The Influence of Co-Occurrence and Inheritance Information on Children's Inductive Generalization

The Influence of Co-Occurrence and Inheritance Information on Children’s Inductive Generalization Karrie E. Godwin (kegodwin@andrew.cmu.edu) Carnegie Mellon University, Department of Psychology 5000 Forbes Avenue, Pittsburgh, PA 15213 USA Anna V. Fisher (fisher49@andrew.cmu.edu) Carnegie Mellon University, Department of Psychology 5000 Forbes Avenue, Pittsburgh, PA 15213 USA Bryan J. Matlen (bmatlen@andrew.cmu.edu) Carnegie Mellon University, Department of Psychology 5000 Forbes Avenue, Pittsburgh, PA 15213 USA Abstract Prior research suggests young children understand that labels serve as category markers and that they can utilize this information to perform category-based induction with both identical and semantically-similar labels (Gelman & Markman, 1986). Recent research suggests that children’s ability to perform category-based induction is limited to a small subset of semantically-similar labels which co-occur in child-directed speech (Fisher, 2010; Fisher, Matlen, & Godwin, in press). However, most of the co-occurring labels used in prior research are not only semantically-similar but they also refer to baby-parent relationships (e.g., puppy- dog). Thus, children may be able to perform induction with these particular label-pairs, because they contain kinship information rather than because they co-occur. The present study aims to disentangle whether young children’s induction performance is driven by kinship information or co-occurrence probability. Results indicate that 4-year-olds’ (but not 5-year-olds, 7-year-olds, or adults) induction performance was influenced by co-occurrence probability; kinship information was found to be insufficient to promote young children’s induction performance. Keywords: Categories. Labels. Induction. Cognitive Development. Introduction It has been suggested that even young children understand that labels denote object categories, and that children rely on this information to make inductive inferences (Welder & Graham, 2001, Gelman 1988; Gelman & Coley, 1990; Gelman & Markman, 1986; Jaswal, 2004). The strongest evidence in support of this argument comes from a study demonstrating that preschool-age children make inferences based on semantically-similar labels (to be referred to as synonyms henceforth for brevity) (Gelman & Markman, 1986). In this experiment children were first presented with a triad of objects and provided with respective labels. For example, children could be presented with a rabbit (Target item), a squirrel (Test item), and another rabbit (Test item) that looked dissimilar from the target. Children were told about the properties of the test items (e.g., that the rabbit ate grass and the squirrel ate bugs). Then children were asked to generalize one of these properties to the target item. Importantly, similarity in category membership was conveyed either by identical labels (e.g., rabbit-rabbit) or synonymous labels (e.g., bunny-rabbit). The results indicated that the rate of category-based inferences was above chance in both conditions (i.e., 67% with identical labels and 63% with synonymous labels). However, more recent findings suggest children’s ability to make inferences using synonyms is limited to a small set of words that not only share meaning but also co- occur in child-directed speech according to the CHILDES database (MacWhinney, 2000). In particular, Fisher, Matlen, and Godwin (in press) found that most 4-year-old children perform category-based inferences with synonyms that are likely to co-occur in child-directed speech (e.g., bunny- rabbit, puppy-dog); however, these same children are unlikely to make category-based inferences with synonyms that do not co-occur (e.g. alligator-crocodile, rock-stone). Importantly, children in this study exhibited near-ceiling accuracy in a task similar to the Peabody Picture Vocabulary Test (Dunn & Dunn, 1997) with both co- occurring and non-co-occurring synonyms (99% correct in both conditions). One explanation for this finding is that children’s inferences are based on co-occurrence probability rather than semantic similarity (Fisher, 2010; Fisher, Matlen, & Godwin, in press). Co-occurrence plays an important role in the formation of lexical associations (Brown & Berko, 1960; McKoon & Ratcliff, 1992) which may facilitate inductive generalization via priming. For example, when children are asked whether a “bunny” shares a property with a “rabbit” or a “squirrel”, children’s responses may be influenced by lexical priming (i.e., “bunny” priming “rabbit” but not “squirrel”) rather than category-based reasoning. In the English language we have only been able to identify a few semantically-similar labels that are not only familiar to preschool age children but also co-occur in child- directed speech. Incidentally, these words can be construed