Frozen Chunks and Generalized Representations: The Case of the English Dative Alternation

E. Conwell, T. O’Donnell, J. Snedeker

BUCLD

Abstract

A fundamental and celebrated property of language is its productivity: using language we can understand and express an unbounded number of novel thoughts. However, it has long been acknowledged that this productivity presents a fundamental problem for the learner (Baker, 1979). How is the learner to know which of the many possible generalizations consistent with the data are actually productive in their own language? How can learners strike the balance between regularity and idiosyncrasy? Two distinct theoretical approaches to this problem make different assumptions about the starting state of the learner and predict different developmental trajectories. One approach assumes that a rich and abstract linguistic system is available very early in development, perhaps even at birth. The learner’s task is to tailor this system to the particular language being learned. We will call this the ―early generativity‖ approach. There are many different versions of early generativity theories (for examples, see Chomsky, 1981; Pinker, 1989; Fisher, 2002; Snyder, 2007). However, these approaches are unified by the assumption that abstractions of approximately the right grain-such as ―noun,‖ ―verb,‖ ―subject‖ and ―object‖-are available to the learner from an early age and that the role of linguistic input is simply to indicate how these abstractions are manifested in the particular language being learned. The fundamental learning problem for theories that posit early generativity is how the child constrains the potential productivity in the system. Such theories provide a starting state that accounts for the productive linguistic processes in all of the world’s languages, but must also provide a way for the learner to determine which particular processes are productive in her own language. Another class of approaches assumes that the learner begins with no linguistic abstractions but instead gradually forms these generalizations from the linguistic input using domain-general cognitive and social abilities. These theories fall under the umbrella of ―usage-based‖ approaches (e.g., Tomasello, 2003; Lieven, Behrens, Speares & Tomasello, 2003; Goldberg, 2003; Braine, 1963). Such approaches typically emphasize the gradual emergence of