Dasher Part III

Efficient communication with buttons: user trials: 

Efficient communication with buttons: user trials 24 May 2005 Ingrid Jendrzejewski with David MacKay, David Ward, Iain Murray, Alan Blackwell, Matthew Garrett, Mick Donegan, Tom Shorrock, Keith Vertanen and Chris Ball Department of Physics, University of Cambridge http://www.inference.phy.cam.ac.uk/dasher/

Text-entry with buttons: 

Text-entry with buttons In Efficient communication with buttons (2004), MacKay, Ball & Donegan discuss several ways in which Dasher might be adapted for use with discrete button pushes. Here we investigate three of these: One-button Static Mode Two-button Menu Mode Three-button Direct Dasher

One-button static mode: 

One-button static mode Arrow falls at a constant rate Clicking causes zoom-in on part of screen indicated by the arrow Clicking when arrow is above or below the red targets ‘backs up’ Dasher canvas only changes when user clicks

Clicker: 

Designed to investigate parameters g: timing accuracy, i.e. offset from ideal click time D: recovery time, i.e. time required by user to adjust to a zoom In each trial: Target is one pixel in width Target height is randomly set Screen is 200 pixels in height Arrow falls at constant rate Delay in arrow fall can be added Arrow speed can be changed Clicker

Clicking accuracy:one user, same mouse, different arrow speeds: 

Clicking accuracy: one user, same mouse, different arrow speeds

Clicking accuracy:several users, different mice, same arrow speed: 

Clicking accuracy: several users, different mice, same arrow speed

Clicking accuracy:several users, same mouse, same arrow speeds: 

Clicking accuracy: several users, same mouse, same arrow speeds

Recovery time: 

Recovery time In previous trials, the target was placed in the bottom 75% of the screen. Setting the target in the top half of the screen gives a first indication of recovery time Offset from ideal click time (ms) Position of target on screen (1=top, 0 =bottom) Here, arrow falls at 100 pixels per second When arrow speed was changed, the target placements for which users could accurately click also changed. Trials with several users using a range of arrow speeds from 50-200 pixels per second suggested a recovery time between 400-800 ms.

Recovery time: no initial delay: 

Recovery time: no initial delay No initial delay; No initial delay; No initial delay; No initial delay;

Recovery time: 400ms delay: 

Recovery time: 400ms delay Trials were repeated using delays of 200-1000 ms after the target was repositioned and before the arrow began to fall. Users clicked accurately with a delay of 400 ms.

Button Dasher user trials: 

Button Dasher user trials Versions of Button Dasher Results of user trials Parameters used to model real users Improving efficiency of text entry How does Button Dasher compare to other text-entry alternatives? Future work to be done

Two-button Uniform Menu Mode: 

Two-button Uniform Menu Mode Uniform boxes B = 5 s = 25 User sets the pace One button zooms One button cycles Like one-button mode: Clicking causes zoom-in on part of screen indicated by the highlighted box Dasher canvas only changes when user clicks

Menu mode: backing up: 

Menu mode: backing up Mistakes can be corrected by cycling through all the menu options. Clicking the ‘cycle’ button when the last menu box causes all the menu boxes to be highlighted, as shown above. Clicking the ‘zoom’ button at this point will back up or zoom out. Clicking the ‘cycle’ button at this point returns the user to the start of the menu. This cycle can be repeated indefinitely.

Two-button Non-uniform Menu Mode: 

Two-button Non-uniform Menu Mode Non-uniform menu B = 6 s = 25 r = 44 User sets the pace One button zooms One button cycles Like one-button mode: Clicking causes zoom-in on part of screen indicated by the highlighted box Dasher canvas only changes when user clicks

Three-button Direct Dasher: 

Three-button Direct Dasher Non-uniform menu B = 2 s = 25 User sets the pace One button selects top half of the screen One button selects bottom half of the screen One button backs up Like one-button mode: Dasher canvas only changes when user clicks

Writing speed: 

Writing speed

Information rate: 

Information rate

User errors: 

User errors

Information inefficient clicking: 

Information inefficient clicking Users will sometimes make non-efficient clicks. This can occur when users: Make a typing mistake Click too fast Become disoriented Cannot find target text Find that their target text lies on a boundary between menu boxes In the above example, a user wanting to spell ‘Dasher works well!’ may not be sure whether to click in the last or next-to-last box in order to zoom in on the correct text. Other cases can be even more ambiguous; unusual words that appear on menu boundaries can be particularly difficult to judge.

Zoom-outs: 

Zoom-outs Direct Dasher users only need to press one button to back up while Menu Dasher users must make a number of clicks equal to the number of menu boxes. Thus, Direct users tend to waste fewer of their clicks on zooming out.

Loops: 

Loops If menu users overshoot their target text, they have to spend a number of clicks equal to the number of menu boxes to through the menu cycle. This causes a significant amount of unnecessary button pressing.

Total percentage of clicks spent oninformation inefficient clicking: 

Total percentage of clicks spent on information inefficient clicking Menu Mode novices spend 40% to 90% of their clicks inefficiently. In contrast, by the end of an hour, both Direct Dasher novices make fewer information inefficient clicks than the fastest menu expert.

