Abstract The study of cross-linguistic word-class variability and its limits needs to be based on clear comparative concepts because languages have different grammatical structures. This chapter shows that the notions of (i) function indicators and (ii) semantic root classes give us a way of formulating universals and of distinguishing cross-linguistic macro-classes in a rigorous way. The root classes are action, property, and object; function indicators are markers that signal an unusual propositional act function (e.g. a copula signals predicative use of an object root, and a relativizer signals modifying use of an action root). The chapter argues for distinguishing clearly between comparison and language-particular analysis, because analysis must be based on language-particular constructions and is thus a rather different enterprise from the search for language universals.