Factive predicates (like ‘-matters’, ‘discover-’, ‘realizes-’) take NPs that refer to facts, propositional predicates (like ‘-seems’, ‘believes-’, ‘-likely’) take NPs that refer to propositions, and eventive predicates (like ‘-occurs’, ‘-take place’, ‘-causes-’) take NPs that refer to events (broadly speaking, including states, processes, conditions, ect.). Logically speaking at least two out of the three categories (facts, propositions, and events) can be eliminated. So, if all three kinds of referents turn out to be required for natural language semantics, their postulation is (...) empirically significant since a priori logical considerations do not require all of them.Pronominalization evidence inter alia raises questions about the distinctness of facts, events, and propositions. Two proposals for resolving the pronominalization dilemmas are, first, that abstract elements exist which contain the genuine antecedents for the pronouns (co-reference remaining both syntactic and semantic) and second, that syntactic co-reference is simply distinct from semantic co-reference. The first proposal hardly works at all, since it requires the postulation of many abstract elements and associated unmotivated deletion (or insertion) rules. The second proposal works for all the examples considered. Prior to discussing the two proposals, I show how any two of the three categories can be logically eliminated, a demonstration which also produces some hypothetical abstract elements of use in discussing both proposals. I conclude with some brief remarks on reference versus coreference. (shrink)