Stephen Bique, Ph.D.
To test the application, visit Link to Chat Assistant.
We formed n-grams from the databases provided by Swiftkey. Visually we represent a picture of 3-grams using a word cloud formed from the actual n-grams created. Evidently, the phrases depicted are quite common. Indeed, words are not randomly put together.
Consider a phrase w x y where w, x, and y denote words. Perform the following steps, halting as soon as we are able to predict:
1. If w x y z is in 4-grams, predict z.
2. If x y z is in 3-grams, predict z.
3. If w y z is in 3-grams, predict z.
4. If y z is in 2-grams, predict z.
5. If w x z is in 3-grams, predict z.
6. If x z is in 2-grams, predict z.
We use so-called 2-grams, 3-grams and 4-grams with a novel skip one strategy whenever we do not find the desired n-gram. Actually, we return a list of all of the completion words z for the observed phrase. We have over simplified here as we separate n-grams depending upon whether or not they were formed from skipping one word.
The GUI included the familiar copy and delete buttons, and has a professionally designed layout suitable for various devices. Experiment with resizing the text input area. Notice the Help and About tabs at the top right of the screen. The online help is summarized on the next slide.
1. Note a space is automatically added before pasting