Hyperdocument Presentation: Facing the Interface - CiteSeerX

Hyperdocument Presentation: Facing the Interface Jörg Hannemann*, Manfred Thüring**, Jörg M. Haake* * Integrated Publication and Information Systems Institute (IPSI) Gesellschaft für Mathematik und Datenverarbeitung (GMD) Dolivostraße 15, D-64 293 Darmstadt, F.R.G. {haake,hanneman}@darmstadt.gmd.de ** empirica GmbH, Communications and Technology Research Oxford Straße 2, D-53 111 Bonn 1, F.R.G. MANFRED%emp–[email protected]

published as: GMD Report #784. September 1993.

ABSTRACT To improve the readability of hyperdocuments, it is not sufficient to concentrate on navigation without supporting comprehension. Focusing on this widely neglected issue in hypertext research, interface design has to reduce the mental effort for comprehension which depends on the coherence and the mental workload required from the reader for concurrent activities such as navigation and orientation. Facilitating comprehension of hyperdocuments can be accomplished by coping with three issues: how to impose a coherent structure on the document, how to convey that structure to the reader, and how to support navigation and orientation. In this paper, we describe an approach which specifies these general issues for designing hypertext interfaces in more detail. Based on the resulting comprehensive issue structure, we describe SPI – a reader interface which demonstrates the close relationship between special design elements for structuring hyperdocuments and their presentation on the screen. It is this tight coupling that is of crucial importance for user interface design aiming to improve readability and to increase acceptance of hyperdocuments.

KEYWORDS hypertext readability, coherence in hyperdocuments, design issues and design rationale for reader interfaces, hyperdocument presentation, comprehension and navigation, SEPIA’s presentation interface (SPI)

1

INTRODUCTION

Because the interface directly affects usability and acceptance, its design is crucial for any hypertext system. In fact, many researchers seem to regard it as the most relevant feature of hypertext overall: according to Nielsen (1990), the classification of a system as a hypertext system should depend more on the “look and feel” of its user interface than on the commands and data structures that it provides. Brown demands that hypertext should be as “simple to use as a television set” (1986, p.176) and Shneiderman et al. (1991) predict a brighter future for hypertext applications if the user interface for browsing can be made more attractive and effective. Despite the obvious importance of interfaces in the domain of hypertext, no systematic and comprehensive view on the issues and problems involved in their design has been proposed so far. Instead, research has been dominated by a single issue: for the last few years, browsing has been regarded as the most 1

central user activity and interface design has focused mainly on the solution of navigation problems associated with this activity. This focus is perfectly appropriate for hyperbases that are used for retrieval and free exploration of information, but it is insufficient for hyperdocuments which aim to provide information in a coherent, reader-oriented way1. Examples for such documents come from a variety of domains, such as hypertext fiction, tutorials, training materials and guides. From a cognitive point of view, a hyperbase is appropriate for satisfying reader requirements that significantly differ from those for a coherent hyperdocument. Depending on their needs, readers may engage in three different kinds of activities (and their combinations): 1. Information scanning serves to decide whether a document contains any information of interest. 2. Information search serves to find specific information that is relevant with respect to a particular interest or task at hand. 3. Information reception serves to comprehend the complete document or parts of it selected from information scanning or information search. While the first two activities require interfaces for exploration and retrieval, the third activity demands interfaces which support readers not only in browsing but also in understanding and mentally representing information: “reading ... hypertext about matters that deeply matter to us” (Bernstein 1991b, p.365) requires more mental effort than exploration. For the purpose of information reception, navigation should therefore not be regarded as an end in itself, but as a mere precondition for the reader ’s actual goal: the comprehension of the hyperdocument. From this perspective, navigating through a hypertext net and reading the content of nodes both serve the same purpose: to find and understand information that fits the reader’s current interests. Problems of orientation and comprehension occurring in this process are not independent of each other. In order to avoid disorientation and insufficient understanding, the reader must build up a coherent mental representation of the hyperdocument. Under this premise, orientation and comprehension are important and interdependent parts of the same overall cognitive activity, i.e., the formation of a mental model which adequately represents the information encountered in moving through the document. Interface design which takes this position and addresses such neglected issues as reading, comprehension and remembering may help to avoid what Foss (1989) has coined the “Art Museum Phenomenon” of hypertext: “[a]fter you have spent a long day in a large art museum gazing at hundreds of paintings . . . at the end of the day you may not be able to tell someone what you have seen” (p. 408). In order to overcome the “Art Museum Phenomenon” together with related problems of insufficient understanding and disorientation, a broader perspective for designing hypertext interfaces is needed which does not focus on isolated problems, but instead addresses the overall readability of hyperdocuments. The development of such a perspective requires the accomplishment of two research tasks: 1.

To establish a theoretical framework on which to base the design of interfaces for the presentation of hyperdocuments. This framework should treat navigation and comprehension as related problems and help us to develop solutions for increasing hypertext readability.

2.

To articulate the design issues which have to be addressed in order to increase the readability of hyperdocuments. These issues should constitute an important part of a detailed design space (MacLean et al., 1991) for hypertext interfaces which is based on the theoretical framework and helps us to identify and implement design options in terms of concepts and tools.

In the present paper, we address both tasks. In section two, we introduce the concept of coherence from cognitive research on text comprehension and extend it towards a theoretical framework for hypertext readability. Based on this framework, we present a systematic theoretically motivated set of issues to 1.For the distinction between hyperbases and hyperdocuments see also Stotts and Furuta (1991).

2

design interfaces for the presentation of hyperdocuments. These issues 2 are extensions of the design questions described in “PHD” (Schuler & Thüring, 1992), a methodology for “Pragmatic Hypertext Design” which specifies design procedures as well as criteria and options for answering design issues 3. Moreover, we demonstrate how some of these answers may be embodied in a construction kit for coherent hyperdocuments as component of an authoring system. The construction kit is part of the cooperative hypermedia authoring system SEPIA (Streitz, Hannemann & Thüring, 1989; Streitz, Haake, Hannemann, Lemke, Schuler, Schütt & Thüring, 1992) which supports authors in coherent hyperdocument design (Thüring, Haake & Hannemann, 1991). In section three, we describe SPI (“SEPIA’s Presentation Interface”), an interface for presenting hyperdocuments which are created by using SEPIA’s construction kit thus providing a first set of possible answers to the issues identified in our design space. We finish with a summary and some conclusions for interface design and future research in section four.

2

ISSUES FOR THE DESIGN OF READER INTERFACES FOR COHERENT HYPERDOCUMENTS

One major purpose – or even the major purpose – of reading is comprehension. In cognitive science, comprehension is often characterized as the construction of a mental model that represents the objects and semantic relations described in a text (Jonson-Laird, 1983; van Dijk & Kintsch, 1983). The readability of a document depends on the mental effort spent on the construction process. This dependency is expressed by the first two issues in figure 1: if we want to increase the readability of a hyperdocument (I1), the main issue is how to support readers in the construction of their mental models (I2). Empirical studies have shown that a reader’s ability to understand and remember a text depends on its degree of coherence. Therefore, psycholinguistic research emphasizes the relation between coherence and information processing (see VanDijk & Kintsch, 1983). A document is coherent if a reader can easily construct a mental model from it that corresponds to facts and relations in a possible world: “The coherence of discourse depends in part on how easy it is to construct a single mental model.” (JohnsonLaird, 1989, p. 472). Two types of processes are especially important with respect to constructing a mental model. In order to understand the relation between clauses and sentences, readers infer ‘small scale’ connections which link pieces of information together and thus establish local coherence. Furthermore, readers infer ‘large scale’ connections which are conclusions drawn from several clauses, sentences, paragraphs, or even chapters. Such conclusions establish the global coherence of a text (VanDijk & Kintsch, 1983). They summarize the meaning of diverse chunks of information as an abstract ‘macroproposition’ and thus represent the common topic of the chunks. In the course of comprehension, a hierarchical macrostructure is built up which is an important part of the reader ’s mental representation since it comprises the main ideas of the document. Empirical studies of linear texts indicate that establishing coherence at a local as well as at a global level is facilitated when a document is set out in a well-defined structure and provides rhetorical cues reflecting its structural properties (see van Dijk & Kintsch, 1983). Note that for reducing the mental effort of comprehension, it is not sufficient simply to impose a coherent structure on a document; it is also necessary to convey that structure to the reader by appropriate cues (Charney, 1987). Authors can establish local coherence by indicating semantic relationships between statements within the text of a node. For instance, they can express a causal relationship between two clauses in terms of conjunctions such as “because” or “so.” Authors can establish global coherence by indicating which chunks of information belong together. For example, authors can 2 The issue structure is an artefact created by the SEPIA system and reflects the authoring process of the present

paper: first, the issues were developed in SEPIA’s planning space. Then a variety of positions answering the issues were generated and their pros and cons were represented in SEPIA’s argumentation space. Based on these structures, texts were written by elaborating positions and arguments thus constituting a complex hyperdocument. The present paper is a linearized version of selected parts of this hyperdocument. 3

A first version of PHD was developed in the ESPRIT project HYTEA. It is currently further developed in the ESPRIT project HIFI (1992).

