent systems Returning to formal explication, three new schemas are presented, defining neutral objects, server agents, and multi-agent systems. The first two of these schemas elaborate on the entity hierarchy already defined, with the multi-agent system schema incorporating all of the entities defined thus far. Neutral objects and server agents serve to refine concepts already introduced by the object and agent concepts (and informally alluded to earlier.) The problem with the concepts of object and agent as already defined is that, since these ideas are hierarchically defined (i.e., every agent is also an object), we do not yet have a term to refer to those objects that are not agents (i.e., neutral objects) or those agents that are not autonomous (i.e., server agents), both of which represent important distinctions in multi-agent systems and transactional theory. Neutral objects are simply objects that have neither goals, nor motivations and server agents are those agents that have no motivations (but, by definition as agents, have at least one goal that has been assigned by an autonomous ganet in its environment). Informally, a neutral object is something that is currently peripheral to theoretical concerns. In the context of transactional theory, neutral objects are those elements in the environment that fall outside of the scope of the literacy transaction. It is, however, relevant to note that the status of a neutral object can change as a readers attention changes. Server agents, unlike autonomous agents, do not have a capacity to generate goals but can acquire goals if they are engaged by another agent. As Rosenblatt notes (1994, p. ix), books are simply objects of paper and ink until a reader initiates a

Multi-Agent Transactional Theory - Page 8 of 35

transaction and, in doing so, provides the book-as-object a purpose or goal and transforms it into a server-agent. One way I might transform a book-as-object into a server-agent, however, might be to use it as a paperweight, but this would not be relevant to a theory of literacy. If, however, on scanning my desktop I pause to read the spine of my paperweight, the book-as-server-agent might acquire a new literacy-related goal that may bring it within literacy theory. With the specification of the entity hierarchy complete, it becomes possible to define the larger concept of a multi-agent system upon which much of the present work builds. A multi-agent system incorporates sets of the entities defined thus far that must, in addition, meet a number of conditions. Specifically, a multi-agent system can include entities, objects, agents, autonomous agents, neutral objects, and server agents. According to the predicate section, the four initially defined theoretical elements (autonomous agents, agents, objects, and entities) must adhere to the hierarchically nested relation illustrated by the Venn diagram in Figure 1. Agents must be restricted to autonomous and server agents; objects are restricted to agents and neutral objects. There must also be a minimum of two agents, at least one of which must be autonomous. Finally, the two required agents must share at least one goal in common. NeutralObject Object goals = { } motivations = { } ServerAgent Agent motivations = { } MultiAgentSystem entities: Entity objects : Object agents : Agent neutralobjects : NeutralObject serveragents : ServerAgent autonomousagents: AutonomousAgent autonomousagents agents objects entities agents = autonomousagents serveragents objects = agents neutralobjects #agents 2 #autonomousagents 1 a1, a2: agents a1.goals a2.goals { } As before, the specification aligns with transactional theory as traditionally understood. It acknowledges the role of readers goals and attention in excluding portions of literacy environments as outside the scope of literacy theory (i.e., neutral objects). It also accommodates traditional literacy theory by including literacy systems that distinguish different roles for text as a server agent and the reader as an autonomous agent. Moreover, while it requires that a literacy system include a minimum of two participating

Multi-Agent Transactional Theory - Page 9 of 35

agents that share at least one common goal and a minimum of one autonomous agent, this leaves open the significant possibility (explored later in the paper) that literacy transactions may be defined that include multiple autonomous agents (and indirectly, the notion that text may itself be autonomous). Before moving on to more technical details, however, it will be useful to review the ontological framework that has been defined and highlight three elements with special roles in multi-agent transactional theory. Specifically, although the general terms entity, object, and agent will continue to serve useful roles as broader more inclusive terms to refer to items about which our knowledge is incomplete or when we want to cast a broader conceptual net, the terms neutral object, server agent, and autonomous agent will occupy special positions in multi-agent transactional theory. Neutral objects in multi-agent transactional Figure 2. An engagement chain that depicts a series of direct engagements (DE) creating an theory are associated with those most basic, engagement chain satisfying the goal of an yet still theoretically significant, elements in autonomous agent (A) using a browser (B) to a literacy environment: artifacts of paper and access a university reference page (C), which ink, and server agents are what those supports access to a remote database (D). objects are transformed into when they are engaged by an autonomous agent in a literacy transaction. Finally, autonomous agents are those entities that initiate literacy transactions, human readers in transactional theory as traditionally understood (but a significantly broadened category in the new electronic literacies.) Having considered these theoretical entities and the framework within which they operate (i.e., a multi-agent system), we are now in a position to consider how they engage and cooperate with one another. Engagements, engagement chains, and cooperations A multi-agent system has been defined in terms of the entities that operate within it and the conditions these entities must meet, but there has not been any specification concerning the kinds of relationships these entities have to one another (Castelfranchi, 1990). It is the purpose of the present section to define and describe two major categories of relationships, referred to as engagement and cooperation, that have special bearing on literacy theory, particularly as this theory is applied in the analysis of on- and off-line literacy transactions. Direct engagement occurs when an agent, referred to as a client, appropriates a nonautonomous entity (i.e., a neutral object or server agent) referred to as a server to achieve a goal by arranging for the server to adopt the goal of the client (dInverno & Luck, 2001, p. 52). Since a neutral object, by definition, has no goals, engaging it is straightforward (i.e., there is no danger of a conflict with another goal). In the case of a server agent, however, it may be that the goal of the client conflicts with one or more prior goals of the server and, under this circumstance, the engagement is more complex. Moreover, as illustrated in Figure 2, achieving a goal may involve several agents acting in an engagement chain anchored by the autonomous agent that initiates the chain.

Multi-Agent Transactional Theory - Page 10 of 35

As an example, use of a remote reference database (e.g., ERIC) by a university student would usually require a series of steps that include A) an autonomous agent establishes a goal to retrieve information from a remote database, B) the agent opens a web browser, C) a connection is opened to a university-sponsored reference site, and D) a connection is opened to the remote database itself through the university subscribed account. Throughout the process, there are conditions that must be met to engage server agents at each step. My workstation requires a certain amount of memory to open a browser. My local network connection must be available so I can open the library web site, and the remote database may admit only a limited number of users based on the university subscription. If these server agent conditions are met, however, these engagements are considered to be obligated , and I need not worry that the agents might add new conditions at the last minute or refuse my request for an engagement for any other reason (something that cannot be guaranteed in attempting to establish cooperation with another autonomous agent.) Unlike the obligated engagements of neutral objects and server agents, cooperation between autonomous agents involves a dynamic negotiation between entities that are capable of generating new goals, perhaps even in the very midst of negotiation. As might be expected, the potential for complication is considerable. Schemas for direct engagement, an engagement chain, and cooperation are provided below. A direct engagement involves a client, a server, and a goal. Although the client may be an autonomous agent, the server must be a non-autonomous agent. Constraints specified in the bottom half of the schema require that the client and server be distinct entities (i.e., an agent cannot engage itself) and that the engagement goal be shared by the client and server. DirectEngagement client: Agent server: ServerAgent goal: Goal client server goal (client.goals server.goals) An engagement chain, like a direct engagement, involves a goal but is more specific in designating an autonomous agent (i.e., autoagent) that is responsible for initiating the chain and a non-empty injective sequence of server agents (servers B, C, & D in Figure 2) such that the goal of the autoagent is also a goal of every server agent in the agent chain. EngagementChain goal: Goal autoagent: AutonomousAgent agentchain: iseq1 ServerAgent goal autoagent.goals goal {s: ServerAgent | s ran agentchain s.goals} Cooperation, like an engagement, revolves around a goal but defines a relationship between two or more autonomous agents, one of which is the agent that initiates the cooperation. Other autonomous agents that choose to participate do so by adopting the

