No abstract

No abstract

No abstract

1. The knowledge to be elicited must be present in the input information stream -nonsensical data can give no (or nonsensical) results.2. It must be known to the system, how to extract the desired piece of knowledge. In order to be able to learn effectively, the system must possess a large amount of procedural knowledge to allow the extraction of the desired information (see [2] for more in depth analysis of this problem).3. If the goal is to obtain a detailed description of the process perceived about a scene, the system must have the appropriate conceptual structures necessary to express the details of the perceived events. Describing (the scene) can be seen as classifying -the more complex (deep and wide) the prototype net (or tree), the more detailed the resulting description. This remark is also valid for systems able to learn about new prototypes -such systems must have a "place" for these prototype descriptions, i.e., the prototype net must also be deep and/or wide with, possibly, empty nodes to be filled during the learning process.4. Multilevel organization of knowledge processing modules allows the simplification of the resulting procedures -instead of one, huge program, a chain of smaller and simpler ones may be obtained. An example of such organization can be found in [1].

This paper presents an overview of research in progress in which the principal aim is the achievement of more natural and expressive modes of online communication with complexly structured data bases.A natural-language compiler has been constructed that accepts sentences in a userextendable English subset, produces surface and deep-structure syntactic analyses, and uses a network of concepts to construct semantic interpretations formalized as computable procedures. The procedures are evaluated by a data management system that updates, modifies, and searches data bases that can be formalized as finite models of states of affairs.The system has been designed and programmed to handle large vocabularies and large collections of facts efficiently.Plans for extending the research vehicle to interface with a deductive inference component and a voice inputoutput effort are briefly described.