Acquiring Knowledge From Encyclopedic Texts

Acquiring Knowledge From Encyclopedic Texts* Fernando Gomez Dept. of Computer Science Univ. of Central Florida Orlando, FL 32816, USA gomez~cs, ucf. edu Richard Hull Dept. of Computer Science Univ. of Central Florida Orlando, FL 32816, USA hull@as, ucf. edu Carlos Segami Math. and Comp. Sci. Dept. Barry University Miami Shores, FL 33161, USA segami~buvax, barry, edu Abstract A computational model for the acquisition of knowledge from encyclopedic texts is de- scribed. The model has been implemented in a program, called SNOWY, that reads unedited texts from The World Book En- cyclopedia, and acquires new concepts and conceptual relations about topics dealing with the dietary habits of animals, their classifications and habitats. The program is also able to answer an ample set of ques- tions about the knowledge that it has ac- quired. This paper describes the essential components of this model, namely seman- tic interpretation, inferences and represen- tation, and ends with an evaluation of the performance of the program, a sample of the questions that it is able to answer, and its relation to other programs of similar na- ture. 1 Introduction We present an approach to the acquisition of knowl- edge from encyclopedic texts. The goal of this re- search is to build a knowledge base about a given topic by reading an encyclopedic article. Expert sys- tems could use this database to tap in for pieces of knowledge, or a user could directly query the database for specific answers. Then, two possible applications could be derived from our research: (a) the automatic construction of databases from ency- clopedic texts for problem-solvers and (b) querying an encyclopedia in natural language. The idea is to build a database from an encyclopedic text on the fly. Then, if a user asks the question, say, Which bears eat seals? the system would reply by saying something like "I don't know. But, wait a minute, I am going to read this article and let you know." In the process of reading the article, the system builds a small knowledge base about bears and calls the *This research is being funded by NASA-KSC Con- tract NAG-10-0120 question-answering system to answer any question posed by an expert system or a human. The long- term goal of our research is to read an entire ar- ticle on, say bears, and to build a knowledge base about bears. Since this goal requires dealing with an extraordinary number of research issues, we have concentrated on a series of topics about animals, in- cluding diet, habitat, and classification. A skimmer scans the article for sentences relevant to a given topic and passes these sentences to the understand- ing system, called SNOWY, for complete parsing, in- terpretation, concept formation, concept recognition and integration in long-term memory (LTM). But, if, in the process of learning about the dietary habits of, say beetles, the program is told that they culti- vate fungi, and the program is able to interpret that sentence, that knowledge will also be integrated in LTM. Consequently, our …

Content truncated to 500 words