Effects of Task and Category Membership on Representation Stability‪.‬

According to Poitrenaud, Richard and Tijus (2005), "the role and nature of categories, as well as the process of categorization itself, are all highly controversial" (p. 151). The conceptual core of categories (i.e., the idea that concepts are stable because the foundation of their coherence is non-perceptual and reflects representations and beliefs about things of the world) is indeed questionable. From one perspective, conceptual representations are temporary constructs realized in a specific context and with a particular goal that is based on to the task at hand (Barsalou, 1993). In addition, representations are "assembled on-line from multiple knowledge sources and specific experience provided by the task context" (Poitrenaud et al., 2005, p. 154). Given this definition, are conceptual representations stable over time? In this study, we examined the withinsubject stability of representations in relation to the type of task (a sorting task versus a property-generation task) and to the nature of the categories (natural versus manufactured categories). Few authors have studied the temporal, within-subject stability of conceptual representations in relation to task. Bellezza (1984a, b and c) and Barsalou (1989) were the first authors to demonstrate such variability in the property-generation task. Bellezza (1984c) asked participants to produce several descriptive properties to qualify various concepts such as personal friends. The properties of a concept are the dimensions (e.g., height) and the features (e.g., tall) that characterize it. The participants in Bellezza's study repeated the task one week later and displayed low within-subject stability. Indeed, only 38% of the generated properties were consistent between the two sessions. These results are in line with the results of Barsalou, Spindler, Sewell, Ballato and Gendel (1987, cited in Barsalou, 1989) and Barsalou, Sewell and Ballato (1986, cited in Barsalou, 1989). Barsalou et al. (1987) showed that, regardless of the proposed task (the production of either ideal or defining properties for the categories), only one-third of the properties were common across all of the participants. Thus, the property-generation task seemed to be an important indicator of within-subject variability.