Relative contributions of kind- anddomain-level concepts to expectationsconcerning unfamiliar exemplarsDevelopmental change and domain differencesPascal Boyera,*, Nathalie Bedoinb, Sandrine HonoreÂcaCentre National de la Recherche Scientifique, Lyon, FrancebLaboratoire d'Etude des MeÂcanismes Cognitifs, UniversiteÂLumieÁre, Lyon, FrancecDoctoral Programme, DeÂpartement de Psychologie Cognitive, UniversiteÂLumieÁre, Lyon, FranceReceived 1 July 1999; accepted 1 July 2000AbstractTwo inferential routes allow children to produce expectations about new instances ofontological categories like ``animal'' and ``artefact.'' One is to generalise information from a``look-up table'' of familiar kind-concepts. The other one is to use independent expectationsat the level of ontological domains. Our experiment pits these two sources of informationagainst each other, using a sentence-judgement task associating properties with images offamiliar and unfamiliar artefacts and animals. ``Strange'' properties are compatible with theontological concept, but not encountered in any familiar kind. A look-up strategy wouldlead children to reject them and an independent expectation strategy to accept them. In bothdomains, we find a difference in reaction to strange properties associated with familiar vs.unfamiliar items, which shows that even young children do use independent domain-levelinformation. We also found a U-shaped curve in propensity to use such abstract information.In addition, animal categories are the object of much more definite domain-levelexpectations, which supports the notion that the animal domain is more causally integratedthan the artefact domain. D 2001 Elsevier Science Inc. All rights reserved.Keywords: Concepts; Domain-specificity; Category-specificity; Animal; Artefact0885-2014/00/$ ± see front matter D 2001 Elsevier Science Inc. All rights reserved.PII: S 0885-2014(00)00022-8* Corresponding author. Washington University-Campus Box 1114, One, Brookings Drive, St.Louis, MO 63130, USA. Tel.: +1-314-935-7280; fax: +1-314-935-8535.E-mail address: [email protected] (P. Boyer).Cognitive Development 15 (2000) 457 ± 4791. IntroductionChildren between 3 and 10 develop a large number of kind-concepts foranimals and artefacts. This is clear in the lexical explosion that occurs at thebeginning of that developmental period (Clark, 1993) and in children's perfor-mance in categorisation tasks (Gopnik & Meltzoff, 1987). Most of these conceptscorrespond roughly to the ``basic'' level of categorisation where between-classessimilarity is reduced and within-classes similarity is maximal (Dickinson, 1988;Hall, 1993; Waxman, 1991). These are, for instance, concepts of ``giraffe,''``cow,'' ``car,'' or ``telephone.'' Count-nouns are generally associated with suchkind-concepts rather than with superordinates, subordinates, part-, or property-concepts (Waxman & Markow, 1995).Children also develop, from infancy, more abstract concepts for superordinatesof the basic concepts, e.g. ``animal,'' ``artefact,'' or ``person'' (Mandler & Bauer,1989). These domain-concepts do not generally correspond to early-acquiredlexical entries. Nor are they connected to maximally integrated features. How-ever, they constrain inferences, for instance, about both familiar and unfamiliaranimal and artefact categories.These two levels of conceptual information (henceforth ``kind-concepts'' atthe basic or near-basic level and ``domain-concepts'' at the ontological level) arealso found in adult semantic memory. They store different kinds of information.Various behavioural measures demonstrate the difference between conceptualassociations retrieved at domain level and those available at the ``entry level,'' i.e.kind-concept level (Chumbley, 1986; Jolicoeur, Gluck, & Kosslyn, 1984;Kosslyn, Alpert, & Thompson, 1995). We now have some evidence concerningboth the sources of domain-level inferences (the database that is activated) andthe processes engaged (inductive generalisation, instance-based generalisation,analogy, theory-like principles). The aim of the present article is to provide abetter description of the developmental course of inferences from familiar tounfamiliar stimuli in the domains of animal and artefact concepts, highlightingdomain and age differences in sources and processes of inference.It is still not entirely clear how domain-concepts are constructed by childrenand how they connect with kind-concepts at or near the basic level. To someextent, an accumulation of features represented in kind-concepts certainlysupports inductive generalisations in domain-concepts (see e.g. Eimas, 1994for the ``animal'' concept). This is the most plausible route whereby childrenlearn, for instance, that most tools are in fact made of hard materials or that mostmammals' bodies are warm. However, at a deeper level, there is considerableevidence that information flows in the other direction as well. First, domain-concepts produce expectations about instances of kind-concepts. Preschoolers,for instance, expect animals to move around of their own accord, but notartefacts, even if the latter look like animals and even though the child cannotexplain why she has these divergent intuitions (Massey & Gelman, 1988).Second, there is evidence that in many cases these expectations are systematic,P. Boyer et al. / Cognitive Development 15 (2000) 457±479458that is, interconnected rather than merely juxtaposed. Preschoolers expect animalsto move around because they are ``driven'' by internal states and to be in thesestates because of external circumstances. Third, expectations from domain-concepts appear in many cases to be theoretical, in the sense that they go farbeyond obvious features of known instances (Gopnik & Meltzoff, 1987). Forinstance, children expect the ``insides'' of animals to be more crucial in producingbehaviour than their ``outsides,'' although the latter are of course critical in theidentification of instances (Gelman & Wellman, 1991).These different phenomena suggest that domain-concepts have direct effects onthe construction of kind-concepts. Kind-concepts may be initially triggered bynonprincipled detection of similar features in different instances, notably by shape(Landau, Smith, & Jones, 1988; Quinn & Eimas, 1997). However, efficientcategorisation would require constraints on similarity (Medin, Goldstone, &Gentner, 1993; Rips, 1989) and such constraints are indeed found in children(Keil, 1994). Identification of