Pattern Recognition
The notion of the ability to recognize patterns in information streams
For John Sowa, there are these two notions in Conceptual Graphs:
- Canonical Graphs
- Graph Mappings
The idea is this:
If you can create Graph X from an input stream, say the simplest graph being a triple {some subject, is the cause of, some other subject}, then if you can find a canonical graph for that pattern to which the subjects and predicate can map, then the canonical graph can tell you more about that scenario.
For George Lakoff, there is the concept of basic level category, a basic level for concepts on which we think. For instance, a basic level category for most automobiles might be "car".
Suppose we map BLC into the realm of canonical graphs in this sense:
- Named entities are given some basic level category
- Leukemia might map to Cancer, say
- Predicates are given some basic level category
- Varieties of saying a causes b are mapped to cause
With those basic level categories in the system vocabulary, then we have a means to fabricate or learn canonical graphs. As an input stream is processed (akin to parsing), it is mapped to the greatest degree possible to BLCs; when that is completed, detection of canonical graphs becomes a kind of case-based reasoning (Schank et al)