Program of the Tenth Annual Conference of the Cognitive Science Society: 17-19 August 1988, Montreal, Quebec, Canada

Program of the Tenth Annual Conference of the Cognitive Science Society: 17-19 August 1988, Montreal, Quebec, Canada

Program of the Tenth Annual Conference of the Cognitive Science Society: 17-19 August 1988, Montreal, Quebec, Canada

Program of the Tenth Annual Conference of the Cognitive Science Society: 17-19 August 1988, Montreal, Quebec, Canada

Excerpt

One reason for scepticism about connectionist models that use distributed representations is that there are currently no convincing demonstrations of how these models can represent complex, articulated structures. Drew McDermott (personal communication) has suggested that the approach would be far more convincing if it could come up with a sensible scheme for representing the meaning of a sentence such as: "She seems to be more at ease with her fellow students than with me, her adviser." This meaning is clearly composed of several major constituents with relationships between them, and each major constituent has its own, complex, internal structure. A representational scheme for dealing with meanings of this complexity must, at the very least, specify how the meanings of whole expressions are related to the meanings of their constituents and how it is possible, in some sense, to have the whole meaning in mind at once.

The example given above is typical of examples from many different domains. It appears that whenever people have to deal with complexity they impose part-whole hierarchies in which objects at one level are composed of inter-related objects at the next level down. In representing a visual scene or an everday plan or the structure of a sentence we use hierarchical structures of this kind. The main issue addressed in this paper is how to represent complex part-whole hierarchies in a connectionist network. Three different methods are described.

Symbols and the conventional implementation of hierarchical structures

It will be helpful to begin by reviewing the standard way of implementing hierarchical data-structures in a conventional digital computer. There are obviously many minor variations, but a suitable paradigm example is the kind of record structure that is found in languages like Pascal (but without the type constraints).

Each instance of a record is composed of a predetermined set of fields (sometimes called "slots" or "roles") . . .

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.