Towards a Substantively Biased Theory of Learning

0. Introduction There has recently been much debate concerning the nature of representations of phonological objects, particularly with regard to the role of abstract phonological constructs such as features and natural classes. This debate has extended to include differing notions about how phonological knowledge is learned and represented in the mind. The goal of this paper is to provide evidence for the psychological reality of abstract representations, including distinctive features. We argue for a theory of substantively biased learning in which learners are biased towards phonological patterns that are abstract, general, and correlate to cross-linguistic tendencies. We review how the poverty of the stimulus paradigm for artificial grammar learning can be used to reveal that knowledge of vowel feature dependencies (height and tenseness) can lead to substantive biases about the level of generality in rule formation. This paper provides further data that substantive biases are abstract, and derived from constraints on the inventory of phonological segments. This paper proceeds as follows. Section 1 provides background on the nature of vowel harmony and the type of substantive biases that are likely to form in vowel harmony systems. Section 2 discusses substantive biases in greater detail, including how it is possible to tap into the adult learner’s biases. Section 3 describes the (Artificial Grammar) POVERTY OF THE STIMULUS PARADIGM, and the results of an experiment on height harmony. Section 4 presents and discusses the results of an experiment testing for generalization to novel suffixes in vowel harmony. Section 5 concludes with prospects for future experiments.