What solutions could we propose to eliminate this procrastination? If adding a person's name [Footnote: For brevity, we'll refer to a person's name, address, phone numbers, e-mail, etc. as a "name". Whether the aggregate information or just a person's first or last name is intended should be clear from context.] to your organizer were fun (say with your choice of inspirational message, gratuitous violence, or a lottery ticket), you might add names more readily. Avoiding this whimsy, we could get the desired effect by just making it faster to add a person's name. For this reason paper organizers use index tabs; electronic organizers use automatic filing. To be faster still, an organizer could read a card or handwritten note (via optical character or handwriting recognition). Applying artificial intelligence ideas, we could even imagine an organizer that predicts what you need to write and does it for you.
This paper describes an electronic organizer that is almost as fast as a slip of paper, and certainly much faster than previous organizers. It uses commercial hardware (Newton, described in Section 2). ts software has three interface components designed to speed adding a person's name (described in Section 3): handwriting recognition, adaptive menus with recent values, and predictive fillin. The primary contributions of this paper are detailed evaluations of the benefits of these three components (described in Section 4).
Adding a person's name into an organizer is a special case of capturing and organizing information. This is a ubiquitous task. Big businesses institute careful procedures with custom forms and databases, but there are billions of smaller, one-or-two-person tasks which could be done more efficiently and accurately if getting information into a computer were easier. Even small gains would be repeated many times over whenever someone needed to collect information to make a decision, monitor a process, or investigate something new. The secondary contributions of this paper are a consideration of how the three components may be applied more broadly (described in Section 5).
These three interface components are robust if familiar. Whether they are "intelligent" is arguable. Some advocate a behavior-based definition which may also apply here, i.e., that the question of whether a device is intelligence or not should be answered by examining a its behavior rather than its internal processes and representations (e.g., Agre & Chapman, 1987; Horswill & Brooks, 1988). Even if it does not, our goal is to address the question of how much intelligence, agency, or support one wants in an interface (Lee, 1990; Rissland, 1984). We assert: as much as will speed the user's performance of the task. Furthermore, much research is directed at automatically learning what we will hard-code in this study (e.g., Dent, Boticario, McDermott, Mitchell, & Zabowski, 1992; Hermens & Schlimmer, 1994; Schlimmer & Hermens, 1993; Yoshida, 1994). Even if learning works perfectly, is the result worthwhile? We claim the answer can be found in an empirical study of the usefulness of various user interface components.