3

aggregate sentences into a paragraph, helping readers to perceive them as related pieces of information that should be distinguished from preceding or following information.

Figure 1: Part of the issue structure in SEPIA’s planning space In a hyperdocument authors must provide cues for coherence at two levels, namely at the node level (i.e., inside a node) and the net level (i.e., outside of nodes). At the node level, authors can rely on conventional cues for linear text. At the net level, authors need to provide cues which parallel the cues for coherence at the node level. In order to establish local coherence, they may provide cues which represent semantic relationships between nodes and indicate what the contents of the nodes have to do with each other. In this respect, a link between two nodes can be regarded as fulfilling a function analogous to a conjunction in a linear text and hence its label should indicate the appropriate semantic relation. If authors want readers to construct relationships exceeding the level of local coherence, they need to incorporate cues for global coherence. In analogy to such cues as “paragraph” or “chapter” in linear text, they need higher order units that will enable readers to identify structures at net level and to represent them as macropropositions. In contrast to traditional text, neither the internal structure of these higher order units nor relations between them need to be linear. Units which fulfill these requirements are composite nodes as proposed by Halasz (1988). Besides processes involving coherence, a second factor affecting the ease with which readers are able to construct a mental model of a text concerns the limited capacity of human information processing. Every additional effort required in the course of processing a document reduces the mental resources available for comprehension. Therefore, all activities which go beyond reading create an extra workload. With respect to hypertext, such activities mainly concern navigation. The more energy the reader must spend on navigation and orientation, the less resources are left for comprehension. In analogy to Conklin’s notion of cognitive overhead (1987, p. 38) which basically concerns authors, one might regard the additional workload resulting from the reader’s interaction with the hyperdocument interface as manual overhead. This does not mean that navigation and orientation are always severe obstacles for comprehension, but that they can hamper comprehension if they interrupt the reader’s “train of thought” (Gordon et al., 1989; Monk et al., 1988). Hence, navigation in a hyperdocument should make as few demands as possible in order to avoid any impediment in the construction of mental models. 4

In summary, the issue of facilitating the construction of a mental model (I2) can be refined into three interrelated sub-issues which are discussed in the next sections: ♦ How to create a coherent hyperdocument (I3) ♦ How to convey coherence in hyperdocument presentation (I4) ♦ How to facilitate navigation (I5).

2.1

How to create a coherent hyperdocument

In order to create a coherent hyperdocument, an author must solve two problems: the creation of a coherent document structure (I6) and the appropriate naming of the created structural elements (I7). Both problems need to be solved keeping in mind that the structure and names will later be presented to the reader and therefore require a comprehensible format. Hence creation and presentation of a coherent hyperdocument are strongly interrelated (although we have distinguished them for analytical reasons). In the following two sections, we discuss issues 6 and 7 assuming that the structural characteristics and names created by the author can be preserved best by using a graphical presentation format which we will describe later (see section 2.2).

2.1.1 How to establish coherence by structuring According to the solution of two main problems identified in writing linear text (Bereiter & Scardamalia, 1987), two different activities may be involved in creating the structure of a hyperdocument: factual structuring and rhetorical structuring. Factual structuring identifies the information units which form the building blocks of the hyperdocument as well as the semantic relations between these units. As an example, consider two hypertext nodes: “Expert systems” and “MYCIN”. While the first one gives an overview of the major characteristics of expert systems, the second one contains a short description of MYCIN (a medical expert system, Buchanan & Shortliffe, 1984). Several semantic relations between both texts may exist (e.g., the characteristics of expert systems may be “illustrated by” the MYCIN description or features of MYCIN may be “summarized in” the text about expert systems). Information units and relations of that kind establish a structure that is basically focussed on the semantics of the document, but does not specify in which way its elements should be processed by a reader. Such a specification requires rhetorical structuring: depending on the author’s anticipation of his readers’ interests and background, different reading structures are required in which selected parts of the factual structure are presented in a linear or non-linear order. While linear rhetorical structures may be sequential, branching or conditional paths (Zellweger, 1989), non-linear ones take the form of partial nets derived from the factual structure and do not specify a particular reading-sequence (i.e. the reader can browse freely). The distinction of factual and rhetorical structures leads to the issue of how these structures can be created (I13 and I14). For structuring a hyperdocument, several models have been advanced in the recent past (e.g., the Trellis Model, Stotts & Furuta, 1989; the Hypertext Design Model, Garzotto, Paolini & Schwabe, 1991; the MacWeb Hypertext Model, Nanard & Nanard, 1991; and the Hypermedia Templates, Smith, Catlin & Garrett, 1991). Although all these models propose sophisticated concepts for structuring, none of them explicitly deals with the problem of coherence and distinguishes between factual and rhetorical structuring. In order to support authors in coping with this problem, we have built a construction kit that provides a set of design objects for both structuring activities particularly dedicated to establishing coherent hyperdocument structures (Thüring, Haake & Hannemann, 1991). Since it is a component of the authoring environment SEPIA (Streitz et al., 1992), it is called SEPIA’s construction kit (SCK). It consists of two parts: (1) A content part offers design objects for creating factual structures. Its information units are atomic nodes and composite nodes. Atomic nodes represent chunks of information (e.g., text, graphics, audio, 5

video). Composite nodes comprise references to other information units and are used for aggregating atomic nodes or other composites. In addition, two types of relations between information units are supported: “Node-node-links” connect two entire nodes (atomic or composite). “Embedded links” connect a selected part of a node with another node. Together, these design objects enable authors to create different kinds of factual structure (i.e., hierarchies as well as flat and layered nets, the latter resulting from nesting composites). (2) An organizational part offers design objects for rhetorical structuring. Its units are called structure nodes and structure links. Structure nodes are composite nodes containing references to design objects of the content-based structure. In addition, each structure node specifies an atomic node as a starting point that is automatically opened when a reader enters the structure node. Two kinds of structure nodes are provided by the kit, each allowing different degrees of freedom with respect to navigation: sequencing nodes represent author-defined reading sequences; exploration nodes represent a subset of the factual structure without specifying an additional ordering. Instances of both node types can be connected by structure links: sequencing links define sequences among content or sequencing nodes. Exploration links provide access to an exploration node and are usually embedded into a sequencing node. These design objects can be summarized in a hierarchy of associated design object classes in which every subclass inherits properties from its superclass (see figure 2). hypertext object node content node atomic node composite node

link structure node

content link

sequencing node-node node exploration link embedded node link

structure link sequencing link

explore link

Figure 2: Hierarchy of design object classes Providing design objects for both factual as well as rhetorical structuring offers an important advantage. The factual structure can be created by authors without accounting for reader-specific needs in detail. In fact, authors can use this part of the construction kit for the mere purpose of expressing their views of the domain that they are writing about. They can completely concentrate on domain specific aspects and postpone all rhetorical issues. Of course, in many cases a structure will emerge from this process which later will also be presented to the readers. On the other hand, different groups of readers often require different pieces of information as well as different orderings. In this case, authors can create readerbased, rhetorical structures by using the design objects from the organizational part of the construction kit and apply them to the results of factual structuring. Thus, multiple reader-specific versions can be derived from a single factual structure. In summary, the components of SEPIA’s construction kit for factual and rhetorical structuring provide support for establishing local as well as global coherence at the level of the hypertext net. Elements for factual structuring enable the author to represent basic information units as atomic nodes and their semantic relationships as links in order to establish local coherence. These structures can be aggregated into higher order units in terms of composites and can be linked together to ensure global coherence. In addition, elements for rhetorical structuring can be used to map factual structures onto reader-specific orders. In analogy to the author of a traditional document who increases global coherence by creating reasonable sequences of paragraphs and hierarchies of chapters, the author of a hyperdocument can accomplish the same by creating hypertext specific structures, such as paths and flat or layered nets. 6

