Step Eight

Consider the following two sentences:

"Derek threw the baseball to Tino. It left his hand, flew through the air and landed in Tino's glove before the runner touched first base."

Here we have two sentences that at the very least model a cause and effect for the verb "throw". It is even possible to model a rule for "throw" based on its context in the two sentences. The rule could be something to the effect:

Forall x, y and z

If a human x throws a baseball z to a human y,

Then z leaves x's hand and travels through the air to reach y.

Suppose a representation were created for the two sentences that modeled neither a cause/effect or ant/cq proposition. The verb algorithm would be unable to independently generate the cause/effect or ant/cq proposition as it has no capabilities to create entirely new consequences or effects for actions. The representation must include its own set of rules so consequences are categorized and found by the verb algorithm.

The verb algorithm uses belief revision (SNeBR) in order to revise consequences when inaccuracies are found. Belief revision leads to a more refined and accurate definitions for verbs. Ehrlich provides seven different knowledge categories with which beliefs are to be labeled. When SNeBR is invoked and a discrepancy has to be resolved the hierarchy of the knowledge categories aids the system in determining which beliefs have the least certainty.

I am unsure of how knowledge categories work with SneBR. It is my hope to create a scenario with the proliferate passage where contradictions exist, SNeBR is invoked and revison takes place based upon the knowledge category hierarchy provided by Ehrlich's algorithm.