3.4 Early Work with Users and the Domain
Seven mutually informing activities take place in parallel during the early stages of the design: user categorization and profiling activities, scenario generation, building of a relatively coarse-grained task model, task description and referent identification, early domain modeling, generation of design sketches, and generation and evaluation of low-fidelity prototypes. User categorization and profiling is not discussed except for a consideration of its consequences in Sections 3.4.2, 3.4.5, and 3.5.1, and low-fidelity prototypes are not discussed, except for a brief mention in Section 3.6.2.
Throughout much of the early design activities, there is a strong emphasis on opportunistic design practice while seeking to avoid over-commitment to early design artifacts. Some of the early design activities are concerned with finding visions and paths for the design—sometimes by looking in detail at the real world and sometimes through the formulation of tentative and partial design solutions that might be very concrete in nature. Subsequent early design activities abstract away from these details. Later, in Section 3.5, the resultant abstract models are used to systematically develop aspects of the user interface design. The processes of abstraction and subsequent model-based design can help in avoiding the retention of early concrete design ideas that might limit the concrete user interface design. In practice, this process must, at the least, be accompanied by user evaluation of the developing design.
Early on, it is often a challenge to formulate abstract descriptions; sometimes it is easier to adopt the design tactic of formulating a description that includes some less than desirably abstract detail, and then to remove the non-abstract detail.
3.4.1 Scenario Generation
In Idiom, scenarios are stories, natural language descriptions of use of the system, written in user terms and language. Since scenarios are expressed in user language, they may be written, reviewed, and modified by both users and designers. Scenarios can be constructed for identified types of users, or scenario data can inform the identification of user types in user profiling activities. Scenarios can be written to capture user goals, expectations, and some of the context in which the users work. Scenarios can also be used for requirements capture, for illustration and visualization of system use, and to facilitate communication between designers and users. Later, in a process of abstraction away from the concrete detail of scenarios, the scenarios provide data that is used in task modeling and referent identification. More generally, scenario generation and task modeling and description are all mutually informing.
Much of the subsequent chat example is centered around a simple scenario depicting the initiation of a conversation, shown in Figure 3.3. Some of the detail in this (monitoring the agreement of invitees) changes as the design progresses.
){kind=link}
Figure 3.3 A scenario illustrating conversation initiation
3.4.2 Coarse-Grained Task Modeling
Idiom approaches the functional side of system specification via an abstract coarse-grained (or high-level) task model that is constructed with the users' goals in mind. A task model is required for each category of user. The level of abstraction utilized within at least the top-level tasks is similar to that of essential use cases—a kind of task model that is free of technology and implementation detail [Constantine and Lockwood 1999] (and see also Chapter 7). As with OVID [Roberts et al. 1998, p. 62], Idiom is not prescriptive as to the notation used for task modeling. Simple textual task decomposition styles are effective for both abstract and more detailed task models. One such style is used throughout this chapter. Collins [1995] recommends a similar, but slightly simpler technique that does not incorporate some of the control structures used here.
For the chat example, a simple high-level task model for a user who communicates with other users is shown in Figure 3.4. This is a rendition of a particular stage in the development of the model, along with comments, introduced with //, to the author and readers. The top-level task, Communicate, is motivated by the user goal to communicate with other users. Some sub-tasks that help realize this goal, such as address book–related tasks, are not included in the model in order to simplify and reduce the size of the example. Similarly, the task model omits the consideration of many exception conditions, such as everyone else leaving a conversation. The high-level task model in Figure 3.4 was constructed with extra detail that was then removed, as suggested in the preamble to Section 3.4.5 The scenario in Figure 3.3 illustrates the InitiateConversation and RespondToConversationInvitation tasks in Figure 3.4.
){kind=link}
Figure 3.4 Coarse-grained task model
A key to successfully using a textural form like the task model in Figure 3.4 is to accept some degree of informality in indicating interleaved task flow and thereby gain a way of indicating unordered sequences of optional activities while still using a relatively tractable syntax. The third (loop...) and fourth (start or resume...) lines in Figure 3.4 express a user-driven choice of interleaved concurrent task execution, as might be performed by users in a modern windowed environment with different windows providing different task execution contexts. Beyond this, there is no task model consideration of the sequencing of interleaved tasks, such as communicating with two sets of users via two different and simultaneously ongoing conversations.