Multi-Agent Transactional Theory - Page 11 of 35

initiating agents goal. An example of a cooperation in a literacy environment is the use of an online search engine. An initiating agent (e.g., a person) submits search terms to the search engine in an effort to locate reading material and the search engine returns a set of links. The set of links returned, however, may reflect goals of the search engine as well as the reader. Some of the links returned may actually be advertising that support the search engines goal of generating revenue that would not otherwise be included in the links returned. Readers generally acknolwedge the cooperative nature of this transaction and are not surprised to find advertising in their search results (as they are not surprised to find advertisements in magazines.) Cooperation goal: Goal generatingagent: AutonomousAgent cooperatingagents: 1 AutonomousAgent goal generatingagent.goals aa: cooperatingagents goal aa.goals generatingagent cooperatingagents Notably, in these definitions nothing has been said about whether the goal that prompted the engagement(s) and/or cooperation has been achieved. In order for this matter to have any bearing on the response of an agent, however, an agent must be able to perceive the environments within which it operates. Indeed, much of what has been presented presumes such perception (otherwise, how could an agent establish an engagement or cooperation?) but this issue has been deferred. It can be deferred no longer and, accordingly, the next section addresses perception by agents. Environments and perception by agents Thus far, little has been said about what agents can and cannot do and on what basis agents might seek out engagements and cooperations, and otherwise act on an environment. The demands of a transactional theory of literacy, however, obviously call for agents that are capable of perception and, ultimately, environmental modeling and manipulation. Fortunately, all of the apparatus needed to extend agents in the directions required have already been defined by dInverno & Luck (2001). To begin with, there is a need to provide enough detail in describing entities to distinguish intrinsic attributes of the entity from those attributes that are a result of the environment within which the entity is situated. A physical entity, for example, has transient features that define its state within a particular context (e.g., spatial location), as well as intrinsic attributes (e.g., its mass) Attributes that do not depend on an entitys environment, are referred to as its attributes while those that are determined at least in part by the entitys environmental context are referred to as that entitys situation. The Env schema defines an environment variable that takes an Environment data type (a set of attributes) with at least one feature. It also contains entities with attributes of their own. Moreover, all the attributes of entities in that environment are a subset of the attributes that define the environment as a whole. In other words, the features of the environment completely describe it, including all of the entities it contains.

Multi-Agent Transactional Theory - Page 12 of 35

Environment == 1 Attribute (restating this definition from an earlier point.) View == 1 Attribute (a View represents environmental attributes an agent perceives.) Env environment : Environment entities : Entity environment { } { e : entities e.attributes } environment When an environment contains entities, the state of the entity must account for both intrinsic features and those attributes that situate the entity within its environment. The EntityState schema therefore defines a situation as a set of attributes and the constraint section requires at least one distinct situation attribute (i.e., independent of the entitys intrinsic attributes) and that the environment must contain the entity defined (i.e., the attributes of the entity cannot be equivalent to those of the environment as a whole since the environment would then reduce to the entity itself.) EntityState Entity Env situation : Attribute situation { } attributes situation = { } attributes situation environment With an EntityState schema defined, it is possible to develop more specialized state schemas that define the states of agents. Before introducing agent states, however, it is important to consider how agents perceive environments within which they operate. In general, agents are not omniscient and therefore, not all of the attributes of an environment will be perceived by an agent. Under these circumstances it is important to distinguish possible percepts, those attributes of the environment that can be perceived, from percepts that actually are perceived, a difference that plays a critical role in multiagent transactional theory since two fundamental constructs in transactional theory, selective attention and stance, are grounded in this distinction. The AgentPerception schema begins with the inclusion of the agent schema and then introduces variables and functions specific to perception. The perceivingactions variable and canperceive function, taken together, define the perceptual capability of an agent, given an environment and the perceptual capabilities defined by perceivingactions. These elements, in effect, define the perceptual competence of an agent and thus play important roles in models of learning predicated on transactional theory, but their role in defining agent performance in specific circumstances (i.e., literacy transactions) are more modest, since they simply indicate what can be, rather than what is, perceived. Rather, the critical kernel of transactional theory, as formally expressed in this system, resides in the willperceive function, a function that incorporates both an agents goals and perceptual competencies in producing a view of an environment, where a view is an agents representation of a set of environmental attributes. More specifically, the

Multi-Agent Transactional Theory - Page 13 of 35

willperceive function is defined as a composite function that can be parsed out into a broader operation based on stance (i.e., ( Goal View) that results in a perceptual field and a more specific mapping (i.e., (( Goal View) View)) from this perceptual field onto those attributes that occupy the attention of the agent. AgentPerception Agent perceivingactions : Action canperceive : Environment Action View willperceive : Goal View View perceivingactions capabilities env : Environment; as : Action as dom(canperceive env) as = perceivingactions dom willperceive = {goals} Following up on the perceptual capabilities defined in the AgentPerception schema (and building on the ObjectAction schema that has not been described see corrigendum to UAS, section 2.3.2), the AgentAction schema defines a mapping across an agents goals, perceptual view, and environment into a set of actions the agent is capable of carrying out in that environment. AgentAction Agent ObjectAction agentactions : Goal View Environment gs : Goal; v : View; env : Environment (agentactions gs v env) capabilities dom agentactions = {goals}

Action

The AgentState schema, relying on a number of previously defined schemas, inherits a wide range of elements from these components (e.g., intrinsic and situational attributes, a capacity to act, etc.) In addition, the AgentState schema explicitly defines two categories of percepts that were alluded to earlier and represent an important distinction in transactional theory. In particular, two distinct types of views are defined, a set of attributes that represents possible percepts (posspercepts) and a set that represents those attributes actually perceived by an agent (actualpercepts). Predicates in the AgentState schema, as suggested earlier, have the effect of associating the posspercepts variable with an agents perceptual field and the actualpercepts variable with those attributes within an agents focus of selective attention. The action that an agent actually takes when in a given state is defined by the willdo function that operates across the goals, percepts, and environment of the agent, drawing on the agentactions function defined in the AgentAction schema.

Multi-Agent Transactional Theory - Page 14 of 35

AgentState AgentPerception AgentAction ObjectState posspercepts, actualpercepts : View actualpercepts posspercepts posspercepts = canperceive environment perceivingactions actualpercepts = willperceive goals posspercepts perceivingactions = { } posspercepts = { } willdo = agentactions goals actualpercepts environment Building on the AgentState schema, the AutonomousAgentState schema is an extension that incorporates the elements of an AgentState schema, while adding perception and action specializations for autonomous agents and defining actions as a function of an autonomous agents motivations, goals, percepts and environment. In the interests of brevity, a number of the state schemas referred to are not defined. These schemas are, however, developed as specializations in a matter analogous to the derivation of autonomous agents from entities. Readers with an interest in formal details are referred to sections 2.3.2 to 2.5.5 of dInverno & Luck (2001). AutonomousAgentState AgentState AutonomousAgentPerception AutonomousAgentAction willdo = autoactions motivations goals actualpercepts environment Interpretive Aside 1 The scope of the framework developed thus far is now broad enough to sustain some genuinely interesting problems and issues with direct relevance to a transactional theory of literacy, particularly as it relates to similarities and differences in traditional print and online reading environments. Although formal concepts specific to transactional theory have yet to be presented, it should be apparent that one essential aspect of the proposed framework is the idea that both reader and text operate as agents within a multi-agent system that defines a literacy environment and this has a number of immediate implications. One implication is that, while some forms of reading will cast the reader in the role of an autonomous agent and the text in the role of a supporting server agent (in traditional print, for instance), some electronic environments (although, by no means all) may create circumstances where both reader and text operate as autonomous entities, resulting in two completely different kinds of transactions, one based on an engagement (or engagement chain) and the other on a relationship of cooperation. In reading a newspaper, for example, I can enforce my goal of learning about the weather by only attending to material that addresses my interests and ignoring the rest (e.g., other articles, advertisements, etc.) When my reading transaction involves a cooperation with another autonomous entity, however, I may be obliged to relinquish some of my authority. As an illustration, consider the agendas search engines and online ecommerce sites sometimes bring to reading transactions. An online catalog like

