Raising, Control and the Subset Principle

One part of the answer to the Logical Problem of Language Acquisition is that children come to the task of language learning equipped with Universal Grammar: this allows them to entertain a limited set of hypotheses about what Language could possibly be like, so they don’t have to consider all the possibilities in the universe. Another part of the answer, typically assumed, is that children use a learning strategy that follows the Subset Principle. The idea behind this principle is that the learner should assume the most restrictive grammar, and then change the grammar when faced with positive evidence in the input that the target grammar is different. If a learner assumed a grammar that was less restrictive, or larger than the actual target grammar, no evidence in the input would force the learner to change the grammar, due to the absence of negative evidence. Although the subset principle has been challenged in the formal learnability theory literature (Angluin , 1980), it is still commonly assumed by researchers studying children’s language development and is thought to play a dominant role in human language acquisition. There are good reasons for this. From a theoretical standpoint, we assume there is no negative evidence in the input that could help a learner to restrict a grammar that is too large. In other words, we cannot require the child to be explicitly corrected, in order to learn the target grammar (Chomsky , 1959). From an empirical standpoint, many of the patterns we find in children’s early language suggest a highly restrictive grammatical system, and the subset principle has been invoked to explain particular phenomena in acquisition, for example in Manzini and Wexler (1987)’s account of the acquisition of the Binding Theory and Hyams (1986)’s account of setting the Null Subject parameter. In this paper, I will look at a domain of language in which the subset principle falls short: namely, the acquisition of the distinction between raising verbs and control verbs. First I will review the syntactic properties of raising and control constructions and I will lay out what the subset principle would lead us to assume about how a language learner should distinguish these verb classes. Then I will show how the predictions of this kind of approach fail on both logical and empirical grounds. Finally, I will suggest an alternative learning strategy based on multiple cues.