In lieu of an abstract, here is a brief excerpt of the content:

  • The Structure of Children's Linguistic Knowledge
  • Andrea Gualmini and Stephen Crain

A recurring theme in arguments from the poverty of the stimulus concerns children's knowledge of linguistic structure. Nativists point to the extensive gap between what children know and what they could have learned from experience, even given optimistic assumptions about children's abilities to extract information from the environment, and to form generalizations. This squib looks at children's knowledge of linguistic structures that involve the semantic property of downward entailment, allowing us to address a recent critique of children's knowledge of structure offered by Lewis and Elman (2002).

1 Structure Dependence and Poverty of the Stimulus

An example of structure-dependent linguistic principles deals with question formation. This phenomenon was originally described by Chomsky (1971), who questioned the extent to which the primary linguistic data could lead children to form the correct generalizations relating declarative sentences and their yes/no question counterparts (see also Chomsky 1980 and discussion in Piattelli-Palmarini 1980). [End Page 463] Consider the declarative sentences in (1) and the corresponding yes/no questions in (2).

(1)

  1. a. The boy who is sitting on the rug is hungry.

  2. b. The boy is sitting on the rug that is being vacuumed by his mother.

(2)

  1. a. Is the boy who is sitting on the rug ________ hungry?

  2. b. Is the boy ________ sitting on the rug that is being vacuumed by his mother?

Even these simple examples suffice to illustrate that structure-blind operations will have a difficult time accounting for question formation in English.

The role of structure dependence in question formation by children was investigated by Crain and Nakayama (1987), who used an elicited-production task in several experiments to evoke yes/no questions from young children. In one experimental trial, children were shown a picture depicting one man who was beating a donkey and a second man who was not beating a donkey. Then children were invited to pose a question to a puppet, Jabba the Hutt. The input to children contained two auxiliary verbs: for example, Ask Jabba if the man who is beating the donkey is mean. The experiment was designed to see whether children would ask adultlike questions like (3) or whether, instead, they would ask incorrect questions like (4), where the first auxiliary verb was "moved" from its position in the request they had just heard.

(3) Is the man who is beating the donkey mean?

(4) *Is the man who beating the donkey is mean?

Observing that no child ever produced questions like (4), Crain and Nakayama (1987) concluded that children entertain structure-dependent operations in forming yes/no questions, not structure-independent ones, despite the lack of overt evidence in the input.

This conclusion was recently challenged on two counts. Pullum and Scholz (2002) argue that Crain and Nakayama (1987) make the unwarranted assumption that children cannot rely on the input in order to learn the correct question forms. They contend that there is sufficient available evidence to support learning (but see Crain and Pietroski 2002, Lasnik and Uriagereka 2002, Yang 2003). Taking a different tack, Lewis and Elman (2002) constructed a simple recurrent network to model question formation in English. The network was trained in a three-stage process, such that the degree of complexity of the input was increased at each stage. The network was presented with yes/no questions with a single auxiliary verb (e.g., Is the big dog in the car scary?), but not ones with two auxiliaries (e.g., Is the farmer who is beating the donkey mean?). Lewis and Elman (2002:364) maintain that "the network does not make the predictions corresponding to the ungrammatical [question, such as (4)]—i.e., the network does not predict [a gerund] following 'who'." In other words, the network consistently [End Page 464] predicts that a substring like (5) should be followed by an auxiliary verb, despite the absence of such strings in the training sessions.

(5) Is the boy who _______

Lewis and Elman (2002) observe, further, that if the network is presented with a substring with an auxiliary verb following who, as in (6), it predicts the occurrence of...

pdf

Share