Multi-Agent Transactional Theory - Page 15 of 35

Amazon.com can attend to my interests and respond in ways that are explicitly intended to shape my actions so as to bring them into alignment with the goals of the ecommerce site, rather than simply satisfy those that originally led me to the site. Another related, but generally less benign, example concerns the practice of online search engines that mix paid advertisements with the search results that they report, sometimes without providing any clear indication that the most prominently displayed links are not actually a result of the search that was motivated by the readers goal, but are displayed specifically to support the goal of the search engine to generate advertising revenue. A second implication that will be explored in some detail at a later point is that the capacity to represent (i.e. model) both the transaction and the agent with which I transact meaning may have considerable significance in determining whether I succeed in achieving my goal. In a sense, this is a case of critical reading, and relates fairly directly to the problem of autonomous search engines noted above. If, as a reader, I have developed the kind of critical reading skills that allow me to recognize when search engines are manipulating transactions in ways that may be incompatible with my own objectives, I am in a position to more effectively achieve my own goals. If, however, I am unable to model the agents with which I transact, it is unlikely I will even be aware that what I perceive to be an engagement is, in fact, a cooperation (and perhaps a manipulative one at that.) Sociological agents and system modeling The preceding section calls special attention to a question concerning multi-agent systems: What is it agents might be said to know about other entities and/or processes that occur in the system and how is that knowledge applied in the pursuit of goals? Although there has been some limited discussion focusing on perception, nothing has been said about how agents use perceptions to achieve goals. As before, although the definitions provided are sufficient to support simple reflexive or tropistic agents (i.e., if a specified perception occurs, respond immediately with a particular action), it is clear that a transactional theory of literacy requires more deliberative agents that are capable of remembering, representing, and manipulating perceptions in more sophisticated ways. The agents required by a transactional theory of literacy must not simply be social (i.e., capable of immediate unreflective response), they must also be sociological (dInverno & Luck, 2001, section 5.3.2); they must be capable of social modeling, and this requires attention to a number of new dimensions of agency. Foremost among the requirements for sociological agents is some form of internal storage or memory where percepts and models of agents and relationships can be retained. The StoreAgent schema responds to this need for a more persistent representation of the environment by incorporating a store variable that takes an Environment data value. With the introduction of additional schemas (see dInverno and Luck, section 5.2 for details) perceptual functions are specialized to distinguish percepts both internal (i.e., memories) and external (i.e., environmental states), and schemas are developed to represent store agent state, actions, and an update operation on the store. In effect, the store provides a persistent internal representation of an environment that can be applied by an agent in subsequent perception, planning, modeling, and action. The stage is now set for the modeling required by sociological agents.

Multi-Agent Transactional Theory - Page 16 of 35

StoreAgent Agent store : Environment Before a sociological agent can model an environment, however, data types must be defined for the models that will be created. Not surprisingly, model data types are defined in terms of the entities they represent. Since, as noted earlier, a store agents perceptual functions differentiate internal and external percepts, models are distinguished from the entities they represent on the basis of the percept source (i.e., the agent store in the case of models). EntityModel == Entity ObjectModel == Object AgentModel == Agent AutonomousAgentModel == AutononmousAgent NeutralObjectModel == NeutralObject ServerAgentModel == ServerAgent The AgentModelEntities schema begins by defining a store agent that incorporates both state and a set of entity models. Such an agent is capable of representing entities in its environment, but does not distinguish objects and agents, nor their respective capabilities, goals, and motivations. An example of such an agent is a robotic or software-based counter that assesses traffic on a city street or hits to a web page. In this case, the granularity of representation is very coarse and does not distinguish one entity from any other; the goal is simply to generate a numerical sequence that maps onto perceived entities in a one-to-one fashion. Subsequent schemas define more complex model-aware agents, progressively expanding the scope of representation to include objects (AgentModelObject) and agents (AgentModelAgents). Extensions to a web page hit counter, for example, might distinguish visits by browser type, recognizing that different browsers may display elements of the site in different ways that may exert subtle but important influences on users experiences. Each of these schemas defines agents that model their environments in ever greater detail culminating in the AgentModels schema that incorporates models for all of the entity types considered thus far, and even supports an optional variable (modelself) to model itself. AgentModelEntities StoreAgent AgentState modelentities : 1 EntityModel AgentModelObjects AgentModelEntities modelobjects : 1 ObjectModel modelobjects modelentities

Multi-Agent Transactional Theory - Page 17 of 35

AgentModelAgents AgentModelObjects modelagents : 1 AgentModel modelneutralobjects : 1 NeutralObjectModel modelagents modelobjects modelobjects = modelagents modelneutralobjects AgentModels AgentModelAgents modelautonomousagents : 1 AutonomousAgentModel modelserveragents : ServerAgentModel modelself : optional [AgentModel] modelautonomousagents modelagents modelserveragents = modelagents \ modelautonomousagents Social modeling, however, requires attention to relationships as well as entities. Engagements, engagement chains, and cooperations organize and structure entities in important ways that must be considered in modeling an environment. The SociologicalAgent schema therefore, incorporates models for relationships (see dInverno & Luck, section 5.3.2) as well as entities and this provides a basis for still more sophisticated relationship modeling that an agent can apply in assessing the likelihood of achieving a goal (e.g., will another agent participate in a cooperation to achieve a goal?) Formally, the SociologicalAgent schema follows the same general pattern of specialization adopted in earlier model schemas by incorporating an agent state, the most directly related schema predecessor (in this case, the AgentModels schema), and the specific additions characteristic of the new schema (i.e., relationship models). Schema constraints assure that the relationship models adhere to existing definitions. SociologicalAgent AgentState AgentModels modeldirectengagements : DirectEngagementModel modelengchains : EngagementChainModel modelcooperations : CooperationModel md : modeldirectengagements (md.modelserver modelserveragents) (md.modelclient modelsagents) mch : modelengchains (ran mch.modelagentchain modelserveragents) (mch.modelautonomousagent modelautonomousagents) mc : modelcooperations (mc.modelgeneratingagent modelautonomousagents) (mc.modelcooperating agents modelautonomousagents)

Multi-Agent Transactional Theory - Page 18 of 35

Interpretive Aside 2 Given the formal apparatus presented thus far, it may be useful to once again consider a more practical perspective. As was noted earlier, an important aspect of multi-agent transactional theory is the recognition that both reader and text may assume roles as autonomous, independently motivated agents, a circumstance that deviates from the role assignment traditionally adopted in print transactions, where text typically is transformed from a neutral object into a server agent through direct engagement by a reader. The schemas that have been developed to support environmental modeling provide agents with an opportunity to represent an environment in a more persistent fashion that affords agents a greater degree of knowledge and thus control. Nothing in the schema specifications, however, dictate whether a given agent will actually choose to create models of its environment, or whether models that might be created are accurate or complete representations of an environment, all of which bear on the efficacy with which a literacy agent pursues its goals. Consider the example used earlier focusing on web-based search engines that mix paid advertisements with search results in an effort to achieve goals created by the site, rather than simply serving in support of reader goals. A nave reader using such a site might conclude that s/he is the only autonomous agent operating in the environment, erroneously treating the relationship with the site as an engagement rather than a cooperation. In addition, given this commitment, such a reader would be likely to limit modeling of the environment to the server agent level (i.e., terminating modeling with the AgentModelAgents schema), since there is no perceived need to provide for other autonomous agents. In each case, however, the nave reader is operating under incomplete information and this puts this reader at a disadvantage in the effort to achieve goals in contrast with more sophisticated readers who more accurately assess and model environmental entities and relationships. A more benign, indeed, useful example of deliberately limited environmental modeling has already been noted in the example of the web page counter. Although autonomous agents (e.g., human readers) will visit such pages with specific goals in mind, those goals may have nothing whatever to do with the goal of the web page counter agent. In fact, from the perspective of the counter agent, there may be no advantage in representing autonomous agents in its environmental modeling, although there is at least one reason to consider not doing so, since there is additional overhead in maintaining a more complex environmental model. Ultimately, the point of environmental modeling is to enhance an agents effectiveness in achieving its goals. In some cases a cruder environmental model may be more than enough. In others, an incomplete environmental model may significantly reduce the effectiveness of an agent. The effectiveness of an agent in an environment, therefore, can be viewed as balancing the benefits of modeling against the costs of building and maintaining models. Transactional theory addresses variability in modeling using the concepts of selective attention and reader stance. Although these terms are used in an informal manner, their use by Rosenblatt (1994) consistently defines stance as a more general orienting function and selective attention as a function that brings specific literary elements out of the penumbra of the perceptual field (generated by stance) into the spotlight of the readers attention. In multi-agent transactional theory, the meaning of a text (i.e., the poem) emerges out of the engagements or cooperations of multiple agents. These