2.1.2 How to establish coherence by naming Structure is a necessary but insufficient precondition for coherence. It informs the reader that a hyperdocument consists of separate information units, some of which are related to each other, but it tells nothing about the meaning of these units and relations. Information of that kind can only be conveyed to a reader by appropriately naming the structural elements, thus leading to the issues of how to label nodes (I15) and links (I16). In general, a node label should represent the content of the node (i.e., it should trigger expectations about the kind of information to be found when the node is opened). A link label, on the other hand, should express a semantic relation between link source and link destination thus helping the reader to understand why there is a connection between both nodes. So far, no conventions have evolved in the hypertext community which could serve as a guide for appropriate naming. In order to provide prima facie support for authors in choosing names that will increase the coherence of hyperdocuments at a local level, SEPIA’s construction kit provides a taxonomy of generic link labels. The taxonomy consists of verbs which describe possible functional relationships between the source and the destination of a link. Authors are advised to represent this relationship in terms of a short sentence by selecting a verb as link label from the taxonomy and by choosing two nouns which characterize the content of the connected nodes (e.g. “Expert systems” [source] “illustrated_by” [link] “MYCIN” [destination]). Since the taxonomy supports labelling, it reduces the “cognitive overhead” of hypertext authoring (Conklin, 1987, p. 38) and guides authors to create net structures that can be easily understood and – if appropriately presented – ensure local coherence.

2.2

How to convey coherence in hyperdocument presentation

Although the creation of structure is important for coherence it is by no means sufficient. The author must also present the structure in an appropriate form (I4). In conveying the structural properties to read ers, an author cannot solely rely on conventional cues such as indicating semantic relations between pieces of information (see also Charney, 1987). Instead, an author must deal with new forms of structural indicators that aim at helping readers to construct a mental model representing the content of the document. Three sub-issues are related to this task: ♦ How to present the structure of a hyperdocument (I8) ♦ How to present the content of a hyperdocument (I9) ♦ How to indicate correspondences between presented information units (I10).

2.2.1 How to present the structure of a hyperdocument An answer to this issue is to present hypertext structures in maps or graphical browsers. Although this solution is apparently not favored by all designers of hypertext systems, empirical studies have shown that content maps improve understanding (Monk, Walsh & Dix, 1988; De Lucas & Larkin, 1991). Maps and browsers provide an overview of central topics and important relationships, thus increasing global coherence. They support the development of a mental macrostructure that represents the gist of what is read. Into this gist, all information acquired from the document can be integrated, thus gradually building up a coherent mental representation. Moreover, a gist that is derived from a graphical structure can serve as a cognitive map that eases orientation. On account of this facilitation, cognitive resources that would otherwise be bound to navigation are set free for comprehension. The graphical presentation of structure greatly depends on the kind of design objects that are used to create that structure. Since we are concerned with the presentation of hyperdocument structures that are developed in SEPIA, we will focus on the presentation of objects belonging to SEPIA’s construction kit. 7

Two sub-issues are important in this context: how to present nodes and links in a graphical browser (I18) and how to present a layered hypertext net arising from nesting composite nodes (I17). The presentation of a node-link structure in a graphical browser usually exploits the graph metaphor. Nodes are displayed as objects (e.g., boxes, circles) indicating their name and type (e.g., a box carrying a name and a type flag or using a type specific color or shape). In addition, indicators for the amount and type of content can be given using icons and different degrees of shades or border width. Links are usually displayed as edges or arrows (if directed). Link types can be shown as special icons or labels carried by the arrow. Displaying node names and link labels are a prerequisite for establishing coherence at a local level (see also section 2.1.2) and thus support comprehension. Several approaches have been developed to facilitate the presentation of hypertext nets: fisheye views (Furnas, 1986), zooming functionalities and multiple browsers. Fisheye views show the entire hyperdocument structure at varying levels of detail. While those parts of the hyperdocument that are close to the reader’s current position are displayed in their usual size, parts that are farther away are presented in a diminished format. Unfortunately, fisheye views alone cannot handle the presentation of a layered hypertext net because they do not distinguish between atomic nodes and composite nodes. Dynamic zooming enables the reader to expand (or “explode”) those parts of the document that are of current interest and to compress (or “implode”) all other parts. In contrast to fisheye views, a zooming functionality can cope with different layers: when the reader opens a composite node this node is exploded and its content is displayed. At the same time, the node he came from is imploded. Thus navigating between layers can be accomplished by a combination of “zooming in” the next composite and “zooming out” the previous one. This supports the reader in smoothly travelling across the layers of a hypertext net at a selfpaced speed. Multiple browsers can be used to present a layered hypertext net by associating one browser with each layer. Therefore, multiple open browsers can display several layers of the hyperdocument at the same time using different windows for each of them. The position and size of these windows can be either defined manually by the reader or automatically by the presentation environment. Both solutions may have advantages as well as drawbacks: 1. Multiple browsers with manually positioned windows may cause a lot of interaction overhead by enforcing the reader to find the adequate location and size for each node opened. Therefore, this solution may use valuable cognitive resources which otherwise could be employed for comprehension. On the other hand, for specific tasks (e.g., searching) and certain users (e.g., experts), individual and task-oriented organizations of the screen can be very useful. 2. System-defined positions and sizes of windows relieve the reader from the burden of positioning and resizing. In this case, the limited screen space must be partitioned among the different browser windows. Adopting a piling strategy with overlapping windows allows for displaying many layers of the document simultaneously, but can make the content of upper layers almost invisible. On the other hand, defining a fixed lay-out without overlapping windows leaves the upper layers visible, but can only deal with a small number of layers due to the spatial limitations of the screen. So far, no guidelines have been developed which help authors to choose between these options. We will discuss one possible choice in section 3.

2.2.2 How to present the content of a hyperdocument and how to preserve contextual information The issue of how to present content (I19) is related to the amount of information that may be stored in an atomic node. Usually the content of a node is displayed in a dedicated window on the screen. Since 8

atomic nodes can contain fairly large chunks of information, the window may not suffice to display the complete content at a time. In this case, facilities must be provided which enable the user to process the information step by step. For example, scrollable windows can be used to display long pieces of text or large graphics. Further details of content presentation are given in section 3.1.1. The segmentation of information into nodes and their display in separate windows mainly contribute to “the fragmentation characteristic of hypertext” (Marshall & Irish, 1989; p.22). Fragmentation may result in a lack of interpretive context (Landow, 1987) and thus lead to the impression that the hyperdocument is merely an aggregation of loosely linked pieces of information rather than a coherent entity. To prevent this impression, the content of a node should not be presented in isolation but in a context which enables the reader to connect it to information outside the activated node. In order to clarify how the context of a node can be preserved (I20), two different kinds of contextual information, which we call structural context and temporal context, can be distinguished. This distinction gives rise to issues 26 and 27. Structural context can be provided by a presentation style in which graphical information about the document structure is presented together with the content of an activated node (see also Nielsen, 1987). In this case, the reader can locate the node in the environment of neighboring nodes and links and can relate its content to this structure. As we have argued, this increases coherence and supports the comprehension of important semantic relations. Additionally, fragmentation can be reduced by a temporal context in which the actual node is displayed together with its predecessor. The display of temporal context conveys a sense of continuity across nodes which is very important for comprehension: in attempting to understand the content of a new node, readers try to extract its new information and relate this information to the content of other nodes which they have visited before. In psycholinguistics, this activity is called “given-new-strategy” (Clark & Haviland, 1974) and is regarded as a basic process of comprehension. With respect to hyperdocuments, it facilitates the construction of an integrated mental representation of information that may be distributed over different nodes. One way to support the given-new-strategy in hyperdocuments is by preserving the content of the predecessor of the current node. Since readers can see the ‘given’ information of the old node, they can detect semantic relations between both sources. As a result, they can join the content of both nodes in a coherent mental representation. Another way to support this strategy is to allow readers to return to any location by accessing an active reading history. Such a history enables readers to simply click on a node they have visited previously and then immediately provides the desired information. This type of computer-supported retrospection is especially important if readers want to inspect information which they have not completely understood before (Alessi, Anderson & Goetz, 1979): equipped with knowledge that they have acquired in the meantime, they can move back, reconsider what they have read and correct an old view.

