Steve Whittaker, Julia Hirschberg
Steve Whittaker and Julia Hirschberg
180 Park Avenue
Florham Park, NJ 07932
The following abstracts are from recent issues and the forthcoming issue of ACM's Transactions of Computer Human Interaction (ToCHI). They are included here to alert interactions' readers to what research is being done in the field of Computer Human Interaction. The complete papers, when published, can be found in ACM's Digital Library at www.acm.org/pubs/contents/journals/tochi/2001-8-2/p150-Whittaker and www.acm.org/pubs/citations/journals/tochi/2000-8/#3.
Over the past 20 years, the popular technology press has repeatedly predicted the demise of paper, arguing that the paperless office is imminent. According to these experts, the growth of the Web, e-mail and personal digital assistants means that paper files, folders, and memos will be swept away by their electronic equivalents. Paper will be made irrelevant by the greater ease of distributing and searching digital data and the emergence of public digital data on the Internet. Yet, even casual observation reveals that paper is still central to office work, and sales of technologies that require paper, such as printers, are strong. Paper is still very much with us. Our study investigates why.
To understand the function of paper in office work, we examined the information available on paper to office workers. Does paper still play a role in the modern office? What uses do paper archives serve? Do they contain irreplaceable data or disposable information? What makes information on paper valuable? How are archives acquired, maintained, and deleted? And what are people's strategies for personal information management?
We collected qualitative and quantitative data from 50 office workers. Our investigation coincided with an office move, and workers had recently sorted through their files on paper to prepare for the move.
Paper was obviously not defunct. People kept large, valued archives, despite the increased availability of digital materials and the growth of the Internet. The scale and growth rate of, and attitudes toward, paper archives suggested they were still a valuable resource; younger workers relied on paper as much as older ones. The value of paper archives partly derived from paper's affordances. Paper is more available and better supports reminding. Although people discarded a small part of their archives, the discarded information was extraneous rather than obsolete. People acquired extraneous paper data for two reasons: they lacked time to process new information, causing unread information to accumulate, and they were unable to gauge the future utility of information, leading them to save data of questionable value.
A major reason for keeping paper data was that it was irreplaceable. Nevertheless, much of people's information consisted of personal archives of publicly available documents because of the greater availability of personal documents along with a mistrust of storage institutions, including the Web. Certain individuals established themselves as informal librarians and have became a resource for others. People also kept data for sentimental reasons, even when there was little direct use for it.
Workers also experienced problems with processing incoming paper documents, deciding their future value, and organizing them for future access. We contrasted two management techniques. Whereas filing applied a formal structure to paper data, filers also incurred costs. Consistent with prior research, people found it difficult to operate filing effectively by applying consistent labels and organizational structure [1, 2]. We also observed premature filing: to keep a clear desktop, filers sometimes spent time organizing information of questionable value. Once filed, information became difficult to discard, given the effort of organizing it. In consequence, filers ended up with greater amounts of information than pilers. In contrast, there were surprising advantages to piling. Pilers benefitted from greater availability of recent information and experienced fewer overheads in managing and cleaning up archives. The main limitation of piling was scale: pilers encountered difficulties once piles had begun to multiply. Many of these observations are common to other types of personal information management, such as e-mail  and voice mail .
Our work has technological implications. The need for personal rather than public data argues against centralized information storage mechanisms such as the network computer. We can improve the usability of digital archives by adding paper's affordances to electronic documents, as well as by applying text indexing and learning techniques to aid the classification and filing of digital documents and avoid premature filing.
1. Kidd, A. The marks are on the knowledge worker. In Proceedings of CHI'94 Human Factors in Computing Systems (ACM Press, New York), 1994, pp. 186191.
2. Malone, T. How do people organize their desktops? Implications for the design of office systems. ACM Transactions on Office Information Systems 1 (1983), pp. 99112.
