Deductive Coordination of
Multiple Geospatial Knowledge Sources

Richard Waldinger - Martin Reddy ¹ - Christopher Culy - Jerry Hobbs ² - Peter Jarvis
Artificial Intelligence Center, SRI International,
333 Ravenswood Avenue
Menlo Park, CA 94025, US
{waldinger, reddy, culy, jarvis, hobbs}@ai.sri.com - Jennifer Dungan
Ecosystem Science and Technology Branch
NASA Ames Research Center
Moffett Field, CA 94035, US
jdungan@mail.arc.nasa.gov

Deductive inference is applied to choreograph the cooperation of multiple knowledge sources to respond to geospatial queries. When no one source can provide an answer, the response may be deduced from pieces of the answer provided by many sources.

Sources so far include

• The Alexandria Digital Library Gazetteer, a repository that gives the locations for more than four million place names.
• The CIA World Factbook, an online almanac with basic information about more then 200 countries.
• The SRI TerraVision 3D Terrain Visualization System, which displays a flight-simulator-like interactive display of geographic data.
• The NASA GDACC WebGIS server, which stores and distributes satellite data.
• NASA MODIS Landcover Data.
• The Northern Arizona University Latitude/Longitude Distance Calculator.

Queries are phrased in English and are translated into logical theorems by the Gemini Natural Language Parser. The theorems are proved by SNARK, a first-order-logic theorem prover, in the context of an axiomatic geospatial theory. The theory embodies a representational scheme that takes into account the fact that the same place may have many names, and the same name may refer to many places. SNARK has built-in procedures (RCC8 and the Allen calculus, respectively) for reasoning about spatial and temporal concepts. External knowledge sources may be consulted by SNARK as the proof is in progress, so that most knowledge need not be stored axiomatically. The Open Agent Architecture (OAA) facilitates communication between sources, which may be implemented on different machines in different computer languages. An answer to the query, in the form of text or an image, is extracted from the proof. Three-dimensional images are displayed by TerraVision. The combined system is called Geo-Logica.

Typical Geo-Logica queries include

• Show the petrified forests in Oregon north of Portland.
• Show the lake in Argentina with the highest elevation.
• Show the land cover for Montana.

Use of a theorem prover allows sources to cooperate even if they adapt different notational conventions and representation schemes and have never been designed to work together. New sources can be added without reprogramming the system, by providing axioms that advertise their capabilities. Future work involves

• Entering into a dialogue with the user, to clarify ambiguities, elaborate on previous questions, or provide new information necessary to answer the question.

• Dealing with data that represents changes over time, with answers displayed in a moving picture.