2.2.3 How to show correspondences between content and structure The presentation of structural context implies that an activated node is displayed together with the structure in which it is embedded. The presentation of temporal context implies that an activated node is displayed together with its predecessor. In both cases, two windows are required: one showing the current node, the other showing the contextual information. In order to reduce the impression of fragmentation, readers must be able to recognize correspondences between these different windows. For example, it should be possible to read the content of the activated node in window A and, at a glance, to identify its position in a structure displayed in window B. Considerations of this kind lead to the issue of how to show correspondences between information units across windows (I10). Of course, correspondences can be indicated by identical names, but additional support is required to save the reader from scanning and reading the vast amount of node labels that might be displayed in a 9

graphical browser. For this purpose, referential devices to indicate correspondences are useful. Such devices can establish relations similar to deictic references in guided tours as proposed by Trigg (1988). Referential devices include arrows or asterisks (Marshall and Irish, 1989). Also, color can be imposed as a cue for correspondences between objects on the screen (i.e., the identity of nodes displayed in different windows can be indicated by the same color). The use of identical colors for identical objects helps readers to detect correspondences at first glance and increases the coherence of a document at a perceptual or visual level. Therefore, color can serve as a valuable supplement to linguistic cues in order to point out relations which are crucial for comprehension. In addition, spatial cues can be provided. If contextual information is always displayed in a window at a fixed position, the reader can use this cue to identify correspondences. Examples of using color and spatial information as cues for correspondences are given in section 3.1.3 and 3.2.4.

2.3

How to facilitate navigation

For the majority of readers, navigating through a hyperdocument is not only a novel and unusual activity, it is also a demanding task that requires a lot of concentration. Two subtasks must be accomplished as part of navigation: 1.

Readers must “move” through the document in order to get any information. For this purpose, they must get familiar with the interface of the document and use its functionality for navigating.

2.

Readers must be informed about their current position and about options for proceeding back or forth in the document.

Depending on the complexity of the interface and on the quality of orientation support, both tasks can be rather demanding. Since navigation can require a considerable mental effort, it may occupy a significant part of the reader’s information processing capacity. In order to reduce the capacity that is required, two design issues are of crucial importance: the reduction of interaction overhead (I11) and the improvement of orientation (I12).

2.3.1 How to reduce interaction overhead To reduce the interaction overhead, it is necessary to take a closer look at the kinds of navigational needs that may occur during reading. These needs can concern the direction as well as the distance of a move. With respect to direction, one can distinguish between forward and backward navigation. While forward navigation usually occurs when readers seek new information by moving to a node they have not yet opened, backward navigation occurs when they try to find old information by moving to a node they have already visited. Relevant issues in this context are how to facilitate forward navigation (I21) and backward navigation (I22). With respect to distance, one can distinguish between local and global navigation. In local navigation, the reader simply wants to follow a link (i.e., to move from the current node to a node that is directly linked to it). In global navigation, the reader wants to reach a node that is not directly linked to the current position. Two cases are relevant for this type of navigation: (a) the desired node has not been opened before, but is currently visible (e.g., as part of a graphical browser); and (b) the desired node has already been visited by the reader, but is not directly linked to the node which is currently activated. In this case, it may or may not be visible on the screen. The combination of these types of moves leads to four types of navigational situations, each requiring specific support to reduce navigation overhead (compare I28 to I31). With respect to interface design, a variety of navigational devices can be used for this purpose. For example, standard facilities, such as 10

‘next buttons’ and ‘back buttons’ allow for local forward and backward navigation; histories and bookmarks can be used for global backward navigation; and indices as well as graphical browsers can be used to support all four types of navigational moves. Wright and Likorish (1990, p. 93) report empirical results supporting the conclusion that “different navigation systems appear suitable in different circumstances.” A way in which these possibilities can be usefully combined is described in section 3.2.

2.3.2 How to improve orientation With respect to improving orientation, Bernstein (1991) has claimed that a severe or prolonged disorientation is uncomfortable, but that the complete absence of orientational challenges is dull. Hence, he argued for a “mild disorientation” which “can excite the readers, increasing their concentration, intensity, and engagement.” (p. 295). Although we wonder what a mild disorientation might be and how it could be accomplished, we agree that hyperdocuments should not have a soporific effect. Nevertheless, the goal of fascinating a reader can be reached by means other than disorientation. Factors related to the content of a document are much more suitable for winning a reader’s interests – as Berlyne (1960) has convincingly shown in his theory of epistemical curiosity. At least for documents that are not primarily post-modern aesthetic works, we argue for diminishing disorientation as much as possible and propose to support orientation by graphical interface components, such as maps and browsers. As we have discussed in section 2.2.1, maps and graphical browsers ease comprehension by increasing coherence at the net level. Moreover, they are likely to improve orientation since comprehension and orientation are not independent of each other. McKnight, Dillon and Richardson (1992; p.73) report several experimental studies which show that readers of printed text develop an incidental memory for the spatial location of information encountered during reading. From this perspective, a mental model that is gradually derived from a coherent document can also function as a “mental map” which to some extent represents the place of information within a document. According to McKnight et al. (1992), tables of contents, structures of paragraphs and chapters, as well as headings in printed text may serve as landmarks that facilitate orientation. In a similar way, graphical structures provided by maps and browsers of hyperdocuments are likely to enhance the mental models of these documents with valuable spatial cues. How well a mental model serves the purpose of a map may be characterized in terms of the reader’s ability to answer three questions: where am I, how did I get here, and where can I go next? Obviously, the difficulty of answering these questions may be influenced by certain features of the interface. To support orientation, interface design should address the following issues: ♦ how to indicate the reader’s current position with respect to the overall structure of the document (I23) ♦ how to show the way that led to the current position (I24) and ♦ how to present navigational options for the next move (I25). In a layered hypertext net, the indication of the reader’s current location should show it in the hierarchy of layers as well as within the layer itself. Depending on the degree of nesting of composites, the complexity of the document structure may make it very difficult to show the reader’s location on both of these dimensions. Due to the limitations of the screen, a large and complex layered net cannot be displayed as a whole in a graphical browser. Since it is often impossible to indicate the reader’s position with respect to the complete document, it is necessary to provide a browser that displays only a segment of the overall structure, but nevertheless provides enough information about the current layer and the current node. Fisheye views (Furnas, 1986) can be considered as an option which provides at least some contextual information about the area which closely surrounds the reader ’s position. A similar difficulty arises for showing the way that led to the reader’s current position: a simple history function (as e.g., in Hyperties (Shneiderman, 1987) and HyperCard (Goodman, 1987)) which leaves the 11

structural information about layers and neighboring nodes aside is obviously not sufficient in documents of great complexity. Little more is gained by showing a list of recently visited places (e.g., in HyperCard the ‘Recent’ option shows postage stamp-sized images of the 42 most recently visited cards). Navigational options for the reader’s next move can be presented in three ways: ♦ as hot words or buttons embedded in the content of the currently open node, ♦ as activatable nodes in a graphical browser, or ♦ as items of a list of potential destinations offered in a standard navigation panel. Each of these possibilities not only displays the reader’s navigational options – it also provides the means for actually making the next move. From this perspective, the reduction of interaction overhead (I11) and the improvement of orientation (I12) are very closely related: the way in which navigational options are presented is usually combined with a certain kind of interface interaction and leads to specific types of reading and navigation. An author who has the choice between hot words, browsers and navigation panels must carefully consider which one to use because each may trigger a different kind of behavior on behalf of the reader. For example, hot words can “seduce” readers to leave the current text and to follow embedded links, an activity that can lead them away from information which they may have actually been interested in. On the other hand, hot words can be used to provide useful background information which readers can draw on when they feel that their knowledge is not detailed enough to understand the content of the actual node. Similar considerations can be made for the other two navigation devices: graphical browsers, which allow for unrestricted navigation, may be useful for supporting free exploration of an information space, but may be inadequate to support sequential information processing, as required for instance in reading a tutorial. A standard navigation panel on the other hand, which only supports simple forward and backward moves, is more suitable to support sequential reading, but is only of limited use for getting an overview of the available information or for interactive searching. The difficulty of decisions of that kind points to a major problem of designing interfaces for hyperdocument presentation: the lack of any guidelines which support authors in creating and presenting hyperdocuments in a coherent format. The issues presented in this section may serve as a first step into that direction. They provide a comprehensive overview of the problems hypertext authors are bound to encounter and can be used as a medium for discussion. Such a discussion may reveal a variety of design options out of which guidelines may evolve. Moreover, it may lead to a variety of different interfaces resulting from the implementation of design options. In the next section, we present an interface of that kind and discuss its particularities with respect to the issues of our design space.

