18-03-2019, 02:22 PM
Is there a measure of similarity which can be used to compare generated texts to the real text? If so, can we establish a minimum sets of generative rules to create maximum similarity?
To me, that would help to demonstrate whether a procedurally-generated text has any chance of being the correct solution. If it's not possible to approach 90%+ similarity without a large and complex ruleset then we must discard it as a theory.
I suppose this idea is complicated by the fact that we can start from an arbitrary base, such as parts of words created by a human, and them only devise a set of rules to combine them. Even so, we could incorporate some measure of "starting position" to balance out the size of the ruleset.
To me, that would help to demonstrate whether a procedurally-generated text has any chance of being the correct solution. If it's not possible to approach 90%+ similarity without a large and complex ruleset then we must discard it as a theory.
I suppose this idea is complicated by the fact that we can start from an arbitrary base, such as parts of words created by a human, and them only devise a set of rules to combine them. Even so, we could incorporate some measure of "starting position" to balance out the size of the ruleset.