Using Lexical Semantics to Predict the Distributivity Potential of Verb Phrases in a Large Dataset

Abstract

Applied to a plural subject (“Alice and Bob”), some predicates are understood distributively (individually true of each member of the subject: “Alice and Bob smiled” conveys that Alice smiled and Bob smiled); some are understood nondistributively (true of the subject as a whole, but not each member individually: “Alice and Bob met”); and some can be understood in both ways (“Alice and Bob opened the window”: distributive if they each individually opened it, nondistributive if they opened it jointly). This paper tackles the open question of which predicates are understood in which way(s) and why: Which other predicates act like “smile”, like “meet”, or like “open the window”? Researchers would agree that a verb phrase's distributivity potential depends on world knowledge about the event that it describes. Making that truism predictive, this paper presents an experimental study providing evidence consistent with several large-scale, theoretically-motivated generalizations in this realm.