RAMSCAR LAB PAPERS LAB PEOPLE RESEARCH LIBRARIES

cognition language & learning lab nal explains... learning and the nature of symbolic thought department of psychology, stanford university

The Feature-Label-Order Effect In Symbolic Learning Introduction Symbolic communication is a defining human characteristic. Yet despite the benefits symbolic thought provides when it comes to organizing, communicating about, and manipulating the world, our understanding of symbols and symbolic knowledge is relatively poor. Centuries of pondering the nature of symbolic representation (in terms of concepts, categories or word meanings) have so far yielded more puzzles than answers. Our impoverished understanding of symbolic learning contrasts starkly with our understanding of other areas of learning, where computational models of behavior have been related to the neuroanatomical structures in which these learning mechanisms are realized. Here we present an analysis of symbolic learning - and in particular, word learning - in terms of error-driven learning, which is the basis of most formal learning models. This analysis reveals that word learning can take two potential forms: learning to predict a label from an object in the world, and learning to predict an object from a label. Because of the intrinsic characteristics of verbal labels, and the complexity of the categories to which labels are applied, the analysis predicts that significant differences in discrimination learning will occur depending on the temporal order in which objects and labels are encountered. In a learning task, when objects precede their labels, the various features of those objects compete for relevance. This allows for the sets of features that are most definitive of each label -and which discriminate the meanings of labels from one another - to be learned. On the other hand, when labels precede objects, this kind of competitive discrimination learning is fatally inhibited.* The results of a computational simulation and a study of adults learning complex artificial categories confirm these differences, as does a study of children learning color categories. In each of these studies a Feature-Label-Ordering (FLO) effect in learning is clearly evident; discrimination learning is facilitated when objects predict labels, but not when labels predict objects. Even when learners are provided with exactly the same information, manipulating the order in which objects and labels are encountered has a dramatic effect on learning. *This has potentially important implications for theories of REFERENCE.