3. Whittaker, S., Davies, R., Hirschberg, J., and Muller, U. Jotmail: a voicemail interface that enables you to see what was said. In Proceedings of CHI2000 Conference on Human Computer Interaction (2000), pp. 8996.
4. Whittaker, S. and Sidner, C. E-mail overload: exploring personal management of e-mail. In Proceedings of CHI96 Conference on Computer Human Interaction (ACM Press, New York), 1996, pp. 276283.
Bradley T. Vander Zanden and Richard Halterman
University of Tennessee
This paper describes a technique that reduces in graphical interfaces the storage required by spread-sheet like constraints. Constraints allow a programmer to express a relationship among objects and then use a constraint solver to automatically maintain the relationship. For example, Figure 1 shows a number of constraints that center a label within a labeled box and that make the labeled box just large enough to accommodate the label. If the user edits the label, then the constraint solver automatically re-evaluates the constraints that compute the width and position of the label and the width of the frame.
Constraints can reduce the complexity of developing graphical interfaces because the constraint solver performs much of the bookkeeping that the programmer would otherwise have to perform. In the foregoing example, the programmer can effect the operation for editing the label by simply changing the label's text property. The programmer does not have to remember to recompute the widths of the label and frame or to recenter the label, because those actions are done automatically by the constraint solver.
The constraint solver performs this bookkeeping by using a dataflow graph to keep track of the relationships among variables and constraints. Figure 1(c) shows a sample dataflow graph for the constraints associated with the labeled box. From the figure you can see a directed edge from a variable to a constraint if the constraint uses that variable. There is a directed edge from a constraint to a variable if the constraint assigns a value to that variable.
A constraint solver uses the dataflow graph to locate the constraints that must be resatisfied when a variable is changed by either the user or the application. For example, suppose the user edits the label's text string. The constraint solver can follow the dataflow edges from the label's text property to determine that constraints C1 to C3 must be re-evaluated. The constraint solver can also use the dataflow graph to determine the order in which the constraints should be re-evaluated. In this instance, C1 should be evaluated first, and then C2 and C3 can be evaluated in any order.
This paper addresses the problem of reducing the amount of storage consumed by a dataflow graph. Studies have shown that constraints can exact a significant storage toll on programs. This storage cost can adversely affect the execution times of programs managing a large number of constrained objects if virtual memory must be used to meet their storage demands. Dataflow graphs account for up to 50 percent of the storage required by constraint systems; thus, reducing the storage cost of these dataflow graphs is an important goal.
This paper's approach to storage reduction is based on the observation that objects that use the same constraints have the same dataflow graph. Consequently, you can store a pattern of a dataflow grapha model dataflow graphin a common place and then use the pattern to derive explicit dataflow edges on demand. For example, a model dataflow graph for a labeled box would keep track of the fact that in every labeled box, constraint C1 depends on the label's font and text properties and sets the label's width property (each labeled box object has an instance of C1). If the user changes either the font or text property of an arbitrary labeled box, the constraint solver can use the model dataflow graph to determine that that labeled box's C1 constraint must be re-evaluated. It can further use the model dataflow graph to determine that the labeled box's C2 and C3 constraints must be re-evaluated.
Because thousands of objects may be created from the same prototypical object, the storage savings can be considerable. For example, a computer-aided-design system might use thousands or tens of thousands of labeled box objects. All these labeled box objects can share the same model dataflow graph. A considerable storage savings can therefore be realized in a large application that manipulates thousands of these objects. Indeed, the experiments that we report in this paper show that more than 75 percent of the explicit dependencies in most applications can be eliminated with the model dataflow approach.
The paper first presents an overview of the model dataflow scheme and discusses detailed algorithms for implementing model dataflow graphs. Next, it discusses several extensions that make model dataflow graphs more effective in reducing storage. Finally, it presents a number of experiments confirming that model dataflow graphs can significantly reduce the storage of a constraint solver.
