In defining use cases, heuristically thinking, it is probably wise to avoid blue sky concepts like Understand Natural Language,  instead (perhaps) thinking in terms of specific kinds of goals which drive specific kinds of use cases.  In many instances, it's reasonable to expect that some use cases will satisfy more than one goal.

Many of the use cases are highly related, as if there might be some taxonomic structure that could be imposed upon them, or sets of relations which wire them together. An overall activity is that of understanding natural language, our blue sky notion. Thus, we introduce Use Case 0, Machine Reading to serve as an umbrella for parsing, understanding, and related activities.