As in Idiom94, task models can incorporate the notion of system feedback, as in
choose and invite participantsfeedback
This is similar to user action notation's (UAN's) use of an exclamation mark [Hartson et al. 1990, Hix and Hartson 1993].
3.4.3 Interaction Exploration Using Sequence Diagrams
UML interaction diagrams (sequence and collaboration diagrams) are normally used to illustrate use cases and depict details of inter-object interaction within the object model [Booch et al. 1999]. As noted in Section 3.2, while very useful for software engineering purposes, the documentation of internal object interactions is not generally useful for interactive system design (Constantine and Lockwood also note this in Chapter 7).
However, sequence diagrams can be used to visualize and illustrate interaction between users and a system. This technique is experimental; it seems useful as a way of selectively visualizing and exploring some aspects of user interaction. UML sequence diagrams that depict user-system interaction can be drawn at a high level of abstraction, or they can have increasing amounts of detail added to them. At a high level of abstraction, with both user and system actions illustrated by message sends as in Figure 3.5, the notation is close to essential use case notation [Constantine and Lockwood 1999] (and see also Chapter 7).
){kind=link}
figure 3.5 Sequence diagram showing the initiation of a conversation and the associated issuing of invitations
The addition of task information to sequence diagrams can be useful in the visualization of the developing task model. Here, these sequence diagrams tend to show only one user, by reflecting the single-user nature of task models. Examples are illustrated in Figures 3.6 and 3.7. The advantages of such task-sequence diagrams seem to be that they do the following:
){kind=link}
){kind=link}
figure 3.6 Task-sequence diagram for a response to a conversation invitation when already online
figure 3.7 Task-sequence diagram for online connection to the chat system and a response to a pending conversation invitation
Combine user tasks with depictions of interactions between the user and the system, and thereby start to "bring the task model to life" as part of the developing user interface and system design.
Help in the examination and further development of the envisioned task model.
While this diagramming technique can be used to illustrate increasingly fine-grained task models, indications are that the point at which to stop this activity is when the high-level task model has been satisfactorily explored. However, since sequence diagrams are notationally heavyweight and have only limited applicability for high-level system visualization purposes, they may either be omitted from future versions of Idiom or only be used for occasional task-model visualization. Certainly, the technique would be tedious and time-consuming in use at fine-grained detail for large systems, or even all the way across a set of high-level tasks.
3.4.4 Describing Tasks and Identifying Referents
Users have goals that they attempt to realize by task performance. During task performance, users manipulate referents, which are user-world objects that are of interest or utility to users in the performance of their goal-directed task behavior. (See the CHI97 framework in [van Harmelen et al. 1997] or see Chapter 10.) Idiom encourages the discovery of referents from scenario data as part of the process of extracting task descriptions from scenarios. These referents are later represented by types and attributes in the domain and core models and are used in the identification and design of a set of contexts for task execution to be implemented in the interface.
The procedure for describing tasks and recording referents in Idiom is to list selected tasks defined in the task model, to describe these tasks in user language (some of this detail may appear in enacted form in the scenarios), and then to find and describe the referents used in each task from the task description. Referents appearing in the task description are underlined and are listed after the task description. Focusing on nouns in the task description is one good way to find potential referents. Each referent description includes (a) a resolution of any naming ambiguities in the task descriptions (if those ambiguities have not, for one reason or another, already been resolved) and (b) a user language description of the referent.
In Figure 3.8, a task has been rendered as a brief textual description; the underlined words refer to objects or relationships. Referents used in the task are listed below the task, together with a resolution of any ambiguities in terminology, the chosen names being unbracketed. Not all of the extracted referents are model objects in themselves. For example, invitee status could be represented in a domain or core model using a UML-association class, or by the use of different associations, as shown in Figure 3.9; the former is the more elegant solution.
){kind=link}
){kind=link}
Figure 3.8 A task description together with referent descriptions
Figure 3.9 Modeling invitee status (with the preferable solution on the left)
This phase of referent identification is not the only source of referents. Scenarios and business input may provide referents, as may users, domain experts, and designers. Whatever the source, referents should be recorded against the task descriptions in which the referents are used. This information will be used to discover if any task execution contexts can be merged during later user interface design (see Section 3.5.3.1).