Multi-Agent Transactional Theory - Page 19 of 35

engagements and cooperations represent the concept of a transaction. Transactions, however, in a multi-agent framework consist of two distinctly different types that, incidentally, serve to distinguish traditional literacy as practiced in the print tradition and a more recent counterpart that only became possible in the dynamic and interactive context provided by an electronic reading environment. This marks an important point of elaboration introduced by multi-agent transactional theory, an elaboration that is consistent with, but not anticipated by, traditional transactional theory, that there is more than one type of transaction and a failure to recognize this distinction may undermine or even subvert the goals of a reader. Defining a Multi-Agent Literacy System With a general framework for modeling the activities of agents within a multi-agent system, it is possible to extend this framework to focus on concepts and relations central to a transactional theory of literacy. The present system will, however, adhere to a more restricted focus than adopted in some other recent work. Specifically, although recent literacy theory attempts to address a wide range of activities under the umbrella of semiotic theory, including video, images, oral language, and even human performance, the present framework sets a more modest agenda by restricting the focus of inquiry to written language. This restriction serves both theoretical and practical objectives as it provides for greater specificity, remains consistent with Rosenblatts (1994, 1995) primary emphasis on written language, and provides for the integration of existing discourse analytic techniques based on text propositionalization (e.g., Kintsch & van Dijk, 1978; Meyer, 1985; Graesser, 1985). There are, therefore, both practical and theoretical reasons to warrant this restriction, which in no way restricts future specializations that may treat other symbol systems. Like the multi-agent framework upon which it is based, the present system begins with a more basic element that serves as a foundation. Characters, defined as specialized sets of attributes, represent the most basic element of written language. In the case of English, characters are the letters, spaces, digits, and punctuation used in written English. Other languages, of course, incorporate other characters. However, since the definition makes no presuppositions concerning the nature of characters, it does not restrict the characters we consider to those of any specific language. Taking characters as a starting point, the String data type is defined as a non-empty sequence of characters. As before, since the definition does not adhere to any logical or spatial convention in ordering characters (e.g., top to bottom, left to right, right to left, etc.) the linguistic generality of the framework is preserved. A TextEntity schema is defined, inheriting the usual entity properties, but also mapping some entities onto text (thus, the partial function). This partial function associates text with some entities, leaving the entity itself unaltered. Following the hierarchical pattern of objects, agents and autonomous agents, the TextEntity schema serves as a foundation for other text schemas that define levels of increasing autonomy and potential to influence the literacy systems within which they operate. Character == Attribute String == seq1 Character

Multi-Agent Transactional Theory - Page 20 of 35

Text text : String TextEntity Entity Content : Entity Text dom Content = Entity TextObject Object TextEntity TextNeutralObject NeutralObject TextObject TextAgent Agent TextObject TextServerAgent ServerAgent TextObject TextAutonomousAgent AgentText AutonomousAgent It is also be useful to define a number of additional schemas that play a special role in defining text in electronic reading environments. As noted early in this paper, although this framework is intended to account for reading transactions in print, it also seeks to address limitations of transactional theory in electronic reading environments (i.e., in hypertext.) The following schemas, therefore, define special features of text in electronic environments, with the TextEntity schema providing the foundation for this development. Hypertext nodes are conceptualized as text entities, with directional links defined as an ordered pair of nodes. The content and links of a hypertext are defined, respectively, by the set of text entities that serve as its nodes and the set of links that connect those nodes. Hypertext is defined in terms of its nodes and links with the additional requirement that every link must begin in an existing node of the hypertext (i.e., dom Links TextEntity). Note, however, that there is no requirement that every link terminate in an existing node in the hypertext, since this would rule out the possibility of broken and dead links, all too common features of the web that have significant consequences for readers and transactions. Finally, the HTPath schema defines a variable htpaths representing all possible paths in a hypertext given a set of links. Note that paths are defined as a sequences of nodes. HTText Nodes : TextEntity

Multi-Agent Transactional Theory - Page 21 of 35

Link == TextEntity x TextEntity HTLinks Links : Link From, To : TextEntity t : Link | t Links (t.From = head t.To = last t) HyperText HTText HTLinks dom Links Nodes #Nodes > 1 Path == seq1 Node HTPath HyperText htpaths : Path p | p htpaths j, k : Nodes | j, k in p j, k Links The MultiAgentLiteracySystem schema follows from the schema defining a multi-agent system through the incorporation of a specialized set of attributes that takes the form of a character string. Moreover, since text is an implicit subset of the original attributes given set, the relationship and modeling framework defined earlier (e.g., cooperation, engagement chains, agent models, sociological agents, etc.) can be interpreted in a new way (i.e., as a literacy system) without any modification to the formal apparatus already defined. In effect, a multi-agent literacy system inherits the conceptual apparatus and all of the functionality of a multi-agent system while supporting a new interpretive perspective. Formally, the MultiAgentLiteracySystem schema is quite brief, defined as a type of multi-agent system that includes text entities among those that make up the system. MultiAgentLiteracySystem MultiAgentSystem text : TextEntity text entities Interpretive Aside 3 Although it may not be immediately apparent, this relatively minor tailoring of dInverno and Lucks framework for conceptualizing multi-agent systems captures a great deal of transactional theory as traditionally presented, as well as introducing a number of genuinely novel concepts that raise some productive questions and perspectives concerning literacy transactions. And perhaps not surprisingly, the most productive questions and perspectives deal with electronic reading environments where text entities can take on a level of autonomy not possible in traditional print.

Multi-Agent Transactional Theory - Page 22 of 35

Briefly, literacy transactions are engagements and cooperations involving one or more text entities in a multi-agent literacy system. Moreover, the hierarchy of text types that have been defined align with transactional theory as traditionally understood while extending the concept of text in an important new direction. Engagements and cooperations are, however, defined in terms of the entities they relate, with a cooperation involving at least two autonomous agents while direct engagements and engagement chains limit participants to a single autonomous entity. It is, therefore, useful to consider how the various text entities that have been defined participate in transactions. Recall that five distinct text entities were derived based on the generic TextEntity schema, ranging from the TextObject schema to the TextAutonomousAgent schema. Three of these schemas have special significance in a transactional framework: the TextNeutralObject, TextServerAgent, and TextAutnomousAgent schemas. From a transactional perspective, the TextNeutralObject schema defines an object that provides the opportunity for a transaction as traditionally understood (e.g., paper and ink.) From this perspective, the book-as-object is, as traditionally understood, only an object that has a potential for meaning, until a reader engages the text in creating an evocation that leads to meaning, which is embodied in the transaction rather than the entities that participate in it. Text does, however, have a special place in the present framework, as written artifacts have potential for literacy transactions that other objects in the environment do not. Multi-agent literacy systems therefore distinguish neutral object text as one type of entity with status deserving special attention within the larger environments where transactions occur. And, having defined objects of special attention, it should also be clear that not everything in a literacy environment can or should be an object of study. Some objects may have nothing whatever to do with text and should be ignored, although it is important to add that even objects that have nothing whatever to do with written language may shape a literacy system (through the larger environment or situational influences) that warrant their inclusion in a theoretical analysis. Summing up, neutral object text is a textual artifact with a potential to engage in a literacy transaction but without the goals and motivations that are the hallmark attributes of agents. On the next rung of text hierarchy are entities defined by the TextServerAgent schema which I will informally refer to as server text. This entity occupies a special central position in traditional transactional theory because this is the level that defines the traditional role ascribed to printed material when engaged in a literacy transaction. What distinguishes server text from its neutral object predecessor is the fact that it has been engaged (either directly or through an engagement chain) by an autonomous agent within the system and has, as a result, acquired one or more goals and these goals are the foundation upon which an evocation is created. Consider, the case of a newspaper laying by a front door in the early morning hours. As a neutral object the newspaper has no intrinsic meaning. It is simply one more physical object in an environment and may (e.g., if the house is unoccupied), never be engaged in a literacy transaction. It may be however, that someone is interested in reading this newspaper and in this case a reader establishes a direct engagement with the newspaper that transforms it into a server agent that supports the goal of the agent who initiated the transaction (e.g., to find out what the weather will be). More prosaically, a text entity acquires a meaning (e.g., sunny, no chance of rain) by virtue of its participation in a

