VERB LEARNING

How do children learn to talk about the objects and events around them? The common sense answer to this question is that they look around the world as they listen to language, and they "simply" notice which words go with which aspects of the world. This commonsense solution is bound to fail, however, because the world provides many interpretations for the words and sentences that the child hears. This "abundance of richness" must be filtered through the child's mental apparatus, which includes his or her non-linguistic knowledge of the structure of objects, motions, and events; and his knowledge of how the linguistic system might be structured, and how the two systems might fit together. Our studies of verb learning (like those of object naming and spatial language) are aimed at discovering how children solve the problem of indeterminacy, or how they come to know the conceptual, semantic, and syntactic structure of words and sentences.

A. Motion events: Sources and goals

Motion events (such as a ball rolling out of a can down to the bottom of a hill) are lingustically expressed in English by a structured set of components: The figure (noun phrase/a ball), motion and manner (verb/ roll), two paths (prepositions/ out of, down to), a source (noun phrase/ a can) and a goal (noun phrase/ bottom of a hill). Each of these components is expressed in production by very young children-- by around age 2 or 3. But when young children are asked to describe motion events in sentences, they often show selective preservation of some elements, and omission of others. In several studies, we found that children as young as 3 years old tend to preserve figures, motions, goals, and their paths, but omit sources and their paths. For example, children might say "The ball rolled to the bottom of the hill", without including the source expression (out of the can; Landau & Zukowski, 2003). This pattern occurs in young children and adults, and over a wide variety of event types, including motion events, events of transfer, attachment/detachment, and change of state (Lakusta & Landau, 2005).

What is the foundation for this distinctive pattern, in which goals are expressed, but sources are dropped? We are now exploring the possibility that this asymmetry may stem from a basic property of event representation which characterizes both non-linguistic and linguistic representation of events. In her dissertation, Lakusta is currently examining this profile in pre-linguistic infants, using a preferential looking paradigm; and in preschoolers and adults, using a change detection paradigm. She is asking whether the source/goal asymmetry has reflexes in non-linguistic representation of events, and if so, what factors modulate this asymmetry.

B. Argument structure and syntactic expression: The problem of implicit objects

Most theories assume that a verb's argument structure projects to its syntax, with the verb's main arguments usually projecting to just one surface NP. This principle is central theories of verb learning, such as syntactic bootstrapping, which propose that the surface NPs accompanying a verb are assumed by the learner to represent its arguments (Landau & Gleitman, 1985; Gleitman, 1990). A particular puzzle arises, however, for verbs that take implicit objects. For example, the verb "eat" can either omit or preserve its external argument in the surface syntax (I ate lunch/ I ate), but a verb such as "want" must specify its external argument (I want a prize/ * I want). How does the learner come to know that a verb like 'eat' has two arguments, with one optionally specified, but a verb like 'want' has two arguments, both of which must be specified in the surface syntax? Tamara Nicol, Philip Resnik, and I are examining a property of the verb semantics which might provide part of the solution for the learner. Resnik (1996) demonstrated that verb selectivity correlates with object omissions in adult speech. Selectivity (a notion defined within information-theoretic framework by Resnik) captures the intuition that verbs like 'eat' tend to take few classes of nouns (hence are highly selective) but verbs like 'want' tend to take many classes of nouns (hence are relatively less selective). We are now investigating whether children are sensitive to the varying selectivity of different verbs, and if so, how they might be able to use this relation to predict which verbs allow implicit objects and which do not. Please see Nicol, Resnik and Landau (2003) and Tamara's web page for more information.