3.4.5 The Domain Model
The domain model uses referent types to represent referents in the users' world and uses associations to represent relationships between those referents. First versions of the domain model should be developed as soon as possible to encourage the designer to rapidly understand the contents and structure of the application domain. Skilled object modelers may wish to start modeling almost immediately, using the later output of the referent identification as an additional source of information for their modeling activity, while less experienced object modelers may wish to start modeling only after the referent identification has produced some output, without worrying too much, at first, about cardinalities. Whichever approach is taken, the domain model must articulate the users' understanding of the domain before the core model is constructed (see Section 3.5.1).
Should users be modeled in the domain model? A domain model for an architectural CAD system would not include the users of that system [van Harmelen 1994]. However, in systems in which users need to communicate and interact with each other, users appear in the domain model. However, it should be noted that modeling provides only a very abstract representation of interactions between users; modeling is no substitute for field work that leads to a rich understanding of ways in which the user activity can be supported.
The chat example includes users who communicate with each other. For these users, initial referent identification leads to the model shown in Figure 3.10(a). Figure 3.10(b) shows a slightly enlarged domain model incorporating another referent—an address book or "buddy list" that provides referents for use while initiating a conversation.
){kind=link}
Figure 3.10 (A) Basic domain model and (B) the domain model used later in this chapter
Different types of users may use different kinds of referents in distinct domain models. Thus, if there were chat system administrators who were responsible for chat server administration, their domain model would have to include referents representing servers and provide operations on those types to enable and support the administrators' activities. To accommodate different kinds of users, designers can think in terms of different compatible domain models, or in terms of different aspects of a single domain model.
3.4.6 Visualization Using Sketches
There is a strong need for the users and designers to gain a feel for a system, and how it might operate, very early in design. Sketches—freehand drawings of the system's appearance—can be used to depict user interfaces while exploring system functionality, behavior, and use. The process of drawing sketches by hand is direct and immediate, without the overhead of computer use. Sketches can be combined to form storyboards, which are sequences of depictions of the system over time, or interaction sequences [van Harmelen 1994], which are sequences of renditions of the screen state separated by depictions of user interactions that cause the transitions between the renditions, see Section 3.6.1. For user interface design, sketches can include ideas of interaction along the lines of the example shown in Figure 3.11. All of these sketching techniques provide user interface representations that can be discussed, compared, mutated, redrawn, and even used as prototypes, all without any need for polished drawing skills. As such, the technique is simple, useful, and laudable. Winograd [1995] makes some similar observations about sketching for software design.
){kind=link}
Unfortunately, there is a danger that, as an early design activity, sketching can lead to (a) inappropriate assumptions on the part of the designers and (b) incorporation of premature and inappropriate low-level interaction detail into the developing design. Once recorded in an early sketch, there can be a strong tendency for ideas of system appearance, interactive characteristics, and functionality to continue unrevised in the developing design. As pointed out by Constantine [2000], avoiding these problems is a good reason for proceeding directly to an essential view of the system. Idiom's approach involves retaining sketching and early concrete visualization while being aware of their limitations.
The original version of Figure 3.11 was drawn on a whiteboard within an hour or two of starting this design study. The figure provides examples of the kinds of problems that early sketches can introduce:
Figure 3.11 Rough early design ideas about interaction flow copied from a sketch on a whiteboard
-
Inappropriate assumptions: The rendition uses, to a computer-literate designer, a useful interaction technique that allows for revision of choices while formulating a list of conversationalists. Interestingly, in showing just how wrong designer choices can be, the very first user who was asked how to initiate conversations rejected this approach in favor of manipulating (in some unknown way) photographs of the users with whom he wanted to chat. 6 Yet other users might want a shortcut in order to chat with a single online user.
Sub-optimal concrete design choices: The status window was soon abandoned as polluting the view space, because, if it were needed, the status of conversationalists (accepted, refused, not online) could be shown in each user's chat window, given a likely-to-be-found solution to the problem of displaying invitee status for a set of invitees.
Even with the danger of these kinds of problems, sketching is extremely valuable and can be used effectively if designers are prepared to revise solutions. Involving users, working with user-derived information in the design process, abstracting and then using the resultant models in further design, and testing prototypes with users are all ways of encouraging suitable revision while gaining a deeper understanding of and feel for the developing interactive system design.