Multi-Agent Transactional Theory - Page 23 of 35

transaction initiated by an agent. More formally, an engagement occurs that results in a change in that literacy system that includes a modification to the agents model of its environment that may have direct bearing on his/her actions. The meaning of this text does not reside within the text entity nor does the initiating agent carry it within. Recognizing that the example I have cited falls in a fairly unambiguous fashion on the efferent end of the aesthetic-efferent continuum, a question naturally arises whether literary transactions can be described in a similar fashion. From a theoretical position, however, the only difference between afferent and aesthetic stances toward text are the goals of the reader-agent. That is, while the newspaper reader wants to know what the weather will be, the literary reader has a desire to engage in an aesthetic experience. Notwithstanding transactional theorys traditional focus on literary goals, I would argue that there is no theoretical basis (or need) for supposing that literary transactions are governed by different processes than those that govern efferent transactions. Ultimately, while significant differences are to be expected in how readers choose and respond to text, there is no reason to suppose that whether a goal is aesthetic or efferent should bear on the theory we apply in analyzing the situation. Indeed, the very fact that this is a continuum would soon create analytical problems related to drawing lines between these poles where none can really be said to exist. Moreover, there simply is not the need since the goals an agent brings to a transaction have a profound influence on the outcomes of that transaction (see AgentPerception schema). At the highest level of the text entity hierarchy is the TextAutonomousAgent schema, introducing a genuinely new concept to transactional theory, that of autonomous text, an entity that may overlap in significant ways with those agents we have traditionally referred to as readers. Specifically, autonomous text has been defined as a specialization of the autonomous agent schema, and this means the traditional distinction drawn between reader and text is dissolved. In a mundane sense this may be as obvious as the name on the back of a jersey or the case of the now-rarely-seen sandwich board man pacing city sidewalks. There are, however, a wide range of more subtle examples of this phenomenon that take on special significance in online reading environments where every attribute and every set of attributes in a digital environment can ultimately be reduced to the form of a text entity. Although human readers engaged in electronic literacy transactions are inclined to perceive their environments as limited to one side of the interface, the transaction crosses the interface and can be deeply rooted on both sides (e.g., when two autonomous agents establish a cooperative literacy transaction.) The hypertext schemas that lead up to the MultiAgentLiteracySystem schema also serve to theoretically ground techniques that have been developed to support analysis of hypertext structure (e.g., Botafogo, Rivlin, & Schneiderman, 1992; Melara, 1996) and user navigation (e.g., Barab, Bowdish, Young, & Owen, 1996; Canter, Rivers, & Storrs, 1985), establishing a clear connection with prior work in human computer interaction. Moreover, when coupled with recent work in user navigation adopting a specifically transactional perspective (McEneaney, 2002), it can be shown that discourse analysis techniques based on text propositionalization (e.g., Kintsch & van Dijk, 1978; Meyer, 1985) can be integrated within the proposed theoretical framework.

Multi-Agent Transactional Theory - Page 24 of 35

Modeling motivations and models of other agents Modeling entities and relationships provides another significant step in understanding how agents operate in environments but still misses two elements that can be quite important in understanding environments that include other autonomous agents. Motivations, the force that actually drive actions in an environment have not yet been modeled, nor have the models that agents create. It is clear, however, particularly in light of the example cited earlier of the nave reader and the manipulative search engine, that the models generated by agents in an environment play an important role in shaping the transactions that occur in that environment. Accordingly, it is useful to model the motivations of agents and the models those agents construct and apply in their transactions. Given the fact motivations are drawn from a given set, without any specification concerning their internal structure, modeling of motivations is limited to assigning relative strengths to motivations in order to resolve conflicts between competing goals and to assess the effect of satisfying goals on agents so appropriate actions can be taken. Briefly (readers are referred to dInverno &m Luck, 2001, section 3.3.2 for details), modeling the motivations of agents is based on a satisfy function that assesses the relative strength of a motivation or, from another perspective, the relative strength of a motivational effect that is expected to result from achieving a goal on an agent in an environment. Specifically, the expression satisfy (g) indicates the motivational effect/strength on agent a of satisfying the goal g, according to agent bs model of the environment. In less abstract terms, if Anne (a) encounters an unfamiliar word when reading, she has the option to either stop and look up the word (s) or simply read on (r) and hope she can figure things out later. Annes teacher Bob represents the strengths of these options for Anne as satisfy (s) and satisfy (r), respectively. Moreover, if evaluation of the satisfy function based on Bobs environmental model shows that satisfy (s) > satisfy (r), Bob will conclude that Anne will stop to look up the word rather than simply read on. It may turn out, however, that Annes evaluation of the satisfy functions according to her own environmental model differs from Bobs. It may be that Anne actually prefers to read on rather than look the word up (i.e., satisfy (r) > satisfy (s)) and in this circumstance the difference between Anne and her teachers environmental models lead Bob to incorrectly predict that Anne will stop to look up unfamiliar words. The satisfy function therefore provides a basis for agents to use what they know about their environment to make predictions, although as illustrated in the example, the predictions thus generated are only as good as the environmental model on which they are based. It is also be useful to consider situations where an agent not only models other agents in its environment but also is prepared to model the models that are generated by other agents. The first requirement for such an agent is that it must recognize sociological agents in its environment (i.e., those agents that have models) and the second is that the agent must be capable of representing models of those sociological agents it encounters. These two requirements are addressed respectively in the ModelSociological and ModelModels schemas.

Multi-Agent Transactional Theory - Page 25 of 35

ModelSociological SociologicalAgent modelsociologicalagents : Agent SociologicalAgent dom modelsociologicalagents modelagents ModelModels ModelSociological modelsociological : Agent ModelSociological dom modelsociological modelagents Interpretive Aside 4 The framework that has been developed supports and extends transactional theory in a number of ways that are important to highlight prior to the analyses and applications that are taken up in the next section. Significantly, nearly every important departure of MATT from traditional transactional theory can be traced back to a pair of related modifications: the introduction of autonomous text entities and denial of a priori assignment of readers and texts to autonomous and non-autonomous roles, respectively. While MATT continues to support traditional role assignments, it does not do so in an a priori fashion. Rather, it defines roles in theoretically principled ways based on the actions entities do or do not take in any given transaction. In effect, what is assumed in traditional theory is explained in the proposed framework, warranting the claim that the proposed system is both more general than, and subsumes, transactional theory as traditionally understood. Moreover, although the proposed modifications to transactional theory appear modest, the consequences for literacy theory are considerable. Transactions occur when an agent engages or cooperates with another entity in its environment to achieve a literacy goal. Traditional transactional theory requires a human reader to serve as the agent initiating the transaction by appropriating a neutral text entity in its environment (e.g., a printed text) in a manner that transforms this neutral object into a non-autonomous agent (i.e., a text server agent) that operates in support of the readers goals. MATT, while still requiring an autonomous agent to initiate a transaction, does not limit transactions to those involving autonomous and server agent pairs, and, as a result, three distinctly different kinds of literacy transactions are possible: those that involve a single autonomous agent, those that involve multiple autonomous agents, and those that are refused outright, with traditional theory restricted to transactions of the first type exclusively (see Figure 3). This division of transactions into types has several rather significant practical outcomes that deserve mention including:

Figure 3. Three types of transactions in the MATT framework including 1) a traditional asymmetric transaction involving one autonomous and one server agent, 2) an interactive symmetric transaction involving two autonomous agents, and 3) a refused transaction that fails.

