Understanding complex idealized cognitive models by the use of a total overlay structure.

摘要

Lakoff (1987) provides general principles about how we reason with conceptual categories in his description of idealized cognitive models (ICMs). Part of his description involves some structuring principles for complex ICMs, such as cluster models and radial categories. However, not all ICMs fall neatly into these two groupings. Another useful method to understand complex ICMs is through the use of an overlay structure, similar to the manner in which transparencies are overlaid on a base diagram to provide a clearer way of understanding the base structure.; There are (at least) three types of overlay structures which can be used in the modeling of complex ICMs; total overlays, partial overlays, and overlaps. This research will focus on total overlays, which are structures in which each layer of the model exhaustively maps the base domain. The use of the total overlay structure is especially interesting, since we make use of it in understanding such cognitively fundamental domains as time and space, as well as in more abstract domains, such as computer science and monetary systems. This research investigates how we make use of structuring principles related to the total overlay in understanding a variety of complex ICMs by examining the components of each of the ICMs, as well as the heuristics we employ in selecting the appropriate term from within these models. Finally, we will explore how these structures and heuristics are useful within the context of natural language processing systems.; One problem especially relevant to this research is on the response generation capability of a natural language processing (NLP) system; specifically, how we choose the appropriate response in a discourse. Consider, for example, the kind of question which even schoolchildren can answer with little difficulty, such as "How old are you?" or "Where are you from?" In most cases, people are able to select a response using a term from the appropriate layer of the model; this research explores the general rules we use in making this selection, as well as how we might implement these techniques in an NLP system.