In my theory of typed Lojban, I distinguish between evaluated numbers and unevaluated numbers, the latter being a subcategory of text.
The types I believe in are:
* predicates (ka and du'u abstractions)
* events (nu abstractions)
* numbers and parsed text (evaluated numbers introduced generally with {li} as well as lu..li'u quotes)
* unparsed text (introduced with zo, lo'u...le'u, zoi, and me'o)
* concrete sumti
The problem I have with this formalization is that I can't seem to pin any good invariants on the class of concrete sumti. Even things which arguably do not have any spatial extent are included in my class of concrete sumti. Therefore, sadly, the defining characteristic of my concrete sumti is "is not any other kind of sumti"; it's a catch-all class.
I believe furthermore in a number of subclasses to each kind of thing, so really, these 5 major types above really are type classes more so than types, and certain expressions can be polymorphic in which instance of the type class they are.
.i mi'e la tsani mu'o