Multi-Agent Transactional Theory - Page 26 of 35

1) agents other than human readers may initiate transactions, 2) transactions involving multiple agents may fail outright (by being refused), and 3) transactions involving multiple agents may result in multiple inconsistent views. Although traditional theory restricts the initiation of literacy transactions to human readers, this restriction does not apply in MATT if other autonomous agents (e.g., software systems) are operating in an electronic literacy environment. In this case, non-human agents may initiate transactions that can range from malicious viral action (e.g., delete local files) to genuinely helpful actions that assist users in locating or managing information (e.g., notify users of available software updates). The fact that transactions may be initiated by non-human agents, however, radically alters the traditional theoretical ontology since it implies an equivalence of reader and text that directly contradicts the neutral-object status of text, a position that explicitly denies text the status of agency except when engaged by an autonomous agent in a transaction. Given the traditional view of text entities as neutral objects in the absence of an ongoing transaction (Rosenblatt , 1994, p. ix) and the obligatory character of engagements with neutral objects (dInverno & Luck, 2001, section 3.4.1), it is not surprising that traditional theory has little to say about transactions that fail outright. While transactions may fail in the sense that a readers objectives are not achieved, it isnt clear what it might mean to speak about the outright failure of a transaction with print since this medium cannot resist engagement in a transaction except when another autonomous agent acts in opposition, and this would not usually be considered a theoretically relevant circumstance. I may, for instance, want to read a text that is currently being read by another person who declines my request to relinquish it, but this outright failure of the transaction would not usually be considered something that a theory of literacy should explain. Outright transactional failure cannot, however, be reasonably excluded from the realm of literacy theory when multiple autonomous agents are involved since establishing a cooperation is an essential aspect of this type of transaction. Consider for example the case of a reader who attempts to initiate a transaction with an autonomous text entity that requires a reader login. In this case, an outright failure of the transaction (e.g., a reader cannot provide required login information) can and will be accounted for by theory since the restriction to specific reader agents is an essential feature of this text. Finally, in a circumstance where multiple autonomous agents participate in a transaction each of those agents may generate models of the literacy environment, including models representing one another and the relationships that they have to one another. There is, however, no guarantee that the models developed by agents will be either complete or accurate, with the consequence that different agents involved in a transaction may generate different and potentially incompatible models of the environment and transaction. Moreover, any global model of a transaction (i.e., interpretation) must be prepared to accommodate as many distinct transactional perspectives as there are autonomous agents involved in the transaction. For example, consider the case of a human reader who has initiated a transaction with an online search engine that surreptitiously mixes paid advertisements with genuine search engine hits. This transaction can be viewed as involving multiple autonomous agents who may generate distinctly different transactional models. The deceptive character of the search engine reveals that it models the reader as another autonomous agent, since otherwise there would be no need for deception. A nave human reader, on the other hand, may not

Multi-Agent Transactional Theory - Page 27 of 35

model the search engine as another autonomous agent with an agenda of its own, but treat it as a server agent, taking the information provided at face value, a circumstance that will generally disadvantage the human reader in achieving the objectives that originally motivated the transaction. Three comparative analyses This section will present three short informal analyses of literacy situations that highlight specific elements of MATT and contrast it with traditional transactional theory. Elements highlighted in these analyses include stance, role assignments, text autonomy, and agent modeling. The last analysis will also raise an important fundamental question regarding the nature of the aesthetic stance that bears on the traditional distinction drawn between aesthetic and efferent stances and an implication of the framework proposed for this distinction. Aesthetic reading in traditional print (asymmetric engagement) The first case considers an aesthetic transaction in traditional print. In transactional theory, although readers initiate transactions, not much is said about why readers choose to read. It seems plausible, however, to presume that a readers specific purpose or objective in reading is captured in a general way by the concept of stance. According to transactional theory, readers who adopt an efferent stance are typically looking for information, while those adopting a more aesthetic stance have a more process-oriented emphasis on the reading experience. Since this example presumes a reader with a primarily aesthetic stance, we will assume an intrinsic motivation to read (i.e., it is a pleasurable activity). As so often noted by Rosenblatt, however, stance is a complex concept that resists the dichotomies that so often drive theoretical discourse. Consider, then, the case of a mystery novel serving as an example of a popular work which, while not necessarily literary art, nevertheless shares an emphasis on the aesthetic stance (Rosenblatt, 1994, p. 143). A mystery novel selected to provide a good beach read highlights the remarkably fluid and generally dynamic character of stance during the reading transaction. The reader of this text generally does not require or want supporting resources (e.g., dictionaries, commentaries, etc.) Rather, there is generally a presumption that the good beach read will be self-contained, providing for a satisfying and not-too-demanding lived-through experience with little need for adjunct aids. At the same time, however, one distinguishing feature of the genre is that it reliance on factual detail and logical (and, usually, not immediately apparent) explanatory structure, both of which are more commonly associated with more efferent forms of reading. The difference in the case of the mystery novel is that the exploration and synthesis of fact may have no referent in the real world, it simply serves to create a special kind of lived-through experience. The suspension of disbelief and literary license that support aesthetic experiences, however, are not without proper epistemological grounding, otherwise we would not experience the work as either meaningful or satisfying. Rather, the literary work relies on more general patterns of coherence (e.g., Are characters believable?, Is the plot plausible?, Do circumstances seem real or contrived?) If the literary work adheres to a coherent internal structure it will generally meet the requirements of readers by providing a good representation of reality as the reader understands it. Not surprisingly, however, in a transactional account the

Multi-Agent Transactional Theory - Page 28 of 35

phrase as the reader understands it covers a great deal of conceptual territory. Some readers may, as a consequence, experience an otherwise unremarkable detective story in ways the elevate it to a genuinely literary experience, while another might consider Crime and Punishment as simply a long-winded variant of the same genre. That, of course, is what distinguishes transactional theory from other approaches. The perspective MATT provides for the example cited generally aligns with the informal analysis described above, with a few notable differences in the absence of a priori commitments to roles and the level of detail supported in the analysis. One difference that emerges very early in the analysis is the fact that MATT need not begin by assigning specific roles to reader and text prior to the start of the transaction since these roles are actually defined by the transaction itself. In the case of human readers, it might seem theoretically rigid and unproductive to withhold the attribution of autonomy to human readers until the transaction itself supports this interpretation, but there are compelling reasons (addressed in subsequent examples) not to privilege human readers in this way. Moreover, a human reader need not wait long to acquire the status of an autonomous agent since the very act of initiating a literacy transaction provides evidence o0f the autonomy of an agent. In a sense, the attribution of autonomy to a human reader when a transaction is initiated is simply the counterpart to the transformation of a text from paper and ink into a transactional agent. Notably, however, this provides a principled basis for role assignment which, as is apparent in subsequent examples, is quite important in online literacies. With regard to the level of detail supported, there are several differences between a traditional analysis and the MATT framework worthy of note. One difference is that, whereas traditional theory tends to rely on a loosely defined family of intentionally underspecified concepts, the MATT framework is quite specific about how readers respond to text. Foremost among these responses are the capabilities of agents to represent their environment, themselves, and other objects and agents they encounter, including literary works (i.e., the evocation or model created to represent the text). In the present example, MATT specifies that reader stance is an expression of reader goals and that selective attention represents a subsequent narrowing. Formally, this is expressed in the willperceive function, defining a preliminary view based on reader goals (i.e., stance) and a more narrowly focused view based on selective attention (i.e., Goal View View; see AgentPerception schema). It is this second view that represents the text as the reader encounters it and forms the basis for the model that becomes the readers evocation of the text. Moreover, the evocation model is subject to continuing refinement and extension, representing secondary responses that shape interpretation. A second difference is that while traditional theory skirts fundamental ontological issues raised by transactions, the MATT framework addresses them directly. What is it that happens to a text when a reader appropriates it in a transaction? Clearly, there is a dramatic transformation. Something that formerly was paper and ink takes on a capacity to interact with readers, but traditional theory says nothing specific about the nature of this transformation. In MATT, on the other hand, this transformation is explicitly defined within an ontological framework that includes neutral objects, server agents, autonomous agents, and other more general terms that define the ontological status of theoretically significant entities on the basis of clearly defined attributes that relate to observable behaviors. By this account, books have no special status in an environment,

