Atext understandingsystemwith learning capabilities is presented.Newconceptsare acquiredbyincorporating twokindsof evidence- knowledgeaboutlinguistic constructionsin whichunknownlexical itemsoccur and knowledgeaboutstructural patterns in ontologiessuch that newconceptdescriptionscanbe comparedwithprior knowledge.Onthe basis of the quality of evidencegathered this wayconcepthypothesesare generated, ranked accordingto plausibility, andthe mostcredibleonesare selectedfor assimilationinto the domainknowledgebase.