2007 Volume 24 Issue 4 Pages 4_2-4_22
In recent artificial intelligence research, one of the fundamental problems is to handle uncertainty in symbolic data. To handle such uncertainty, one promising way is to construct a probabilistic model that represents the data well, and then to make probabilistic inferences based on the model. Furthermore, as the data get more complicated than traditional data matrices, like sequence data or relational databases, the importance of the expressivity of the modeling language has been increased. From such a background, there have been proposed plenty of formalisms that attempt to integrate first-order logic and probability. In this paper, we present a probabilistic logic programming system called PRISM, which is expected as an efficient tool for probabilistic modeling which features declarative semantics, high expressivity originated from logic programs, and built-in fast routines for various probabilistic inferences. We describe PRISM as an implemented tool with program examples, its functionalities, and the result of a benchmark evaluation of computing performance.