3

DESIGNING THE INTERFACE

In this section, we describe the “SEPIA Presentation Interface” (SPI), a reader interface for hyperdocuments constructed with SEPIA’s construction kit. As an example, we show how SPI presents a hyperdocument capturing the debate about the intelligence of computers that was triggered by John Searle’s article “Minds, Brains, and Programs” (1980). Since its first publication, Searle’s position has provoked a multitude of replies leading to a complex net of issues, sub-issues, contrary positions and opposing arguments. Most of the resulting argumentative net is represented in a hyperdocument entitled “John R. Searle contra AI”. The document is based on the design objects and naming conventions of SEPIA’s construction kit and covers the issue of how to create a coherent hyperdocument (I3) by structuring (I6) and naming (I7) together with their sub-issues (I13 – I16). Since these issues have already been addressed elsewhere (see Thüring, Haake & Hannemann, 1991), we will only discuss issue 4 (in section 3.1) and issue 5 (in section 3.2) together with their sub-issues in more detail. 12

Let us start with an overview of SPI’s main features. The overall interface for displaying a hyperdocument in SPI is presented in figure 4. It provides several navigation and help facilities: the buttons “Navigator” and “System Info” at the top, and the arrow shaped buttons at the right bottom. Presentation is achieved by four windows, two displaying parts of the document structure (I8), the other two displaying parts of its content (I9). The position of the windows is fixed and adheres to the following principle: the screen is divided into four areas, each dedicated to display a specific type of information. The screen presents structural information in “structure windows” in the left column and content information in “content windows” in the right column. The screen displays currently active nodes in the bottom row and their immediate predecessors in the top row. To ensure continuous reading, one content node is always open. For example, opening a structure node instantly activates its start node and opens it. A similar principle has been applied in KMS (Akscyn, 1988), where either one or two nodes can be presented at the same time. Following a link in the twonode-mode of KMS leads to the replacement of the other node by the destination of the link. This system feature also ensures continuous readings, but since KMS does not display the structure of the document it implements only the vertical dimension of SPI. The major benefit of SPI’s fixed lay-out is that no information can be hidden by other windows. Therefore, readers do not need to move or resize windows and can concentrate on other activities. This simplification should significantly reduce interaction overhead (I11) as indicated by several empirical studies which investigated the impact of different window lay-outs. They showed that tiled windows, as opposed to overlapping windows, are easier to use and lead to higher accuracy and speed in accomplishing certain tasks (Bly & Rosenberg, 1986; Instone, Teasley & Leventhal, 1993).

Figure 4: The SPI reader interface

13

3.1

User-interface components for conveying coherence

With respect to increasing the coherence of a hyperdocument, the partition of the screen into four areas yields three advantages: 1.

It establishes a close correspondence between the structure of the document and its presentation thus conveying the coherent structure directly to the reader (I4).

2.

It gives an overview of important part-of relationships (I8) which is essential for comprehension and navigation.

3.

It reduces the impression of fragmentation because it preserves the temporal and structural context of active nodes (I26 and I27) by displaying their predecessors in another window.

Altogether, SPI reaches these advantages by providing an interface which integrates the presentation of content and the presentation of structure.

3.1.1 The presentation of content The content of atomic nodes (i.e., text, graphics, etc.) is displayed in a scrollable content window (I19). In the screendump of figure 4, two content windows are located in the right half of the screen. They show the text of two atomic nodes and carry the names of these nodes as labels. The content window at the bottom presents the currently activated node (i.e., the latest node opened by the reader). The content window at the top presents the node that was activated before. This means that whenever a new node is opened, it is displayed in the bottom window and the former content of this window is moved to the top. Additionally, the fontsize of the node which is no longer ‘active’ is reduced thus indicating its diminished actuality.

3.1.2 The presentation of structure The structure of a hyperdocument that is developed on the basis of SEPIA’s construction kit may consist of sequences or nestings of composite nodes. These are presented as graphical browsers in structure windows on the left side of the screen. Since a composite node can contain a net of considerable size, structure windows provide a zooming functionality: by pressing one of the three buttons at the top panel (“tiny”, “small”, “large”) the internal scale of a structure window can be enlarged or reduced. Additionally, the reader can move the whole graph in any direction to render hidden parts of the structure visible. In each structure window, nodes are presented as boxes carrying the node label and links are presented as arrows carrying the link label (if any). If an atomic node contains embedded exploration links they are indicated by a short blue arrow at the lower right corner of the node rectangle. (e.g. ‘Introduction’ in figure 4). These design decisions provide answers to issue 18. The interface offers the opportunity to visualize hierarchically nested structures and thus allows for a partial presentation of a layered hypertext net (I17): in figure 4, the upper left window displays the context of an activated sequencing node (‘Replies’) and the lower left window displays the internal structure of this node. The relation between both windows is analogous to the windows showing content information: while the bottom window presents the structure in which the reader is actually located, the top window presents its predecessor which (for the left column) belongs to a higher hierarchical level. This means, the reader has reached the current position by opening the composite ‘Replies’ in the linear sequence. As a consequence, the content of this node is displayed in the bottom window and its predecessor (‘John R. Searle contra AI’) is shown in the window above. 14

As mentioned in section 2.1, SEPIA’s construction kit offers two kinds of structural composites: sequencing nodes and exploration nodes. Each of them represents a specific part structure and entails a different type of browsing behavior. For their presentation, we distinguish between two graphical modes: pathview and netview. In the pathview, sequencing nodes are presented. To the author, these nodes provide the opportunity to define different kinds of paths. Following Zellweger (1989), we distinguish between sequential, branching and conditional paths. While sequential and branching paths are static, conditional paths are dynamic and depend on the reader’s previous actions (i.e., at a specific point in the path, the reader cannot reach all next nodes any more, but only a subset which is automatically computed). A sequential path is shown in the upper window, a branching path in the lower window of figure 4. Figures 5 and 6 show a conditional path:

Figure 5: Conditional path (a) Starting from the node ‘First Replies’, the reader can choose between three alternatives for the next move: ‘Robot Reply’, ‘Systems Reply’ and ‘Brain Simulator Reply’. If the reader follows the link to ‘Robot Reply’ for instance, the link between ‘Searle’s Rejections’ on the one hand and ‘Searle: System Reply’ resp. ‘Searle: Brain Simulator Reply’ on the other hand are removed. The result is shown in figure 6. When the reader later reaches ‘Searle’s Rejection’ only the node ‘Searle: Robot Reply’ will be immediately accessible.

15

Figure 6: Conditional path (b) Sequencing nodes and their presentation in the pathview enables authors to guide a reader in the way which they consider is best. Authors will use these concepts whenever they believe that parts of the document must be read in a specific sequence to be comprehensible. Compared to linear and branching paths, the concept of conditional paths provides additional support for tailoring information to the specific needs of a reader. Since a conditional path dynamically adapts to the actual navigation, it can be employed to construct comprehensible paths which are determined by prior information. Together, these different kinds of sequencing nodes offer powerful means for selecting information from a factual structure and for ordering it according to reader-specific, rhetorical considerations. In the netview, exploration nodes are presented. In contrast to sequencing nodes, exploration nodes do not present paths, but a net of nodes and links (see figure 7). The nodes of an exploration node can be visited in any order. Links have no impact on navigation, but are exclusively used to indicate semantic relations specified in factual structuring. If the exploration net contains no composite nodes, the lower left window is automatically expanded and covers the complete left side of the screen (as in figure 7). The reason for this design decision is obvious from our example in figure 7: since the exploration node does not contain different layers that might have to be displayed, the space of the former upper left window can be used to display a larger proportion of the net.