Anthony J. Hornof
Dept. of Computer and
University of Oregon
This paper demonstrates that a clearly labeled visual hierarchy will significantly reduce visual search time and motivate a more systematic search strategy. It also refines an understanding of Fitts's Law.
To improve the speed and quality of visual communication, screen layout design guidelines recommend that screen layouts be organized with a clear visual structure that allows the user to visually navigate the layout systematically and somewhat predictably. Although this might seem an obvious recommendation, many screen layouts in computer interfaces and Web sites do not impose a clear and useful structure and thus make visual tasks unnecessarily difficult. Few experimental studies have verified or measured the effect of incorporating a clear visual hierarchy.
The experiment presented in this paper verifies and measures the effect. In the experiment, participants were presented with screen layouts similar to those shown in Figure 1. Note that the layout on the left does not incorporate group labels to help guide the viewer in his or her search and the layout on the right does incorporate such labels.
To emphasize fundamental mechanisms and strategies involved in visual search rather than semantic issues, the visual layouts used in the actual experiment were simpler than those shown in Figure 1. All of the items in the layout were more concise, and participants were told the exact target item and target group label for every layout they were shown.
Participants found and clicked on the precued target item. In some blocks of trials, the groups had labels that helped guide the participant to the target. In other blocks of trials, there were no labels. The number of groups, each of which always contained five items, also varied from block to block.
The observed visual search times demonstrate that a labeled visual hierarchy can be searched significantly faster than an unlabeled visual hierarchy. Details in the search time data, such as a more pronounced number-of-groups effect for unlabeled layouts, also suggest that people adopt fundamentally different search strategies for labeled versus unlabeled visual hierarchies.
Visual search of a computer screen is often followed immediately with the user clicking on the target item with the mouse. In order to develop more accurate models of visual search and manual selection, detailed and precise measurements of both subtasks are required. To that end, this paper also introduces a new experimental paradigm for separating visual search time from mouse pointing time for experiments that incorporate these two subtasks. The paradigm incorporates a point-completion deadline. Once the participant begins to move the mouse from the cued position, he/she has a limited amount of time to click on the target before the trial is interrupted. Search time can thus be measured from when the layout appears to when the participant starts moving the mouse to the target.
The mouse-pointing times reported in this paper contribute to the ongoing refinement of Fitts's Law as a means of predicting mouse pointing times by providing evidence that (1) target width is best measured along the line of approach and (2) typical pointing experiments may underestimate the parameters of Fitts's Law, and (3) Shannon's formulation of Fitts's Law does not fit data better than Welford's version.
Looking to the future, this research will contribute to the construction of a tool that designers could use to evaluate the efficiency of visual layouts. The tool will take as input (1) a screen shot of a visual layout and (2) a description of the useful visual targets in the layout. The tool will predict the visual search time required to find each target and perhaps the visual scan pattern that a person would follow to find each target. This experiment contributes to the theoretical underpinnings of such a tool by carefully measuring the difference in search times for two types of layoutslabeled and unlabeled visual hierarchies.
Figure. A labeled box
(a), the constraints that lay out the elements of the labeled box
(b), and the dataflow graph that the constraints generate. The constraints make the frame slightly larger than the text label and center the label within the frame. Assumed in the figure is that the labeled box consists of three objects: (1) a grouping object, called labeled_box; (2) a rectangle, called frame; and (3) a text object called label. Also assumed is that the frame and label objects have a parent property that points to labeled_box. Finally, compute_width is a function that computes the width of a string on the basis of its font.
Figure. A real-world analogy to the experimental task used in the visual search experiment. The layout on the left does not incorporate a labeled visual hierarchy to help the user find a story about the Mets baseball team. The layout on the right does incorporate a visual hierarchy to help with such a task (adapted from NYTimes.com, 5/30/01).
©2001 ACM 1072-5220/01/0700 $5.00
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee.
The Digital Library is published by the Association for Computing Machinery. Copyright © 2001 ACM, Inc.