Multi-Agent Transactional Theory - Page 29 of 35

they are just paper and ink, until they are transformed by an autonomous agent into server agents applied in support of a readers goals. Moreover, when goals are achieved or abandoned, or the reader terminates the relationship with the server agent, the text returns to its former status as a neutral object. Agency as applied in MATT expresses an empirically grounded ontology based on functional roles and, thus, holds true to the phenomenological roots of transactional theory. Although the formalism it is based on is unfamiliar, particularly when compared to the natural leaps of insight and intuition that mark Rosenblatts articulation of transactional theory, MATT, by virtue of its precision and detail, offers explanatory and predictive powers that go beyond that of traditional theory. Aesthetic reading online (online asymmetric engagement) Suppose now, sticking to the same genre (i.e., the mystery novel), the setting is altered so that our reader is reading a hypertext mystery rather than traditional print. In a hypertext, blocks of text referred to as nodes or pages are connected to one another with links. For the purpose of the present example, let us also assume that the hypertext selected is static, meaning that all links are permanent and do not depend on choices that readers make or on other dynamic variables. Another way of putting this is to exclude scripts and other programming that might support dynamic features. The text is not, therefore, altered by its use and does not model its readers or environment, operating as a nonautonomous server agent. What kinds of analyses do traditional theory and MATT provide for this situation? Not surprisingly, the account provided by traditional theory in the case of hypertext is indistinguishable from that provided for print, with the obvious reason that traditional theory does not acknowledge forms of text other than traditional print. Moreover, while it would be inappropriate to call this a flaw, since hypertext goes beyond what traditional theory sets out to address, this limitation does not merely constrain our theoretical view, but leads to a genuine theoretical crisis that Rosenblatt foreshadows in her major theoretical work (1994). Specifically, in responding to the problem of idiosyncratic or pathological interpretation, Rosenblatt focuses on the relationship of the work-as-evoked to that of the text itself, noting that a reader may not do justice to it [i.e., the work] by failing to point to the elements of the text that have given rise to the interrelated or interwoven whole or structure. (1994, p. 91). Significantly, she recognizes both the static character of structure in traditional print and the dynamic character of structure as perceived by readers, arguing that transactional theory requires a dynamic, reader- as well as text-oriented understanding of rhetorical or critical terms such as form or structure (1994, p, 91). The problem with hypertext is that, not only must we acknowledge the role of readers in defining the structure of the work-as-evocation, we must also be prepared to acknowledge a role for readers in defining the text-as-object, something traditional theory has not done. In effect, hypertext links introduce a virtual structure that defines a wide range of structural possibilities for the text-as-object . This problem does not arise in the trivial case of linear hypertext that enforces a single specific path through a narrative, or in a narrative that takes the form of a single text node. In the case where readers have genuine choices to make, however, the number of structural possibilities is an exponential function of the frequency with which alternative links are provided. As a result, different readers may be able to generate logically incompatible but fully justified interpretations

Multi-Agent Transactional Theory - Page 30 of 35

that are based on completely different plot lines, undermining the notion of chronological, causal, and other relationships that are traditionally understood to organize events into a coherent narrative (Rosenblatt, 1994, p. 91). Although traditional print sometimes employs techniques (e.g., flashbacks, multiple perspectives, etc.) that suggest these problems and some forms of literature intentionally capitalize on these kinds of ambiguities (see, for example, Rosenblatts analysis of the anti-novel, 1994, p. 95) there is a persistent underlying presumption that, whatever constellations of meaning readers may evoke, the text-as-object at least provides one fixed point in our literary heavens. The framework developed, on the other hand, addresses virtual structures directly, acknowledging that hypertext provides readers not a single narrative path, but a set of narrative paths, as defined by the htpaths variable in the HTPath schema. The path an individual reader takes during a transaction (referred to as an episodic path, see McEneaney, 2002) is one that has been selected from htpaths, a variable that serves to define a virtual navigational space, bounding all possible paths. Even more significantly, an extension of the MATT framework beyond the scope of this paper but described in detail by McEneaney (2001), provides analytic techniques for resolving problems related to plot, coherence, and interpretation on the basis of agreed upon criteria for what constitutes sound methods of inquiry and judgment (Rosenblatt, 1994, p. 183). Briefly, methods are defined that allow the paths of individual readers to be summed, generating classes of aggregated paths that represent structural abstractions based both on the observed paths of individual readers and on methodological criteria that are subject to empirical assessment. The openness of hypertext, therefore, while certainly greater than traditional print is not unconstrained, with empirically grounded techniques providing a basis for systematic validation. Aesthetic reading online (symmetric interactive cooperation) The previous two examples illustrate some important distinctions between transactional theory as traditionally understood and the agent-based modeling framework developed in this paper. In each case, however, the role assignments that characterize traditional theory were adopted. In the first case, an autonomous human reader (i.e., an autonomous agent) engaged a printed text operating as a server agent. In the second example, a human agent once again engaged a static text, but in this case the server agent was an electronic text, a hypertext consisting of nodes and links. While the engagement of a hypertext altered certain elements of the transaction, particularly those related to defining text structure and the text-as-object, this case did not alter role assignments, with the second example presenting a transaction again based on an engagement of a server agent by an autonomous reader. The purpose of this third example, is to explore one of the more radical consequences of the multi-agent framework developed in this paper: transactions founded on cooperations between multiple autonomous entities. Transactions founded on cooperations operate according to fundamentally different principles than do those based on engagements since they involve autonomous agents who may choose to decline a transaction. As noted earlier, autonomous text introduces what has been called outright transactional failures in which an agents attempt to initiate a transaction by establishing a cooperation with an autonomous text entity (or a server text acting on ones behalf) is denied. Password protected access to an online chat area

Multi-Agent Transactional Theory - Page 31 of 35

or bulletin board, for example, could be viewed as an event that illustrates this kind of outright failure. Furthermore, as was noted in considering the example of the search engine mixing paid advertisements with hits, transactions based on cooperations potentially increase the risks resulting from inaccurate or incomplete agent models since two potentially incompatible sets of goals are involved. The purpose of this third example is to consider how these issues might be expressed in the context of a literary work. As before, the mystery is the genre under consideration. Unlike the second example, however, where an electronic reading environment was an option selected to highlight differences, in the present case, an electronic environment is required in order to support an autonomous text entity. The autonomous text entity under consideration is, like the previous example, a hypertext mystery. Unlike the previous example however, as an autonomous text entity, it has goals and motivations of its own and may model its environment and its relationships in ways to help it better achieve those goals. Generally, such a hypertext will not be malicious, since most authors and publishers want readers to have satisfying experiences. A commitment to providing a satisfying experience, however, may not entail full support of the readers objectives at all times, especially in the case of this specific genre. Although most mystery readers do not expect or seek out literary works that require great effort, most are disappointed if there is no element of uncertainty or surprise, or if there is no opportunity for critical reading that anticipates and explains. What fundamentally distinguishes this example from the second, however, is that the text entity involved is autonomous and this means it has motivations and is capable of generating new goals (e.g., be less direct, avoid conversations related to the murder, etc.) depending on its motivations and the models it creates. In this mystery the text plays an active role, both supporting the reader and pursuing its own goals based on the internal agenda that motivates it. At the same time that the reader is constructing a model of the text, the text is creating a model of the reader and the models created by both reader and text shape the transaction as a whole. In the beginning, a readers model will typically be sparse, but as the transaction develops it fills out. The texts model of the reader will depend on a number of design considerations. If the story is presented by an omniscient narrator, it may be that the text already has access to the story line as a whole and the model of the reader that is generated may be framed within this larger model. In this case, the model building of the text will likely focus specifically on the reader, with the objective of responding to the reader in ways designed to enhance her experience. Models that the text builds of the readers model of the text may also be important here since understanding what a reader does and does not know about the story line is likely to have important implications for what should be revealed and how the reader is likely to respond. Alternative forms of presentation can also be adopted representing a perspective of a single character in the story, or multiple representations of the story line based on views of different characters. Although these approaches to literature have been implemented in print the capacity of text to dynamically model both itself and its reader is something genuinely new, as is a texts capacity to reorganize and revise its agenda. Traditional transactional theory is simply unprepared for this kind of text. Paper and ink do not lead to this kind of transaction.

Multi-Agent Transactional Theory - Page 32 of 35

Implications and conclusions The examples presented illustrate a number of ways the MATT framework supports and extends traditional transactional theory, particularly as applied in electronic reading environments. As has been noted, the MATT framework distinguishes different types of transactions that are crucial in understanding the new, genuinely interactive, literacies that are appearing online. It also provides a theoretically principled accounting for a priori role assignments in traditional transactional theory, ensures a basis for warranting transactional interpretations on the basis of transaction-specific modeling and on the basis of aggregated user paths (McEneaney, 2001), and can be integrated with existing discourse analytic techniques that rely on text propositionalization. Finally, in addition to these literacy-specific benefits, the framework proposed supports a meta-theoretical framework capable of integrating an agent-based theory of interactive system design with a transactional theory of literacy that more completely and accurately addresses central issues in theory, practice and pedagogy of literacy in electronic environments. As a result of defining this larger theoretical framework, the design and implementation of interactive systems can be more adequately grounded in the phenomenological and social contexts that define literacy for human readers. Engineers and programmers who design interactive literacy environment may no longer be compelled to rely on an ad hoc conceptualizations of reading and writing in making crucial decisions about system design. Likewise, literacy researchers and educators are likely to find that a better understanding of the logic and structure of electronic reading environments will allow for deeper understanding of readers experiences, as well as enhanced capacity to design pedagogical strategies to assist readers in effectively using those environments. From a pedagogical perspective, the proposed framework defines a new type of online literacy transaction that has profound implications for the way literacy educators think about and design instruction, particularly as it relates to developing skills in critical literacy. While traditional pedagogical practices related to critical literacy will continue to be useful, traditional practices cannot account for the dynamic environment that marks multi-agent transactions. But it is precisely these transactions that most demand the critical skills of readers since it is in this kind of transaction that readers are most at risk. In multi-agent transactions reader and text are obliged to negotiate a transaction on the basis of multiple agendas and environmental models. The introduction of multi-agent transactions to online literacy environments, therefore, creates a significantly more complex literacy environment. Of course, the power of multi-agent literacy environments can be applied to the direct benefit of readers. Autonomous text can be explicitly designed to adapt itself to the needs and interests of users to create levels of support that have never before been possible (Anderson-Inman & Horney, 1998; Leu & Hillinger, 1994; Leu, 2000). But the same technologies that can be used to create support for readers can also been applied in manipulative zero-sum agendas that seek to achieve objectives in direct conflict with reader agendas. How should we prepare students for this kind of dynamic literacy environment? The blunt and not-so-reassuring truth is we dont really know, in part because our theoretical frameworks in literacy cannot accommodate the dynamic nature of these new environments. This work is offered as a preliminary step toward addressing this need.