16

Figure 7: Exploration node displayed in a netview

3.1.3 Indicating Correspondences The predefined locations in the horizontal vs. vertical dimension matrix support the reception of content and structure of the current location as well as the next upper level of the document. While color is mostly used to give documents a more lively or interesting appearance, its function in SPI is not exclusively aesthetic. Instead, color is employed as an additional cue for orientation and serves as an indicator for important correspondences between visual objects of the interface. In the graphical browsers of our example (see figure 4), four colors are used 4: ♦ Red indicates the reader’s actual atomic node (‘Replies’ in the bottom window). ♦ Pink indicates nodes which have been visited before, but are no longer activated (e.g., ‘Searles’s Thesis’ in the upper window). ♦ Orange indicates the reader’s actual structure node (‘Replies’ in the top window). ♦ White indicates all nodes which have not been opened yet (e.g., ‘Conclusion’ in the top window). This consistent variation of colors helps readers to see where they are (red resp. orange), where they have been (pink), and where they can go for new information (white). The identity of nodes which are displayed in different windows is indicated by the same color (and of course by identical names). In figure 4, for example, the actual structure node ‘Replies’ in the left top window and the label ‘Replies’ of the left bottom structure window are both orange, thus pointing out that the bottom window displays the content of the orange node that is contained in the top window. Other correspondences exist between graphical browsers and content windows. For example, the currently activated atomic node ‘Replies’ is represented by the red rectangle in the left bottom window and by the red label of the right bottom content window. 4 Unfortunately, we cannot present colored figures here, but we hope that black, white and different shades of grey

are sufficient to illustrate the interface.

17

The use of identical colors for identical objects helps readers to detect correspondences at first glance and increases the coherence of a document at a visual level. Therefore, color can be used as a valuable supplement to linguistic cues in order to point out relations which are crucial for comprehension and navigation.

3.2

User-interface components for supporting navigation and orientation

An interface which efficiently supports navigation (I5) should require as little effort as possible for its operation and should provide good facilities for orientation (I12). To reduce interaction overhead in operating the interface (I11), forward navigation as well as backward navigation should be supported by dedicated facilities. In SPI, these consist of graphical browsers, a specific tool called navigator and standard navigation facilities.

3.2.1 Navigation in graphical browsers Navigation in a graphical browser is accomplished by clicking on nodes and is constrained by the kind of structure node that is currently active: if the browser displays the content of a sequencing node, the reader can only follow predefined paths in terms of local forward navigation (I28). For example in the browser of figure 3, the reader could reach the node ‘Replies’ only after visiting its predecessor ‘Searle’s Thesis’. A different situation arises for exploration nodes (see figure 6), where the reader can move to any node at any time simply by clicking on it (global forward (I29) and backward navigation (I31)). An exploration node can only be reached by executing an exploration link that is embedded in a node on a path represented as a sequencing node. Following such a link, the reader arrives at a specific node which serves as a starting point for the exploration. This landmark node is closely bound to the anchor of the exploration link: when finished with an exploration, the reader is automatically taken back to the anchor of the exploration link and then can move further on the path. This feature of the interface reestablishes the context of the path after exploration and supports the reader in going back to a predefined reading sequence.

3.2.2 The Navigator The Navigator is designed to support global backward navigation (I31). It can be activated by clicking on its button in the top panel of the interface and then displays a list of all nodes previously visited. Global backward navigation is accomplished by simply clicking on a name in the list. When the Navigator is opened, it completely overlaps the graphical browser in the left top window of the interface (see figure 8) and displays three types of information which are very helpful for orientation and navigation: 1.

It shows the history of a reading session by chronologically listing each node that has been visited during a session, thus preserving the temporal context of the visit of each node (I26) and enabling the reader to reconstruct the way (I24). Atomic nodes which have been visited while browsing through an exploration node are marked as ‘explored’ (e.g., ‘Luminous Room’ in figure 7). Therefore, users can tell from this list wether nodes have been visited as part of an exploration node or as part of a path. In order to ensure continuity in reading the content of atomic nodes, structure nodes are not included in the list. Instead, their start node is listed – usually bearing the same name – which is automatically selected when opening a structure node. 18

2.

It shows the currently activated atomic node, which is simply denoted by the last name on the list, thus indicating the reader’s position within a layer (I23).

3.

It shows the number of hierarchical levels of the document and the reader’s current position (I23): (a) The number of levels is given by a scale at the top and the bottom of the navigator. In figure 8, this scale indicates that the presented document has four levels (starting at level 1). (b) The name of each node in the list is indented in order to indicate the level to which the node belongs. For example, ‘Introduction’ is at level 1 while ‘Robot Reply’ is at level 3 (see figure 8). Nodes visited within an exploration are indented to the next half level to show that the exploration belongs to the next higher level path. With respect to the current location, readers can therefore easily recognize their position in the hierarchy and can see how much deeper they can go.

Since the Navigator window completely overlaps the graphical browser in the left top window of the interface, the correspondence between top and bottom windows is temporarily broken. However, because the Navigator is showing the complete history of a reading session, the structural context of the current position (i.e., the name of the next higher level of the network) is already displayed. Therefore, it is possible to replace the graphical browser temporarily without degrading orientation and navigation too much.

Figure 8: Navigator In order to cope with the problems arising from issues 23 and 24 (i.e., how to indicate the reader ’s current position and how to show the way that led to it), we propose a combination of graphical browser (i.e., the pathview, see figure 4) and history function as shown by the Navigator (see figure 8): the browser displays the reader’s position in the current layer and also shows the relation of this layer to the one above. The history function shows the overall number of layers of the document, lists all nodes (composite as well as atomic) that have already been opened and indicates their position with respect to the level of layers. Since the reader can actively select nodes in the history for browsing, it is called a navigator. In combination, both facilities not only provide the required information about the reader ’s position and recent navigation, but also indicate all available options for the next move: the browser displays the cur19

rent node together with all departing and arriving links and thus supports forward and backward navigation within the present layer. The navigator, on the other hand, lists all nodes already visited and thus supports global backward navigation across layers. Further details about the navigator and its combination with graphical browsing facilities are given in section 3.2.4.

3.2.3 Standard navigation facilities The navigation devices discussed so far are complemented by another facility which allows for local forward and local backward navigation (I28 and I30): the button panel at the right bottom of the interface. The panel consists of two buttons, one for backward navigation on the left and another for forward navigation on the right. Each button contains the names of nodes which can be reached from the reader’s current location. For navigating, the reader chooses a name from the list and activates it by a mouseclick. Which nodes are listed in the buttons depends on the kind of structure node that is currently active. In a linear path, both buttons contain a single name. The forward button displays the name of the next node on the path while the back button shows the name of the predecessor of the actual node. In a branching or conditional path, both buttons contain more than one name. If the reader stands at a branching point, the forward button displays the names of all alternative next nodes (see figure 4). If the reader has already passed a branching point, the back button displays the name of the latest node and the name of the branching point. Therefore, the reader can directly return to the latest point of decision. This is extremely useful if a reader wants to leave the current branch and take another choice. When moving through an exploration node, readers usually prefer to use the graphical browser. Therefore, the button panel does not support forward or backward navigation in the net. Instead, it must be used to leave the exploration node: the back button of the panel contains the name of the atomic node from where the reader started the exploration. Figure 7 shows for example, that the exploration was started from the node ‘Introduction’. When the reader clicks on that name in the panel, the exploration node disappears and he is taken back to the starting point.

3.2.4 Orientation facilities Since support for navigation and support for orientation are closely related to each other, most system features described in the previous section also aim at minimizing disorientation: 1.

The reader’s current position (I23) is indicated in the graphical browsers and in the Navigator.

2.

The reader’s way to the current position (I24) is recorded in the Navigator and recent moves can be easily reconstructed from the information presented in the graphical browsers.

3.

The reader’s navigational options (I25) are presented in the button panel, the browsers and the Navigator.

Orientation is further supported by the regular navigation semantics of the interface: selecting a node (either by clicking in the map, by selecting one in the button panel, or by clicking on an embedded link anchor) always makes this node the actual one. Thus, the lower part of the screen shows the structure and content of the actual location. The former location is marked as visited and its content displayed at the upper side. In case of a change in the hierarchy (i.e., opening an exploration or path node), the former layer is displayed in the upper part, too. Opening an exploration node reserves the whole left half of the screen for the netview browser. All navigation facilities are updated accordingly.

