Accelerated learing of chording with spatial mnemonics

I have studied to what degree spatial mnemonics can be used to assist users to memorise or infer a set of text input chords. Users mentally visualise the appearance of each character as a 3x3 pixel grid. This grid is input as a sequence of three chords using one, two or three fingers to construct each chord. Experiments show that users are able to use the strategy after a few minutes of instruction, and that some subjects enter text without help after three hours of practice. Further, the experiments show that text can be input at a mean rate of 5.9 words per minute (9.9 words per minute for the fastest subject) after 3 hours of practice. On the downside, the approach suffers from a relatively high error rate of about 10% as subjects often resort to trial and error when recalling character patterns.

The spatial mnemonics allow subjects to derive the chords and hence achieve accelerated learning. Symbols are entered as sequences of three finger chords. The principle behind the chording task is as follows. First, the subject visualises the character as a 3x3 grid of pixels. Then the subject inputs this character grid in three “horizontal” steps from top to bottom. In each step one chord is entered representing one scan-line of the character grid. For each chord one, two or three keys are used simultaneously. After the three chords are entered the required character is retrieved and the process is repeated. The following chording alphabet was developed.

The experiment software is implemented as two Java Applets running in a web browser. One applet comprises the character pattern reference and is presented as a virtual QWERTY keyboard. By clicking on one of its keys a graphical representation of the corresponding 3x3 pixel map for the character is displayed for 1 second. The second applet comprises the text entry application, and has two components. The top line shows a phrase to be copied and at the bottom a multi-line text entry area is provided. As users enter text the text is displayed in the text entry area. The text scrolls automatically so that one can see a trail of previously entered phrases. The right side of the screen was used to display short instructions on how to operate the system. Both applets logged user activity locally to disk.

The applet is configured to accept keystrokes from the SPACE, J and K keys as this is a hand position touch typists are used to that naturally fits the thumb, index finger and long finger of the right hand. The SPACE, J and K keys represented the left, middle and right chord-keys, respectively. A timeout mechanism is provided such that if the subject halted for more than 500 ms the system would reset the construction of the current character and start over. This mechanism allowed subjects to redo a character by waiting if they got confused.

The left figure below shows the mean words per minute (WPM) based on the means for the fourteen subjects plotted across the four sessions. The plot reveals that there is a steady increase in productivity from a mean of 2.9 WPM during the practice session to 5.9 WPM during the fourth session.

In order to assess learning the number of accesses the subjects made to the character reference during each text entry session was analysed. The right figure below shows the mean learning curve for the 14 subjects across the four sessions. The vertical axis represents the number of accesses made to the character reference table.

You can read more about the text entry speed an error rates of the technique in:

1. Frode Eika Sandnes, "Can Spatial Mnemonics Accelerate the Learning of Text Input Chords?", Proceedings of the Working Conference on Advanced Visual Interfaces AVI 2006, Venezia, Italy, May 23-26, ACM Press, pp. 245-249, 2006.

An account of how errors can be detected an corrected for the proposed technique is given in:

2. Frode Eika Sandnes, Yo-Ping Huang, "Chording with Spatial Mnemonics: Automatic Error Correction for Eyes-Free Text Entry", Journal of Information Science and Engineering, Vol. 22, No. 5, pp. 1015-1031, 2006.

An account of the physical characteristics of chord sequences are under consideration.