Representing and Reasoning with Intervals Erica Yuan-Ting Huang 1 , Clinton Smyth 1 , David Poole 2 1 Georeference Online Ltd., Vancouver, B.C., Canada 2 Department of Computer Science, University of British Columbia E-mail: ehuang@georeferenceonline.com 1. Abstract Geologists around the world often specify numerically measured information such as age or speed with non-numerical words such as “Jurassic” or “fast” rather than actual numerical values or ranges of numerical values. These qualitative descriptions, which may adhere to different standards, can be applied at different levels of detail or abstraction. This presents difficulties in building knowledge-based systems that have to compare temporally and spatially related information such as geological age or landslide movement rates. This paper discusses the representation of intervals and the reasoning required when comparing interval information. It proposes a format, which can work with different interval domains, for representing intervals based on a current software system developed for qualitative probabilistic matching. This is a necessary first step in being able to reason with various interval specification standards that have evolved in different parts of the world. Not only can this framework be used in the geological domain, such as mineral exploration and geological hazard evaluation, but it can also be used for intervallic reasoning in other scientific disciplines. 2. Introduction A software system for matching semantic network descriptions of instances (e.g. a particular mineral deposit) and models (e.g. a type of mineral deposit) has been developed using a qualitative probability approach (Smyth and Poole, 2004). The commercial implementation of this system is called MineMatch  and deals with mineral deposits. The system is specifically designed to reason correctly when making comparisons between deposits, described at various levels and abstractions, with hierarchical classifications such as the taxonomy of rocks (BGS, 1999). Taxonomic reasoning can be used, to a limited extent, to reason with hierarchical representations of interval attributes, such as geological age or landslide movement rates. However, it limits flexibility in describing ranges such as “from Jurassic to middle Triassic” or “from slow to medium speed”. It is also not sensitive to overlapping intervals, or to the degree of separation between non-overlapping intervals. Hence special representations and reasoning algorithms are necessary for correctly working with interval data.