In generative grammar, the principles of binding theory are defined as constraints on reference that hold within sentences (e.g. Chomsky 1981), but not across speakers’ utterances. An example is Principle C, which constrains coreference between an r-expression and a pronoun within a sentence.

(1) *He₁ went to Chuckie₁’s house. (2) He₁ went to Chuckie₂’s house.

In (1), coreference is prohibited because the pronoun c-commands the r-expression. Does Principle C in effect when a referent is outside the utterance? This situation would occur in fragment sentences, as shown in (3) and (4), where A and B are speakers in a discourse:

(3) A: I know where he₁ went! (4) A: I know where he₁ went!

B: Me too! *To Chuckie₁’s house. B: Me too! To Chuckie₂’s house.

Adults know that in (3), ‘he’ and ‘Chuckie’ can not corefer. This paper investigates whether or not children know this fact, and the implications this knowledge has on the learnability problem.

From a theoretical perspective, one could ask whether or not the same mechanism underlies fragments and full sentences. Merchant (to appear) describes an ellipsis account, in which the fragment in (5) would be generated from a full sentence as follows: the fragment answer is obligatorily reconstructed as an IP. The PP fragment is moved to a focus position, and the remaining IP constituent is ‘sluiced’, or hidden in the PF component. On this account, Principle C applies within the sentence before sluicing.

(5) A: Where did he₁ send the letter?

B: *[he₁ sent the letter to Chuckie’s₂ house]

*[ FocP to Chuckie’s₂ house]t k[IP he₁ sent the letter tk]

*[ FocP to Chuckie’s₂ house]t k[IP he sent the letter tk]

The ellipsis theory invites a prediction about children’s acquisition of coreference relations in examples (1-4). If coreference relations across speakers in discourses like (3) are constrained by Principle C, then the expectation is that children should demonstrate mastery of the facts in (1-4) as early as they can be feasibly tested. Children should disallow coreference between the pronoun and the r-expression in both cases.

Non-generative accounts of language development make different predictions. A recent account of language development, the ‘constructivist account’, claims that children acquire the grammar ‘construction’ by ‘construction’, with the order of development determined by the frequency of the particular construction in the input (e.g., Goldberg 2003, Tomasello 2003). Since constructions are learned individually, no deep generalization (such as Principle C) can underlie a number of syntactic constructions. On this account, there is no reason to expect children to treat (2) and (4) in the same way. Therefore, we would expect to see young children who have mastered coreference relations in full sentences, but not in fragments (or vice versa), since these are different constructions.

The learnability problem is at issue: since a child will never hear a sentence where an r-expression is coreferential with a pronoun over a fragment as in (3), how can the child learn that coreference in these situations is disallowed? Considering impoverished input children receive with regard to fragment sentences, it does not seem that these binding relations could be learned. The alternative to learning rules is that pronoun relations are governed by an innate constraint that is part of Universal Grammar. If children disallow coreference of the pronoun and r-expression in utterances like (1) and (3), which are different constructions, this would be evidence that there is an underlying constraint (Principle C). UG predicts that young children will disallow the coreference interpretation when Principle C is violated in both full sentences and fragments.

Experiment: Using a Truth Value Judgment Task (Crain and Thornton 1998), 20 children, ages 3;3 to 5;5 were tested in two experimental sessions. In a TVJT, an experimenter acts out a story with some toys, observed by the child and a puppet. When the story is complete, the puppet says what he thinks happened. The child’s task is to decide whether the puppet’s description was true or false. In both sessions, children were presented with a scenario and an utterance which was false on the non-coreference reading, and true on the coreference reading. If Principle C is constraining interpretations, then they will reject the sentence, since the coreference reading is not available. Session one had two puppets and the utterance was split across two speakers, as shown in (4), so that the pronoun and r-expression were divided into separate sentence fragments. Session two had one puppet which said the full sentence, as in (2). Sentences like (6) and (7) were used as controls to verify that children were not creating ‘positional’ rules, such as ‘an r-expression can not be coreferential with a pronoun on its left’, and that they do accept coreference as a possibility.

(6) His₁ mother bought Chuckie₁’s present

(7) Chuckie₁ threw his_1/2 cup

Results: The experimental findings show that (a) children treat both fragments and full sentences in the same way, and (b) that children and adults pattern alike. Children rejected coreference across speakers in discourses like (3) 89% of the time, compared with (1) 91% of the time. A group of 14 adult controls also rejected discourses like (3) 93% of the time. The differences between these data are not statistically significant, leading to a conclusion that children, as young as 3;3, hold the same binding conditions over a discourse as adults, and also the same conditions as in full sentences. Acceptance rates for the control conditions are shown in the table below. The results of the control conditions confirm that children do accept coreference and that children are not using positional rules for coreference relations. Considering the poverty of the input, and the young age at which children can compute these binding relations, these results support the idea that children access an underlying condition on binding to determine coreference, as opposed to learning construction by construction.

PERCENT ACCEPTED	Children – Fragment	Children- Non-Fragment	Adult - Fragment
Target	11%	9%	7%
(4)	100%	80%	71%
(5)	75%	88%	100%

Additionally, these data are in line with what we would expect from the ellipsis theory of fragments. Analyzing fragments in such a way explains how it is possible that children can apply Principle C (an innate constraint) to a construction (such as fragments) they may have never heard before.

These results strongly suggest that Principle C is the underlying constraint on pronoun reference in fragment sentences. Due to the early age at which children reject coreference violations, it appears that this constraint is innate, and could not have been learned due to the poverty of the input and lack of negative evidence.

Chomsky, N. (1981) LGB; Crain, S. and Thornton, R. (1998) Investigations in Universal Grammar; Goldberg, A.E. (2003) Trends in Cognitive Science; Merchant, J. (to appear) Linguistics and Philosophy; Tomasello, M. (2003) Constructing a Language.