Multi-Agent Transactional Theory - Page 33 of 35

References Andersen, P. B. (1997). A theory of computer semiotics. Cambridge: Cambridge University Press. Anderson-Inman, L. & Horney, M. A. (1998). Transforming text for at-risk readers. In David Reinking, Michael C. McKenna, Linda D. Labbo, & Ronald D. Keiffer (Eds.), Handbook of Literacy and Technology. Mahwah, NJ: Lawrence Erlbaum. Barab, S. A., Bowdish, B. E., Young, M. F., & Owen, S. V. (1996). Understanding kiosk navigation: Using log files to capture hypermedia searches. Instructional Science, 24, 377-395. Botafogo, R. A., Rivlin, E., & Shneiderman, B. (1992). Structural analysis of hypertexts: Identifying hierarchies and useful metrics. ACM Transactions on Information Systems, 10(2), 142-180. Bowen, J. P., Dunne, S., Galloway, A., & King, S. (2000). ZB 2000: Formal specification and development in Z and B, First International Conference of Z and B Users, Lecture Notes in Computer Science, 1878. Berlin: Springer-Verlag. Bradshaw, J. M. (1997). Software agents. Cambridge, MA: MIT Press. Canter, D., Rivers, R., & Storrs, G. (1985). Characterizing user navigation through complex data structures. Behaviour and Information Technology, 4(2), 93-102. Castelfranchi, C. (1990). Social power. In Y. Demazeau and J.-P. Mller (Eds.), Decentralized AI Proceedings of the first European workshop on modeling autonomous agents in a multi-agent world, 49-62. Amsterdam: Elsevier. dInverno, M., Kinny, D., Luck, M. & Wooldridge, M. (1998). A formal specification of dMars. In Intelligent Agents IV: Proceedings of the Fourth International Workshop on Agent Theories, Architectures, and Languages. Lecture Notes in Artificial Intelligence, 1365(155-176). Berlin: Springer-Verlag. dInverno, M. & Luck, M. (2001). Understanding agent systems. Berlin: SpringerVerlag. Ferber, J. (2001). Multi-agent systems. New York: Addison-Wesley. Graesser, A. (1985). Conceptual graph structures. In Bruce K. Britton & John B. Black (Eds.), Understanding expository text. Hillsdale, NJ: Lawrence Erlbaum. Halasz, F. & Schwartz, M. (1994). The Dexter hypertext reference model. Communications of the ACM, 37(2), 30-39. Harm, M. W. & Seidenberg, M. S. (1999). Phonology, reading acquisition, and dyslexia: Insights from connectionist models. Psychological Review, 106(3), 491-528. Just, M. A, & Carpenter P. A. (1992). A capacity theory of comprehension: individual differences in working memory. Psychological Review, 99, 122-49. Kay, J. & Kummerfeld, B. (1997). User models for customized hypertext. In Charles Nicholas & James Mayfield (Eds.), Intelligent Hypertext. Berlin: Springer Verlag. Kintsch, W. & van Dijk, T. A. (1978). Toward a model of text comprehension and production. Psychological Review, 85, 363-394. Leu, D. J. (2000). Literacy and technology: Deictic consequences for literacy education in an information age. In Michael L. Kamil, Peter B. Mosenthal, P. David Pearson, & Rebecca Barr (Eds.), Handbook of Reading Research, Vol. 3. Mahwah, NJ: Lawrence Erlbaum. Leu, D. J. & Hillinger, M. (1994, December). Reading comprehension in hypermedia: Supporting changes to childrens conceptions of a scientific principle. Paper presented at the annual meeting of the National reading Conference, San Diego.

Multi-Agent Transactional Theory - Page 34 of 35

Marshall, C. C. (1998). Toward an ecology of hypertext annotation. In Proceedings of the ninth ACM conference on Hypertext and hypermedia, 40-49. New York: ACM Press. McClelland, J. L., Rumelhart, D. E. (1988). An interactive activation model of context effects in letter perception: Part 1. An account of basic findings. In: Allan M. Collins & Edward E. Smith (Eds.), Readings in cognitive science: A perspective from psychology and artificial intelligence, p. 580-596. San Mateo, CA: Morgan Kaufmann. McEneaney, J. E. (2001). Graphic and numerical methods to assess navigation in hypertext. International Journal of Human-Computer Studies, 55, 761-786. McEneaney, J. E. (2002). A Transactional Theory of Hypertext Structure. Paper presented at the National Reading Conference, Miami, FL. December 4, 2002. Available online at: htttp://www.oakland.edu/~mceneane/ Melara, G. E. (1996). Investigating learning styles on different hypertext environments: Hierarchical-like and network-like structures. Journal of Educational Computing Research, 14(4), 313-328. Meyer, B. J. F. (1985). Prose analysis: Purposes, procedures, and problems. In Bruce K. Britton & John B. Black (Eds.), Understanding expository text (pp. 11-64). Hillsdale, NJ: Lawrence Erlbaum. Reinking, D. (1998). Transforming texts. In David Reinking, Michael C. McKenna, Linda D. Labbo, & Ronald D. Keiffer (Eds.), Handbook of Literacy and Technology. Mahwah, NJ: Lawrence Erlbaum. Rosenblatt, L. M. (1995). Literature as exploration. New York: Modern Language Association. Rosenblatt, L. M. (1994). The transactional theory of reading and writing. In Robert B. Ruddell, Martha Rapp Ruddell, & Harry Singer (Eds.), Theoretical models and processes of reading. Newark, DE: IRA. Rosenblatt, L. M. (1978). The reader, the text, the poem. Carbondale, IL: Southern Illinois University Press. Rumelhart, D. E. (1994). Toward an interactive model of reading. In Robert D. Ruddell, Martha Rapp Ruddell, & Harry Singer (Eds.), Theoretical models and processes of reading. Newark, DE: International Reading Association. Seidenberg, M. (1992). Dyslexia in a computational model of word recognition in reading. In P. Gough, L. Ehri, & R. Treiman (Eds.), Reading Acquisition. Hillsdale, NJ: Lawrence Erlbaum. Singh, M. P., Rao, A. S., & Georgeff, M. P. Formal methods in DAI: Logic-based representation and reasoning. In Gerhard Weiss (Ed.), Multiagent Systems, 331376. Cambridge: MIT Press. Spivey, J. (1992). The Z Notation: A Reference Manual. Hemel Hempstead: Prentice Hall. Available online at http://spivey.oriel.ox.ac.uk/~mike/zrm/zrm.pdf. Stotts, P. D., Furuta, R., & Cabarrus, C. R. (1998). Hyperdocuments as automata: Verification of trace-based browsing properties by model checking. ACM Transactions on Information Systems, 16(1), 1-30. Weiss, (1999). Multiagent systems. Cambridge, MA: MIT Press.

Multi-Agent Transactional Theory - Page 35 of 35

Appendix Sorry this section has not been written yet. dinverno & Luck (2001) include a semi-technical overview that may be useful. Formal Specification in Z Formal specification Introduction to Z Notation Syntax

Vous aimerez peut-être aussi