4

SUMMARY AND CONCLUSIONS

In the present paper, we have addressed a problem which until recently has not been in the focus of hypertext research, i.e., the design of reader interfaces for the presentation of coherent hyperdocuments. 20

Such interfaces must cope with requirements that differ from those arising when a hyperbase is assessed for exploration, search or information retrieval. In order to structure the design tasks related to their development, we have proposed a hierarchy of issues that are crucial in designing interfaces for the presentation of coherent hyperdocuments. The main issue of our approach addresses the question of how to support comprehension in order to increase hypertext readability. Subissues resulting from the main issue address such well known problems as navigation and disorientation. In section two, we have specified these issues in more detail and derived a variety of specific topics which interface design in the domain of hypertext should deal with. Based on this “design space”, we have proposed an interface called “SEPIA’s Presentation Interface” (SPI). The description of SPI in section three has illustrated a variety of design decisions which aim at increasing the local and global coherence of hyperdocuments. SPI is an integral part of the hypermedia authoring system SEPIA (Streitz, Hannemann, & Thüring, 1989; Streitz et al., 1992) and maps elements of “SEPIA’s construction kit” (Thüring, Haake & Hannemann, 1991) onto dedicated interface components. The design decisions taken in SPI are based on the insight that the quality of hyperdocuments cannot be improved by concentrating only on issues of navigation and orientation. Recent research has pointed out that the ability to navigate through a document greatly depends on the reader’s understanding its content (McKnight, Dillon & Richardson, 1991). Taking this into account, the navigation problem has to be reconsidered (Bernstein, 1990a) under the perspective of easing the comprehension of hyperdocuments. Authors should be able to accomplish a deeper understanding of a document at their readers by imposing a coherent structure on it, conveying this structure to the reader and providing adequate navigation facilities. SPI exactly follows this idea: using SEPIA’s construction kit, an author can generate a factual hyperdocument structure which represents semantic relationships between nodes. In addition, he can establish a variety of rhetorical structures which offer reading paths for sequential reading or exploration nodes for free browsing. To convey the coherent hyperdocument to a reader, these structures can be mapped onto SPI which supplies interface components for presentation and navigation. SPI uses a static screen lay-out that provides structural information to support both (1) the detection of correspondences between content and structure of the hyperdocument, and (2) orientation. It uses a combination of graphical browsers presenting nested hypertext nets and content windows displaying the content of atomic nodes to preserve the structural and temporal context of a node. To facilitate navigation, SPI reduces interaction overhead by supporting multiple ways of moving through the document: clicking on nodes in browsers, following embedded links in content windows, selecting links in the button panel, and using the Navigator for global backward navigation. Orientation is improved by indicating the reader’s current position in the graphical browsers and the Navigator. Navigational options are consistently presented in the browsers, the button panel and the Navigator. Orientation is further supported by the regular navigation semantics of the interface which maintains the structural and temporal context of the current node. It is this tight coupling of the different user interface components with the coherent hyperdocument structure that is of crucial importance for user interface design in hypertext (Waterworth, 1990). So far, only few designers have tried to develop an integrated approach which accounts for this close relation of structure and presentation (e.g. Marshall & Irish, 1989). In this paper, we have mainly focussed on the task of reading a hyperdocument. In addition, exploration and retrieval as parts of information scanning and information search are important topics of designing interfaces for presenting hyperdocuments. Several approaches addressing these topics have been discussed by others. Full text search has been implemented in Intermedia (Yankelovich et al., 1988) and SuperBook (Egan et al., 1989a). Indexes have been realized in a number of systems, for example, SuperBook (Egan et al., 1989b, and may range from manually constructed indexes (Glushko et al., 1988) to automatic latent semantic indexing (Dumais et al., 1988) and the combination of index and belief networks (Frisse & Cousins, 1989). Query facilities and information retrieval techniques (Salton, 1989) could be used to facilitate finding information in large hyperdocuments (Nielsen, 1990). Since information search can benefit from structural information and spatial contexts, SPI provides some support in that respect, too. Nevertheless, it will be useful to introduce additional mechanisms which support 21

search and retrieval processes more explicitly. Therefore, points on our agenda concern the integration of indexes, search/query functionalities. Another topic of future research is the evaluation of our design decisions. So far, we have not investigated the adequacy of our interface in a systematic study, but first experiences with several readers support at least some of our design decisions. Especially, the graphical browsing facilities and the navigator seem to give valuable assistance. Of course, evidence from these first observations is not sufficient for evaluating SPI. Therefore, a major part of our future work will consist of empirical studies testing the interface that was described in this paper. These studies will investigate which of the interface components support the construction of a coherent mental representation and thereby improve comprehension and navigation. Empirical results of that kind will help to create hyperdocuments of increased readability, and we hope that readers who access such documents will find them more comprehensible and enjoyable than today’s.

5

ACKNOWLEDGEMENTS

This work is part of a larger effort in the WiBAS department at the Integrated Publication and Information Systems Institute (IPSI). We wish to thank Helge Schütt, Wolfgang Schuler and Norbert Streitz for helpful discussions. We would also like to thank Christian Schuckmann und Boris Bokowski for supporting the implementation of SPI and for their stimulating ideas about graphical browsers and navigation facilities. And finally, we owe special thanks to Chris Neuwirth for her detailed comments and suggestions.

6

REFERENCES

(Akscyn, McCracken & Yoder, 1988) R. Akscyn, D. McCracken & E. Yoder. KMS: A distributed hypermedia system for managing knowledge in organizations. Communications of the ACM, 31(7):820–835, July 1988. (Alessi, Anderson & Goetz,1979) S. M. Alessi, T.H. Anderson, & E.T. Goetz. An investigation of lookbacks during studying. Discourse Processes, 2,1979, pages 197–212. (Bereiter & Scardamalia, 1987) C. Bereiter & M. Scardamalia. The Psychology of Written Composition. Hillsdale, NJ: Lawrence Erlbaum, 1987. (Berlyne, 1960) D.E. Berlyne. Conflict, arousal, and curiosity. New York: McGraw-Hill, 1960. (Bernstein,1991a) M. Bernstein. The navigation problem reconsidered. In E. Berk, J. Devlin (Eds), Hypertext/Hypermedia Handbook, pages 287–287. New York: McGraw-Hill, 1991. (Bernstein, 1991b) M. Bernstein. Structure, Navigation, and Hypertext: The status of the navigation problem. In Proceedings of the 3nd ACM Conference on Hypertext (Hypertext ‘91), pages 363–367, San Antonio, Texas, December 15–18, 1991. (Bly & Rosenberg, 1986) S.A. Bly & J.K.Rosenberg. A comparison of tiled and overlapping windows, CHI ’86 Proceedings, pages 101–106. New York: ACM Press, 1986. (Brown, 1986) P.J. Brown. Viewing Documents on a Screen. In S. Lambert & R. Ropiequet (eds.) CDROM: the new papyrus, pages 175–184, Redmond, WA: Microsoft Press. (Buchanan & Shortliff, 1984) B.G. Buchanan & E,H. Shortliff. Rule-based experts systems: The MYCIN experiments of the Stanford Heuristic Programming Project. Reading, Mass.: Addison-Wesley. (Charney, 1987) Comprehending non-linear text: The role of discourse cues and reading strategies. Proceedings of the 1st ACM Conference on Hypertext, pages 109–119, Chapel Hill, North Carolina, November 13–15, 1987. New York: ACM Press, 1987. 22

