ABSTRACT
Anelementarydistinctionthatcanbemadeintheartificialintelligence literatureisbetween"fact-retrieval"systemsand"question-answering"systems.In fact-retrievalsystems,theuserwishestostorevastquantitiesofdata(e.g.,indexing keywordsofabstractsofpapersonorganicchemistry)andthenretrieverelevant partsoftheinformationbysendingafewkeywordsintothesystem.Insuch systems,thereisalmostnoinferencetobedone,nodeduction,nosubtlesemantic interpretationofthequestionitself.Theuserisusuallyrestrictedtoasmall vocabularyofspecifiedkeywords,andthesyntaxofhisretrievalrequestsisusually restrictedtoonlyafewstandardsyntacticframes,withnopossibilityofcomplex embeddingsorthelike.Theissueinsuchsystemsisefficiencyofagivencoding (keying)andorganizationoftheinformationfiles,sothattheuserobtainsrapid andmaximumreturnofdocumentsorfactsrelevanttohisrequest,witha minimumofirrelevantinformationthatmustbescannedforsignificance.Theseare "fact-retrieval"or"document-retrieval"systems.
