Script of Demonstration of Dasher (using Lotus Screen Cam) - by David Ward and David MacKay http://www.inference.phy.cam.ac.uk/dasher/movies/dasherlotusdemo.zip http://www.inference.phy.cam.ac.uk/dasher/Demonstrations.html 1. Dasher is a text-entry interface that uses continuous gestures to select sequences of letters from a language model. The user drives the interface using a two-dimensional pointer. Here we use the mouse. 2. Before we describe the interface in detail, we will show you what it looks like when the user slowly enters the phrase "introduction to dasher". 3. 4. We will now describe the two ideas behind Dasher - first, what the objects in this two-dimensional world are, and how their sizes are determined; and second, how the pointer is used to navigate in the world. What the objects in this two-dimensional world are, and how their sizes are determined ====================================================================================== 5. When the user starts writing, all possible starting letters are displayed in boxes at the right-hand side of the screen. Here, the letters have been arranged in alphabetical order from a to z, and space is represented by the underscore character "_". 6. The size of each box is proportional to the probability of that letter according to a language model. At the start of a document, all letters are roughly equally probable, but "q", "x", and "z" are slightly less probable than the others. 7. Within each of the starting letter's boxes are smaller boxes, one for each of the possible second letters. 8. The size of each of these second-letter boxes is again determined by the language model. If the first letter is "h" then the vowels a, e, i, o, and u are the most probable second letters. The other letters are less probable, so they are not yet visible at this resolution. 9. If the first letter is "i", the consonants d, f, l, m, n, and so forth are the most probable letters. By making the size of a letter's box proportional to its probability, we make it easier for the user to find and select the more probable letters. 10. The user selects a sequence of letters by moving the point of view into the box of the first letter then into the sub-box of the second letter, and so forth. Let's write "introduction to dasher" again. 11. Because "n" is a probable letter after "i", it is easy to select "in". After "in", the most probable letters are "t", "s", and the space character "_". We move into the "t". Notice how the language model is now offering words like "interest" and "into". We want the word "introduction" so we hunt for the "r" box. Once we have got close to the "r" box, the words "introduce" and "introduction" both appear. It takes only a very small gesture to select the word "introduction". 12. Because the language model has been trained on correctly spelled English, it is very hard to mis-spell words. If we do want to write a string that is relatively improbable according to the language model, we can write it, but it will takes us longer to select the improbable letters. For example, let's write "kournikova_". How the pointer is used to navigate in the world ================================================ 13. The intuitive idea behind Dasher is that the user should simply point where they want to go. The vertical dimension of the pointer controls where on the right hand side we are zooming in on. The horizontal dimension controls the speed at which we zoom in. Note that the dynamics of the interface are independent of the letters and the language model. 14. Now I shall write some text at a faster speed.