1. Introduction and Motivation

2. Performance Task

The primary function of the note-taking software is to improve the user's speed and accuracy as they enter notes about various domains of interest. A note is a short sequence of descriptive terms that describe a single object of interest. Example 1 shows a note describing a particular personal computer (recorded by the first author from a Usenet newsgroup during 1992):

Example 1:
4096K PowerBook 170, 1.4MB and 40MB Int. Drives, 2400/9600 Baud FAX Modem

Example 2 is a note describing a fabric pattern (recorded by the first author's wife):

Example 2:
Butterick 3611 Size 10 dress, top

Tables 5 through 11 later in the paper list sample notes drawn from seven other domains. The user may enter notes from different domains at their convenience and may use whatever syntactic style comes naturally.

From the user's point of view, the software operates in one of two modes: a contextual prompting mode, and an interactive graphical interface mode. In the first mode, the software continuously predicts a likely completion as the user writes out a note. It offers this as a default for the user. The location and presentation of this default must balance conflicting requirements to be convenient yet unobtrusive. For example, the hand should not hide the indicated default while the user is writing. Our solution is to have a small, colored completion button follow to the left and below where the user is writing. In this location, it is visible to either right- or left-handed people as they write out notes. The user can reposition the button to another location if they prefer. The default text is displayed to the immediate right of this button in a smaller font. The completion button is green; the text is black. The completion button saturation ranges from 1 (appearing green), when the software is highly confident of the predicted value, to 0 (appearing white), when the software lacks confidence. The button has a light gray frame, so it is visible even when the software has no prediction. Figure 2 portrays a screen snapshot of the software operating in the contextual prompting mode for a PowerBook note.

Figure 2: Screen snapshot of the note-taking software in contextual prompting mode for a PowerBook note. The two triangles in the lower left are scroller buttons.

The software's second mode presents an interactive graphical interface. Instead of requiring the user to write out the text of a note, the software presents a radio-button and check-box interface (what we call a button-box interface). With this, the user may select from text fragments, portions of notes called descriptive terms, by tapping on radio-buttons or check-boxes with the pen interface device. Each selection from the button-box interface is added to the current note. Intuitively, check boxes are generated to depict optional descriptive terms, whereas radio-button panels are generated to depict alternate, exclusive descriptive terms. For user convenience, the radio-buttons are clustered into panels and are sorted alphabetically in ascending order from top to bottom. To allow the user to add new descriptive terms to a button-box panel, an additional blank button is included at the bottom of each. When the user selects a radio button item, the graphical interface is expanded to depict additional choices corresponding to descriptive terms that follow syntactically. The software indicates its predictions by preselecting the corresponding buttons and highlighting them in green. The user may easily override the default selection by tapping the desired button. Figure 3 portrays a screen snapshot of the software operating in the interactive graphical interface mode for a PowerBook note.

Figure 3: Screen snapshot of the note-taking software in button-box mode for a PowerBook note.

The software is in prompting mode when a user begins to write a note. If the learned syntax for the domain of the note is sufficiently mature (see Section 6, Constructing a Button-Box Interface), then the software can switch into the button-box mode. To indicate this to the user, a mode switch depicted as a radio button is presented for the user's notice. A convenient and unobtrusive location for this switch is just below the completion button. In keeping with the color theme, the mode switch also has a green hue. If the user taps this switch, the written text is removed, and the appropriate radio buttons and check boxes are inserted. The system automatically selects buttons that match the user-written text. As the user makes additional selections, the interface expands to include additional buttons. When the user finishes a note, in either mode, the software returns to prompting mode in anticipation of another note.Of the functionality described here, our prototype implements all but the transition from button-box to contextual prompting. The mechanism for such a transition is machine dependent and is not germane to this research. Because the interface is constructed from a learned syntax, as the software refines its representation of the domains of the notes, the button-box interface also improves. On-line Appendix 1 is a demonstration of the system's operation in each of its two modes.

3. Learning a Syntax