Key Structural Properties of VR Frame Representations

[josh]

Computational efficiencies relate to both time & space. In some situations, it may be a good strategy to adopt data access/storage algorithms that optimize speed of answering “Is THIS feature combination an occurring, possible, or empty set case?” If it is occurring or possible, we can ask how likely/common it is or how to create it. If different feature sets are of interest, then overall question may be speeded by a series of sub-queries with abbreviated feature vectors. This point includes the observation that data access is not just like conventional database querying because flexible reality & thoughts about is not like a fixed dimension vector space that can be held in a fixed set of table joins.

[josh]

Consider the VR structure view of adding a sentence/(1 or more sentences of predicated logic) to create modified VR structures. The before condition includes VR structures in discourse context focus & general back ground knowledge. The predicate structure of the sentence can be coarsely understood as an information sharing maneuver directing us to add to our VR discourse context, including new additions with feature regions that conform to the logical structure of predication. In other words, if the logical statement expresses R(x,y,z) then the result of adding includes some region of VR with non-empty intersection with the support of R(x,yz) – or ~R(x,y,z) can require that the intersection is empty.

One way of testing the adequacy of the VR solution is via the ability to successfully summarize & paraphrase using only the VR forms as records.

[josh]

Consider work on representations of language about smells as a helpful exercise. It has these features:

a) The level of detailed complexity & structural complexity is low compared to many other types of communication.

b) Humans mostly resort to similarity-based language rather than direct-to-sensation adjectives – smells like a fruit rather than smells like THIS PARTICULAR COMBINATION OF ESTERS AND ….

c) Visualization of textual representations can’t make as much use of analog mappings (compared to vision)

It should be easier to get smell talk “right” in the sense of adequate performance while not being fooled about artifactual successes (i.e. interesting pictures that are artifactually creative rather than capturing a communication).

[Author]

Posts