Each categorical array has two parts: a list of category names, and a list of "codes" (indices) of the current contents of the array.
Taking a subset of a categorical array does not change the categories that are allocated. For example,
A = categorical({'dog', 'human'});
[A(2), A(1)]
The A(1) and A(2) each extract a single element of the categorical array, but doing so does not create a new categorical array in which the respective values are the only possible categories. If it did create a new categorical array restricted to the one category, then [A(2), A(1)] would have to synthesize a new categorical array with new code numbers and category order 'human', 'dog', but that is not what happens: A(1) is a categorical object with a single code but with category names left intact.
