Conjunctions and knowledge acquisition from text
Date
1990
Authors
Proctor, Laura Jane
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
A simple model for interpreting subordinating conjunctions is presented and applied to the task of automatic knowledge acquisition from written documents. Knowledge acquisition involves identifying features of the written document which correspond to the basic units of the underlying representation. The relationships between basic units then establishes an organizational component of interpretation. Subordinating conjunctions are not only indicators of structural (syntactic) form but also organize the text's interpretation. The organizational component of a text's interpretation is complex and may involve a number of different levels such as temporal sequence, physical consequence or cause, physical location, and logical contingency. In the application presented here, only the logical contingency level is addressed. However, the basic model may be applied to any other level of organization.
The basic units of the representation used are called objects and correspond to clauses or clause complexes. The internal organization of these basic units is not important to the model as it is presented here. What is important, however, are the relationships among objects. These relationships are represented by directed links which capture the ordering of objects in terms of logical contingency among them. Such links can be indicated by subordinating conjunctions which can be classified into three groups, based on the ordering they signal between their subordinate and main clauses. The three groups are defined as pre-ordered, post-ordered, and parallel-ordered, corresponding to the position of the subordinate clause relative to the main clause in the representation. The initial research reported here has successfully generated a knowledge base from a sample Bylaw of the City of Victoria. Such a preliminary knowledge base is intended to provide a basic knowledge schema for further use by Expert Systems developers. This knowledge schema efficiently extracts the important relationships between objects, and at the same time provides the basis for a hypertext interface to the on-line document.