Discovering the significant relations embedded in documents would be very useful not only for information retrieval but also for question answering and summarization. Prior methods for relation discovery, however, needed large annotated corpora which cost a great deal of time and effort. We propose an unsupervised method for relation discovery from large corpora. The key idea is clustering pairs of named entities according to the similarity of context words intervening between the named entities. Our experiments using one year of newspapers reveals not only that the relations among named entities could be detected with high recall and precision, but also that appropriate labels could be automatically provided for the relations.

No abstract

No abstract

No abstract

Eighty-two Klebsiella O1 strains belonging to various K types were tested for virulence for mice by intraperitoneal inoculation. They comprised 49 strains newly isolated from various clinical specimens, 31 reference strains, and 2 laboratory strains. Of 9 Klebsiella O1:K2 strains, 7 were highly virulent inasmuch as their 50% lethal doses per mouse were less than 10 CFU, whereas the other 2 strains were avirulent even when encapsulated just like the virulent strains as revealed by the quellung test. Klebsiella K2 reference strain B5055 and strain Chedid, which were maintained in vitro for a long time, were highly virulent. Of 8 Klebsiella O1:K1 strains, 2 showed moderate virulence, and the other 6 strains had relatively low or no virulence. We found no definite correlation between the virulence of Klebsiella K1 and K2 strains for mice and the kind of clinical specimens in which they originated. All of the Klebsiella O1 strains tested belonging to K types other than K1 and K2 showed very low or no virulence for mice.