The Dictionary based break iterator implementation make use of the type of the break iterator - word, line, etc. - to decide whether the dictionary should be invoked.
The main (non-dictionary) break iterator character classification data also includes a dictionary bit, which the rule based code uses to decide when to invoke the dictionary code.
These two mechanisms are redundant.
I propose removing the explicit use of the break type, and rely solely on the dictionary bit in the rule data. This is needed in any event. Also, the 'type' of a break iterator is not known for those created directly from rules, and may not even correspond to a type that we know about, but the rule data must define the set of dictionary characters if dictionary processing is wanted.