[ CogSci Summaries home | UP | email ]

Anderson, M. & McCartney, R (2003) Diagram processing: Computing with Diagrams. Artificial Intelligence. 145. pp 181--226.

  author = 	 {Anderson, Michael and 
                  McCartney, Robert},
  title = 	 {Diagram processing: Computing with Diagrams},
  journal = 	 {Artificial Intelligence},
  year = 	 {2003},
  key = 	 {},
  volume = 	 {145},
  number = 	 {},
  pages = 	 {181--226},
  month = 	 {},
  note = 	 {},
  annote = 	 {}
Cite this paper for:

Author of the summary: Jim Davies, 2006, jim@jimdavies.org

Text evolved from pictorial representations. [182] Then other words used those symbols based on phonetic similarity. The text has an example of a bull turning into the letter "A".

Sloman1995,2002: analog representations resemble, in some way, the thing represented. Fregean representations do not. the USA is a Fregean rep of a geographic entity, but you can't see it from a photo of the landscape.

A good deal of attention has been paid to language understanding, since an AI that has to function in human environments must be able to understand these things. By the same reasoning AI should be able to understand diagrams. [184] Diagrams are 2D analogically represented abstractions (maps but not photographs.)

The way they approach the problem is to treat the input as an actual diagram, which gets encoded as greyscale values on the tiling of a planar area. [185] This is a domain-general approach. The Interdiagrammatic Reasoning (IDR) approach "specifies a general diagram syntax and set of operators that leverage, for computational purposes, the spatial and temporal coherence often exhibited by groups and sequences of related diagrams." The output is diagram and text, intended for human cognitive consumption. [186] "Although much of the following has been previously reported in a piecemeal fashion, it is presented here for the first time in its entirety, recast under the most mature version of our approach, with the newly gained perspective of its relationship to the notion of diagram processing. Seen in its entirety, the true breadth of the domains and variety of uses of our approach is made evident, bringing its generality into relief in a way not possible previously." "Although diagrammatic representations may need to be augmented by other representations to completely represent many problem domains, we attempt to rely upon diagrammatic representations as much as possible to bring our understanding of them on par with other, better understood representation methods. Only when parity is achieved in our understanding of all representation methods can we make informed judgments concerning their respective uses."

IDR compares multiple diagrams of the same thing or sequential diagrams. [188] IDR can take the CMY colors of two diagram's pixels (called tessarae) and do operations like and, not, sum, overlay, etc. [189]

Diagrammatic Information Systems (DIS) are constrained diagram processing systems that permit users to pose queries concerning diagrams, seeking responses that require reasoning about diagrammatic as well as non-diagrammatic data. [191] So given a map of the USA depicting areas with grassland, and another map with the states, a DIS should be able to show which states have grassland, with a new map that colors states differently.

Diagrammatic SQL (DSQL) extends SQL for querying diagrams.[193]

Operations like "overlay" are used to help with decision making in the battleship game. [198] Also with n-queens. [201] Which can be done as a kind of diagrammatic CBR [206]

Summary author's notes:

Back to the Cognitive Science Summaries homepage
Cognitive Science Summaries Webmaster:
JimDavies (jim@jimdavies.org)