|
|
|
Chapitre d'ouvrage:
|
ID no. (ISBN etc.): 0-521-82417-6
|
Clé de citation BibTeX: Doumas2005
|
|
Doumas, L. A. A., & Hummel, J. E. (2005). Approaches to modeling human mental representations: What works, what doesn't, and why. In K. J. Holyoak & R. G. Morrison (Eds.), The Cambridge handbook of thinking and reasoning. (pp. p73–91). Cambridge, UK: Cambridge University Press.
Ajoutée par: Sterenn Audo 2008-01-15 10:45:54 Dernièrement modifiée par: Sterenn Audo 2008-01-15 10:53:15
|
|
|
|
|
|
|
|
Résumé
(from the create) A fundamental aspect of human intelligence is the ability to acquire and manipulate relational concepts. Relational thinking is so commonplace that it is easy to assume the psychological mechanisms underlying it are relatively simple. They are not. Along with language, the human capacity for relational thinking is the major factor distinguishing human cognition from the cognitive abilities of other animals. Central to understanding human relational thinking is understanding the nature of the mental representations underlying it: How does the mind represent relational ideas such as "if every element of set A is paired with a distinct element of set B; and there are still elements of B left over, then the cardinal number of B is greater than the cardinal number of A," or even simple relations such as "John loves Mary" or "the magazine is next to the phone"? Two properties of human relational representations jointly make this apparently simple question surprisingly difficult to answer (Hummel & Holyoak, 1997): As elaborated in the next sections, human relational representations are both symbolic and semantically rich. Although these properties are straightforward to account for in isolation, accounting for both together has proven much more challenging. An explanation of human mental representations--and the human cognitive architecture more broadly--must account both for our ability to represent the semantic content of relational roles and their fillers and for our ability to bind roles to their fillers dynamically without altering the representation of either. Traditional symbolic approaches to cognition capture the symbolic nature of human relational representations, but they fail to specify the semantic content of roles and their fillers. Traditional distributed connectionist approaches have the opposite strengths and weaknesses: They succeed in capturing the semantic content of the entities they represent but fail to provide any basis for binding those entities together into symbolic (i.e., relational) structures. This failure renders them incapable of relational generalization. Connectionist models that attempt to achieve symbolic competence by using tensor products and other forms of conjunctive coding as the sole basis for role-filler binding find themselves in a strange world in between the symbolic and connectionist approaches (i.e., on the implicit relations continuum) neither fully able to exploit the strengths of the connectionist approach nor fully able to exploit the strengths of the symbolic approach. Knowledge representations based on dynamic binding of distributed representations of relational roles and their fillers (of which LISAese is an example)--in combination with a localist representations of roles, fillers, role-filler bindings, and their composition into complete propositions--can simultaneously capture both the symbolic nature and semantic richness of human mental representations. The resulting representations are neurally plausible, semantically rich, flexible, and meaningfully symbolic. They provide the basis for a unified account of human memory storage and retrieval, analogical reasoning, and schema induction, including a natural account of both the strengths, limitations, and frailties of human relational reasoning. (PsycINFO Database Record (c) 2007 APA, all rights reserved)
Ajoutée par: Sterenn Audo Dernièrement modifiée par: Sterenn Audo
|
|
Informations supplémentaires en ligne :
|
|
|
|
|
|
|
|
|
|
|
Idées
pdf dispo
Ajoutée par: Sterenn Audo
|
|
|
