The evolution of adjectival monotonicity

Abstract

A striking property of scalar adjectives is that they allow for inferences like the following: If John is tall, and Mary is taller than John, then Mary is tall. This inference can be made because if an individual falls in the extension of the adjective, then any individual that has the property to a greater degree also will. We call this universal of adjectival semantics monotonicity. In this paper, we present an evolutionary account of monotonicity and support it with three computational models. In the first model, we study which of the possible meanings of scalar adjectives evolve under a pressure for simplicity alone, and we observe degenerate meanings that are unlike natural language adjectives. In a second model, we combine the pressure for simplicity with a pressure for communicative accuracy. Under these pressures, mostly non-monotonic meanings evolve. In the third model, we equip the agents with pragmatic reasoning skills. In the third condition monotonic meanings prevail. We conclude that adjectival monotonicity is caused by a combined pressure for semantic simplicity and communicative accuracy, given human pragmatic skills.