How do children use syntax to learn verb meanings?

Previous work finds that children use the syntax of a sentence with a new verb in order to make a guess about its meaning. One hypothesis proposes that children do this by using the number of arguments in a sentence: they expect a sentence with two arguments (a subject and an object) to describe an event that they perceive as having exactly two participants, a sentence with one argument (just a subject) to describe an event that they perceive as having only one participant, and so on. Another hypothesis proposes that children instead expect subjects and objects to correspond to specific participant roles: for instance, they might expect the subject to name the “do-er” of an event and the object to name the thing being acted upon, and that there may also be additional participants that are not mentioned in the sentence. We tested these hypotheses with 20-month-olds, who are just starting to learn new verbs.

20-month-olds heard a new verb in a sentence with a subject and an object: The girl pimmed the truck. They heard this sentence as a description of a video in which a girl takes a truck from a boy. An earlier experiment established that infants view this event as having three participants: the girl, the truck, and the boy. We then tested what infants think pim means by showing them two new videos side-by-side: one in which the girl takes the truck from the boy again, and one in which the girl grabs the truck and moves it towards herself, without the boy. We asked them to find another instance of “pimming,” and measured which video they preferred to look at. If infants were expecting that a sentence with two arguments describes an event with exactly two participants, they should think that the 2-participant “grabbing” scene is a fine instance of “pimming.” But if they did not have this expectation, they should prefer to look at the 3-participant “taking” scene again. We found that they did prefer the “taking” scene, and did so to a greater extent that in a control condition. This argues against the first hypothesis: it does not appear that verb-learning children expect clause arguments to match event participants in number, and it suggests that they might instead be more flexibly linking specific types of clause arguments to specific participant roles.