Sentences in natural language are routinely interpreted as stronger than would be expected from the lexical meanings of the overt lexical items alone. This has led to the postulation of exhaustification (strengthening) mechanisms in pragmatics and semantics. Such exhaustivity effects have largely been discussed for logical vocabulary, focused expressions, and predicates forming entailment scales with other predicates. Relying on recent work on additive particles, I argue that exhaustivity is at play in a significantly broader array of meanings than previously appreciated: all predicates are exhaustified, in all sentences. That is, the intuited meanings of predicates in sentences are stronger than their lexical–conceptual meanings. I focus on 'taxonomic' predicates, which do not form entailment scales with other predicates. I make this case first and foremost based on apparently banal contradictions like This comedy is a tragedy or The white flag is green. While these contradictions are intuitively due to the meanings of the predicates, the interaction of these predicates with additive particles (This comedy is also a tragedy) and conjunction (This play is both a comedy and a tragedy) is argued to show that the predicates are underlyingly consistent. As such, the contradiction observed in the basic case must result from exhaustification.

In addition to demonstrating the existence of exhaustification in the meaning of taxonomic predicates, I also show that this exhaustification behaves in a hitherto undescribed way. The exhaustification of a given predicate is not only obligatory, but it is also obligatorily local to the predicate. Modelling exhaustification through an Exh(aust) operator, roughly equivalent to a covert only, predicates are claimed to 'control' Exh: they both require its presence and roughly dictate its syntactic locus. These constraints on Exh give its semantic output the flavour of lexical meaning. I argue that the locality requirement on Exh is best understood as it needing to be in the predicate's maximal projection, and I model this by postulating an Agree relation between derivational morphemes (n⁰, a⁰, etc.) and Exh.

For Exh to exhaustify predicates in a non-trivial way, predicates must come with alternatives; similarly to expressions like some or or, they bear alternatives even without being focused. I make two claims about alternatives. First, concerning the alternatives borne by predicates, I suggest as a first approximation that these are the sisters of the predicate in a given conceptual taxonomy. I then propose a notion of 'predicational jurisdiction'—the kind of information provided by a predicate—to suggest that predicates are alternatives iff they share a jurisdiction. For example, green and table are not interpreted as mutually exclusive (i.e., are not alternatives for controlled exhaustivity) because they contribute different kinds of information; but table and chair, comedy and tragedy, and green and white are alternatives because they share a jurisdiction. This both explains why taxonomic sisters are alternatives, and, as I will show, manages to capture a broader range of data. The second claim about alternatives pertains to how Exh and additive particles interact. One of the key datapoints motivating the view that taxonomic predicates undergo exhaustification is their interaction with additive particles. Building on work suggesting that additives serve to avoid unwanted exhaustivity effects, I suggest that additives are directly involved in pruning alternatives from the domain of Exh. They do not prevent exhaustification by removing Exh, but can weaken Exh by making it exclude fewer alternatives.

The claim that there is a systematic and principled mismatch between the lexical–conceptual meaning of taxonomic vocabulary items and the meaning intuited from these expressions in actual sentences challenges what appears to be a tacit consensus in linguistics, psychology, cognitive science, and philosophy. Work on concepts takes for granted that the nature of concepts can be researched from the meanings of predicates in natural-language sentences. This thesis shows that this is not straightforwardly the case, because grammar systematically interferes with the basic meanings of predicates.