(Clark & Haviland, 1974) H. H. Clark & S.E. Haviland. Psychological processes as linguistic explanation. In D. Cohen (Ed.), Explaining linguistic phenomena. Washington: Hemisphere, 1974. (Conklin, 1987) J. Conklin. Hypertext: An Introduction and Survey. Computer Magazine, September, 1987, pages 17–40. (Dumais et al., 1988). S. T. Dunais, G. W. Furnas, T. K. Landauer, S. Deerwester and R. Harshman. Using latent semantic indexing to improve access to textual information. Proc. ACM CHI’88, pages 281–285, Washington, DC, 15–19 May 1988. (Dee-Lucas & Larkin, 1991) D. Dee-Lucas & J.H. Larkin. Content map design and knowledge structures with hypertext and traditional text. Poster at the 3rd ACM Conference on Hypertext, San Antonio, Texas, December 15–18, 1991. (Egan et al., 1989a) D. E. Egan, J. R. Remde, T. K. Landauer, C. C. Lochbaum and L. M. Gomez. Acquiring information in books and SuperBooks. Proc. Annual Meeting American Educational Research Assoc., San Francisco, CA, 27–30 March 1989. (Egan et al., 1989b) D. E. Egan, J. R. Remde, L. M. Gomez, T. K. Landauer, J. Eberhardt and C. C. Lochbaum. Formative design-evaluation of ’SuperBook’. ACM Transactions on Information Systems, 7(1):30–57. (Frisse & Cousins, 1989) M. E. Frisse and S. B. Cousins. Information retrieval from hypertext: Update on the dynamic medical handbook project. In Proceedings of the 2nd ACM Conference on Hypertext (Hypertext ‘89), pages 199 – 212, Pittsburgh, PA, November 1989. (Foss, 1989) C.L. Foss. Tools for reading and browsing hypertext. Information Processing & Management, 25(4):407–418, 1989. (Furnas, 1986) G.W. Furnas. Generalized fisheye views. Proceedings of the ACM CHI ’86, pages 16–23. Boston, MA, 13–17 April, 1986. New York: ACM Press, 1986. (Garzotto, Paolini & Schwabe, 1991) F. Garzotto, P. Paolini, D. Schwabe. HDM – A model for the design of hypertext applications. In In Proceedings of the 3nd ACM Conference on Hypertext (Hypertext ‘91), pages 313–328, San Antonio, Texas, December 15–18, 1991. (Glushko et al., 1988) R. J. Glushko, M. D. Weaver, T. A. Coonan and J. E. Lincoln. Hypertext engineering: Practical methods for creating a compact disc encyclopedia. In Proceedings of ACM Conference on Document Processing Systems, pages 11–19, Santa Fe, NM, 5–9 December 1988. (Goodman, 1987) D. Goodman. The Complete HyperCard Handbook. New York: Bantam, 1987. (Gordon, Gustavel, Moore, & Hankey, 1988) S. Gordon, J. Gustavel, J. Moore, and J. Hankey. The effects of hypertext on reader knowledge representation. Proceedings of the Human Factors Society 32nd Annual Meeting, pages 296–300. (Halasz, 1988) F. G. Halasz. Reflections on Notecards: Seven issues for the next generation of hypermedia systems. Communications of the ACM, 31(7):836–852, July. (HIFI, 1992) HIFI: Hypertext Interface For Information: Multimedia and relational databases. Technical Annex of the ESPRIT Project 6532. (Instone, Teasley & Leventhal, 1993) K. Instone, B. M. Teasley, & L.M. Leventhal. Empirically-based re-design of a hypertext encyclopedia. In INTERCHI ’93 Conference Proceedings, pages 500–506. New York: ACM Press, 1993. 23

(Johnson-Laird, 1983) P.N. Johnson-Laird. Mental models. Cambridge: Cambridge University Press, 1983. (Johnson-Laird, 1989) P.N. Johnson-Laird. Mental Models. In M. I. Posner (ed.), Foundations of Cognitive Science, pages 469–499, Cambridge, MA: MIT Press, 1989. (Landow, 1987) G.P. Landow. Relationally encoded links and the rhetoric of hypertext. Proceedings of the 1st ACM Conference on Hypertext (Hypertext ’87), Chapel Hill, North Carolina, November 13–15, 1987. New York: ACM Press, 1987. (MacLean et al., 1991) A. MacLean, R. M. Young, V. M. E. Bellotti & T. P. Moran. Questions, Options, and Criteria: Elements of Design Space Analysis. Human-Computer Interaction, 6(3&4):201–250, 1991. (Marshall, Irish, 1989) C. C. Marshall & P. M. Irish. Guided Tours and On-Line Presentations: How Authors Make Existing Hypertext Intelligible for Readers. In Proceedings of the 2nd ACM Conference on Hypertext (Hypertext ‘89), pages 15–26, Pittsburgh, PA, November 1989. (McKnight, Dillon & Richardson, 1991) C. McKnight, A. Dillon & J. Richardson. Hypertext in Context. Cambridge University Press, 1991. (Monk, Walsh & Dix , 1988) A. Monk, P. Walsh & A. Dix. A comparison of hypertext, scrolling, and folding as mechanisms for program browsing. In D. Jones & R. Winder (eds.) People and Computers IV. Cambridge: Cambridge University Press, 1988. (Nanard & Nanard, 1991) J. Nanard & M. Nanard. Using structured types to incorporate knowledge in hypertext. In Proceedings of the 3nd ACM Conference on Hypertext (Hypertext ‘91), pages 329–344, San Antonio, Texas, December 15–18, 1991. (Nielsen, 1990) J. Nielsen. Hypertext and Hypermedia. San Diego: Academic Press, 1990. (Nielsen, 1987) J. Nielsen. Hypertext ’87 Trip Report. Hypertext for McIntosh Computers. (Salton, 1989) G. Salton. Automatic Text Processing: The Transformation, Analysis, and Retrieval of Information by Computer. Addison-Wesley, 1989. (Schuler, Thüring, 1993) W. Schuler & M. Thüring. Pragmatical Hypertext Design (PHD). HYTEA Project Technical Report. ESPRIT Project 5252: Hypermedia Authoring Tools (HYTEA). Darmstadt: GMD-IPSI. (Shneiderman, 1987) B. Shneiderman. User interface design and evaluation for an electronic encycopedia. In G. Salvendy (ed.), Cognitive Engineering in the Design of Human-Computer Interaction and Experet Systems. Amsterdam: Elsevier, pages 207–223, 1987. (Smith-Catlin & Garrett, 1991) K. Smith-Catlin & L.N. Garrett. Hypermedia templates: An author’s tool. In Proceedings of the 3nd ACM Conference on Hypertext (Hypertext ‘91), pages 147–160, San Antonio, Texas, December 15–18, 1991. (Searle, 1980) J. R. Searle. Minds, brains, and programs. The Behavioral and Brain Sciences, (3):417–457, 1980. (Stotts & Furuta, 1989) P. D. Stotts, R. Furuta. Petri-Net-Based Hypertext: Document Structure with Browsing Semantics. ACM Transactions on Office Information Systems, 7(1), 1989. (Stotts & Furuta, 1991) P. D. Stotts & R. Furuta. “Hypertext 2000: Databases or Documents?” In Electronic Publishing, Vol. 4(2):119–121, June, 1991. 24

(Streitz, Hannemann, & Thüring, 1989) N. Streitz, J.Hannemann, & M. Thüring. From ideas and arguments to hyperdocuments: Traveling through activity spaces. In Proceedings of the 2nd ACM Conference on Hypertext (Hypertext ‘89), pages 343–364 , Pittsburgh, PA, November 5–8,1989. New York: ACM Press, 1987. (Streitz et al., 1992) N. Streitz, J. Haake, J. Hannemann, A. Lemke, H. Schütt, W. Schuler & M. Thüring. SEPIA: A cooperative hypermedia authoring environment. In D. Lucarella, J. Nanard, M. Nanard, P. Paolini (Eds.), Proceedings of the 4 th ACM Conference on Hypertext (ECHT ‘92), pages 11–22, Milano, Italy, November 30 – December 4, 1992, New York: ACM Press, 1992. (Thüring, Haake & Hannemann, 1991) M. Thüring, J. M. Haake & J. Hannemann: What’s ELIZA doing in the Chinese Room? Incoherent hyperdocuments – and how to avoid them. In Proceedings of the 3nd ACM Conference on Hypertext (Hypertext ‘91), pages 161–177, San Antonio, Texas, December 15–18, 1991. (Trigg, 1988) R. H. Trigg. Guided Tours and Tabletops: Tools for Communicating in a Hypertext Environment. ACM Transactions on Office Information Systems, 6(4):398–414, 1988. (van Dijk & Kintsch, 1983) T. A. van Dijk & W. Kintsch. Strategies of Discourse Comprehension. Orlando: Academic Press, 1983. (Waterworth, 1990) J.A. Waterworth. Hypermedia interfaces for hypermedia documents. In A. Rizk, N. A. Streitz & J. Andre’ (Eds.), Hypertext: Concepts, Systems and Applications, (Proceedings of the European Conference on Hypertext, ECHT ’90, Paris, France, November 1990), pages 356–358. Cambridge: University Press, 1990. (Whright & Likorish, 1990) P. Wright & A. Likorish. An empirical comparison of two navigation systems for two hypertexts. In R. McAleese and C. Green (Eds.) Hypertext: State of the art, pages 84–93. Oxford: Intellect, 1990. (Yankelovich et al., 1988) N. Yankelovich, B. J. Haan, N. K. Meyrowitz and S. M. Drucker. Intermedia: The concept and the construction of a seamless information environment. IEEE Computer, 21(1): 81–96. (Zellweger, 1989) P. T. Zellweger. Scripted Documents: A Hypermedia Path Mechanism. In Proceedings of the 2nd ACM Conference on Hypertext (Hypertext ‘89), pages 1–14, Pittsburgh, PA, November 5–8, 1989.

25