Modeling Documents Based on User Performance: An Alternative to Author Intention and a priori Information Model Approaches

This paper proposes an approach to representing the structure of a document based on the way readers or users interact with it in the context of a deliberative task. This approach contrasts with other ways to model the structure of documents including approaches which map authorial intention and those which rely upon a well-known information model or genre. This presentation will highlight the benefits of understanding the structure of documents based on the rhetorical reading/using strategies of those who interact with them. These structures can be rendered as paths through a given set of information resources, offering insight into the way that objects and relationships that make up a document can mediate, or complicate, activity. The paper will conclude by showing some examples which make use of user-performance data to create task-appropriate views of complex (multiscale) documents.

There is a great deal of interest today in the idea of building models of texts. One reason is that, with the growth of the Web as a way to reach a wide and diverse audience, publishers of information of many types are now interested in building information structures that support multiple-audience adaptation. Another reason is to maximize the value of content by delivering information that is tailored to a particular task.

For example, imagine the day-to-day work of a claims processing agent for a large insurance company. The agent is responsible for making decisions based on information in documents of various types - claim forms, telephone call records, police and adjustor reports, medical records, even photographs - all stored in electronic policyholder files. There may be discernable patterns in these types of workflows which can be documented and used as the basis for information models that structure information at or below the level of an individual document. The model could allow information contained in all the documents associated with the policyholder file to transform to suit the decision-making needs of the specific users who interact with it.

Creating a system like the one described above would require analyzing and modeling fundamental patterns of document use, defining a basic modeling language for document-mediated interaction that can capture recurrent patterns of user performance. As Moser & Moore (1996) point out, most semantic modeling approaches construct formal text structures based on either author intention or, alternatively, on an information structure presumed to be instantiated in the text. Neither of these approaches is entirely appropriate for creating effective displays of information for potential users. Creating such displays requires a model of the text-mediated interaction between writers and readers which can then be used to define display conditions for a range of information "views" that a given document might support.

Performance-based text structure models differ from other types in that they are not primarily representations of a stable core semantic structure that is assumed to be either domain independent (e.g, Mann and Thompson, 1987), or genre specific, as suggested by the work of Bazerman(1988) and others. Nor are these structures maps of author intentionality and/or struggles in creating intentional relationships similar to analyses by Van Wijk and Sanders (1999). Rather, the models emphasize structures that constitute the resources authors and the specific users or readers of a document share in order to come to some kind of agreement about an issue or question that all parties have a stake in. What gains status as a unit or object in user-performance based models depends upon the deliberative activity that the document is meant to support. Relationships among objects are similarly defined by how the objects mediate a given decision. In this way, we can expect the model to account for both the regularities in text structures which correspond with similar texts doing similar mediational work, as well as quite specific and arbitrary text structures associated with any given deliberative activity as it unfolds in a social context. This modeling approach comes very close to a process described by Phelps (1985) who articulated an approach to structural analysis drawing on and responding to work by Faigley & Witte (1981) and Van de Kopple (1985) in composition studies, as well as Halliday & Hassan (1976) and Van Dijk (1976) in linguistics. The process, broadly, understands texts as objects with histories, requiring us to study them in process if we are to understand how they shape the experiences of a reader.

Bibliography

Moser, M.
Moore, J.D.
Toward a synthesis of two accounts of discourse structure
Computational Linguistics
22.3
409-419
1996

Mann, W.C.
Thompson, S.A.
Rhetorical structure theory: Toward a functional theory of text organization.
Text
8.3
243-281
1988

Bazerman, C.
Shaping written knowledge
University of Wisxonsin Press
Madison
1988

Van Wijk, C.
Sanders, T.
Identify writing strategies through text analysis
Written Communication
16.1
51-75
1999

Phelps, L.W.
Dialectics of coherence: Toward an integrative theory
College English
47.1
12-31
1985

Faigley, L.
Witte, L.S.
Coherence, cohesion, and writing quality
College Composition and Communication
32
189-204
1981

Vande Kopple, W.
Given and new information and some aspects of the structures, semantics, and pragmatics of written texts
Cooper, C.R.
Greenbaum, S.
Studying writing: Linguistic perspectives
Sage
Beverly Hills, CA
1986
72-111

Halliday, M.A.K.
Hassan, J.R.
Cohesion in English
Longman
London
1976

Van Dijk, T.
Text and context: Explorations in semantics and pragmatics of discourse
Longman
London
1976