Slide23: 

Model parameters novice users Tback > Tsame & Tback > Tdiff for Novice A in all but the first exercise and for Novice P in the last six exercises.

Model parametersexpert users: 

Model parameters expert users For each user, in almost all cases, Tback > Tsame & Tback > Tdiff. For user R with over four hours of experience, Tdiff > Tsame.

Improving Performance: 

Improving Performance Recent software improvements Increase rate at which screen updates Screen no longer has distracting flash Users can keep keys depressed or press more than one key at once without causing crash Text wrapping no longer causes error Better language model Reduce safety factor

Language model: 

Language model Breath Dasher user trials: Dasher is driven by the continuous motion of one’s abdomen during breathing. Expert user shows higher writing speeds are obtained when using a Jane Austen training text rather than one trained with various English-language documents. All Button Dasher trials used the sub- optimal English training text, suggesting better writing speeds are attainable. Breath Dasher work by Tom Shorrock (2004), http://www.inference.phy.cam.ac.uk/ths23

Comparison of language models:expert user, two different training texts: 

Comparison of language models: expert user, two different training texts

Reduce the non-uniformity parameter: 

Reduce the non-uniformity parameter The non-uniformity parameter provides an overlap or ‘safety margin’ Reducing the safety margin increases the information content of each button press Theoretically, setting s=0 would improve results by ～7% For real users, there may exist a trade-off between the number of zoom-outs and the size of the safety margin such that a non-zero safety margin is preferable. Graphic courtesy of MacKay, Ball & Donegan, Efficient communication with buttons (2004).

How does Button Dasher compare?: 

How does Button Dasher compare? Graphic appeared in Ward, Blackwell & MacKay, Dasher - a Data Entry Interface Using Continuous Gestures and Language Models (2000) with the caption, ‘Comparison of Dasher to other devices. Hand printing required the user to print letters on a touchscreen. In QWERTY-tapping, a stylus selects characters from an on-screen QWERTY keyboard. ABC-tapping uses a stylus with letters in alphabetical order. Half-QWERTY uses half a QWERTY keyboard, with the other half accessible with a special key. Bellman is a keyboard designed with a probabalistic character layout strategy. TCK2 is a version of the ternary chorded keyboard. OPTI is a soft-keyboard with an optimized layout’. Button Dasher is driven by the discrete motion of pressing at most three buttons. How close can we get to the writing speeds attainable using text-entry alternatives driven by continuous gestures?

Regular Dasher and Keyboards: 

Graphic showing writing speeds using Dasher versus on-screen keyboard courtesy of Ward, Blackwell & MacKay, Dasher - a Data Entry Interface Using Continuous Gestures and Language Models (2000). Regular Dasher and Keyboards Regular Dasher is driven by two-dimensional, continuous motion (mouse, touchscreen, stylus, eyetracker). Results shown below are for novice users Each exercise involved five minutes of dictation with an optimally-trained Dasher Keyboard typing refers to ten-fingered typing on a QWERTY keyboard. Despite using only three buttons for text entry, novice Direct Dasher users approach the performance of the slowest (regular) Dasher users, even using a sub-optimal training text!

Eyetracking results: 

Eyetracking results Eyetracking results courtesy of Ward & MacKay, Fast Hands-free Writing by Gaze Direction (2002), Nature 418, p. 838. Button Dasher writing speeds are comparable with those achieved using a gaze-driven onscreen keyboard, but are somewhat slower than those achieved using regular Dasher driven by an eye tracking device.

Breath Dasher: 

Breath Dasher Breath Dasher work by Tom Shorrock (2004), http://www.inference.phy.cam.ac.uk/ths23 Even though they are using discrete button presses instead of continuous movement, Button Dasher novices outperform their Breath Dasher counterparts--even with a sub-optimal language model!

Next trials under consideration: 

Next trials under consideration Direct mode, expert users, Austen training text, updated software The results presented here do not represent the upper bounds on the writing speeds that learning or experienced users can achieve. It would be useful to determine the maximum writing speeds and information rates attainable using updated software and an optimal training text. Direct mode, variable safety margin For the purpose of these trials, the safety margin was set somewhat arbitrarily. More work should be done to quantify exactly how small this overlap can be before it has a negative effect on user performance. Menu mode, B=2 It was initially thought that one of the advantages of the direct mode was that it would operate essentially as a two-button mode, relying primarily on only the two zooming-in buttons with the third button used only rarely. As this third button is turning out to be an important feature in direct mode, it could be interesting to reconsider ways in which users who can't perform time-critical tasks could most effectively communicate with only two buttons.

Future trials: new button modes: 

Future trials: new button modes Chris mode, novice & expert users When an expert user spent 10 minutes using Direct Dasher with no safety margin and a very large non-uniformity, information rates and writing speeds were comparable to Direct Dasher results. This suggests a two-button adaptation of Direct Dasher in which one of the zoom-in buttons is controlled by Dasher, leaving the user to control the zoom-out button and the timing of one zoom-out button. Compass mode A four-button mode allows users to navigate in both dimensions of the Dasher canvas.

Special thanks: 

Special thanks David MacKay Chris Ball Tom Shollock The staff of the Whipple Museum of the History of Science and other volunteers Work on Dasher has been supported by the Gatsby foundation and by Applied Science Laboratories.