10.3 The CYC Browser
The most distinguishing feature of the system presented in the this section is its underlying knowledge base, CYC, which tries to model the common-sense knowledge of the average US-citizen. The CYC project [Len90] is a large scale project that started at MCC[10] extending over a period of ten years. Its goal is to build a very large knowledge base containing a broad range of common-sense knowledge. The CYC knowledge base consists of a network of units or frames, where each unit corresponds to a physical object or an abstract concept. CYC currently encompasses a few hundred thousand rules and is projected to have several million when it is completed. The CYC browser [Tra89] uses a visual representation for navigating knowledge structures, based on a virtual museum metaphor. The browser named MUE (Museum Unit Editor) operates in an ideal environment in the sense that the knowledge structures to be modeled are already fully machine understandable. In contrast to MUE, the knowledge bases of all other navigation systems have to be brought either manually or automatically into a format that can be understood by the computer.
The following CYC unit (adapted from [Tra89]) represents the class of all people, specifying that #%People has specializations (#%specs) such as #%workers and #%FemalePeople, generalizations (#%genls) such as Mammals, etc..
#%People:
#%english: ("The class of all human beings")
#%specs: (#%Workers #%FemalePeople #%MalePeople #%USCitizens ...)
#%genls: (#%Mammals #%IntelligentEntities)
#%allElements: (#%MikeTravers #%DougLenat ...)
#%canHaveSlots: (#%citizenship #%languagesSpoken...)
...
Figure I.42 shows the MUE version of the #%people unit of above.
Figure I.42 CYC #%people unit displayed in MUE
As can be seen in figure III.42, each CYC unit is represented in MUE as a box, while units within units are represented as nested boxes. Color is used to indicate the type and the nesting depth of a relationship. It is the goal of MUE to not only display statistical relationships and units as exemplified in figure I.42, but to allow navigation through the knowledge structure. To do so, MUE walks through the graph constructed by the CYC units. The actual unit is always put at the root of the graph to be displayed. This means, that for a walk through the graph the graph is constantly re-rooted to put the actual node at the root (fig. I.43).
To construct a hierarchy, directions within the graph are inverted by inverting the relationships between units. Consider for an example the "#%people" unit from figure I.42. Figure I.44 contains the original graph structure extracted directly from the knowledge base.
Figure I.44 Class structure containing the #%people unit
Figure I.45 Class structure re-rooted such that "#%people" is root
Figure I.46 Re-rooted class structure with inverted directions to get hierarchy
MUE offers an elegant graphical way of navigating in a very large knowledge base. But contrary to most real world examples, the CYC knowledge base has a very strict syntax and thus makes it easy for MUE to reshape the display graph based on the CYC units. This means that the ideas of MUE are hard to apply to unstructured knowledge bases. The long term goal, of course, is to use CYC for structuring unstructured information. When CYC will encompass the full common-sense knowledge, it hopefully will be possible to use it jointly with an extended MUE to model unstructured knowledge bases.
While the main goal of MUE is the visualization, browsing and editing of the CYC rules and units, the system introduced in the next section uses CYC as a tool to implement content-based navigation in plain-text documents.