If you do this, then your overall description has to include separate models for categorical and continuous statistical observations. I personally think this is fine, but I'm curious what others have done in this scenario. There is discrete and continuous data to simultaneously consider. 5/n

For instance, consider that you are looking at glottalization in English coda stops. Maybe it occurs with [k] less often than with [t]. Glottalization is feature "X." It's either present or absent in addition to being measureable when it is found. 7/n