In this module we meet our first machine learning approach. Classification and regression trees (CARTs) are widely-applicable models for making predictions. We can use them for letter-to-sound, prosody, and many other tasks.
Start with these two blog posts. The first one will give you the general idea of a decision tree. Most of the video is about building the tree (this phase is called “training” in machine learning). Training is done just once, when we build a system. We can then make predictions for new samples using this tree (this phase is sometimes called “testing” in machine learning). The second blog post will help you understand entropy; this is used when training a CART, and is an important concept that you need to understand.
Now work through the videos in this module. For the worked example, you can download and print the training data set, then cut the rows of the table into strips: each row is a data point. The last column is the predictee (i.e., the sound) and the first 7 columns are the predictors (i.e., the letters). Try to work through the example yourself in addition to watching the videos.