|
Abstract:
|
[ENG] This paper proposes that the representation of the referent of discourse might be characterized as the construction and updating of a non-linguistic mental model. Mental models of the referent allow the reader to set up some implicit parameters such as temporal, spatial, and behavioral scales, magnitude of quantifiers, etc. The set up of these parameters is critical to appropriately understand and integrate new pieces of information from the incoming text. A second function of modelling is to compute the consequences of the events described by the text. There is psychological evidence demonstrating that comprehenders are very efficient in calculating on-Iine the consequences of described events or changes. However, it is difficult for a purely deductive processor (e.g., a declarative data base plus some rules) to compute appropriately the effects of change. This is the so called «frame problem». Such problem might be overcome by means of a mental model account. Mental models are sensitive to the implicit knowledge of world's regularities and, as a consequence, some courses of modeling are more likely than others. This implies a functional advantage: The number of potential models for a givem linguistic message can be drastically reduced. Traditionally the mental models framework has focused on the representation of spatial contents. However, I suggest that mental models involve also the representation of psycho-social and self traits. Comprehension, is a knowledge state that potentially involves an ecological, a psycho-social, and a self model. Ata given moment certain level of modelling is more prominent (foreground) than the others (background). Some computational principles are proposed in order to guide an eventual implementation of mental models. First, the building blocks that underlies models are primitives. Primitives are variable rather than discrete and fixed semantic values. Secondly, the building and updating of models demand cognitive resources, therefore mental models are sensitive to the structural limits of the system (e.g., working memory, attentional resources). Thirdly, the updating of mental models is cyclic rather than continuous, and it takes place at the end of the main text constituents (clauses, sentences and paragraphs). Fourthly, some models parameters are structurally isomorphic (some spatial relationships, the temporal structure of events, etc.). Finally, the computation of primitives in the model is based in a multiple constraints principle. |
Mental models and referential processing