@incollection{
    BergSchleimerSarelaHonkela05,
    editor = "Magnani, Lorenzo and Dossena, Riccardo",
    author = {Berg, Mikko and Schleimer, Jan-Henrik and S\"arel\"a, Jaakko and Honkela, Timo},
    publisher = "College Publications",
    title = "Category learning by formation of regions in conceptual spaces",
    url = "http://www.cis.hut.fi/tho/publications/berg_schleimer_sarela_honkela.pdf",
    booktitle = "Computation, Philosophy, and Cognition",
    address = "London, UK",
    pages = "381-396",
    year = "2005",
    abstract = "In this paper, we discuss the issue of conceptualization. The traditional view is that concepts are essentially linguistic. Recently, Gardenfors has proposed a contradicting view stating that the concepts get associated to language terms, but essentially belong into other domain called conceptual spaces defined by quality dimensions. These dimensions form meaningful representations of the concept domains in hand and they should be formable by mappings from the sensory input and possibly from other more basic quality dimensions as well. In the space spanned by the quality dimensions, natural concepts form convex regions. The borders of these regions can be hard or soft and can vary according to the context. In the present work, we have decided to code the regions by prototypes, so that instances closest to a particular prototype in the conceptual space form a region. In other words, the regions are defined by the Voronoi tessallations of the prototypes, which later define hard-bordered regions. In the case of soft borders, the prototypes can consist of probabilistic density functions defining graded membership function for each point in the conceptual space. This paper explores the idea of quality dimensions by trying to realize contextual categorization in such a domain. That is, trying to form prototypes and regions. As addition, the connections to the lower, connectionist level and to the higher, symbolic level are discussed briefly."
}