IDyOT: A Computational Theory of Creativity as Everyday Reasoning from Learned Information - Computational Creativity Research: Towards Creative Machines - page 139

Fig. 7.4 Confusing IDyOT: how priming resolves speech analysis in the wrong direction. As
previously, points where there is a distribution containing a potential choice in the example are
shaded. This example is different from the earlier ones, because it includes a semantic context
level, conditioned by previous input, that adjusts the prior provided by the language model. The
assumption here is that speech recognition has been the topic; so the distribution associated with
/r3k/ leads away from “wreck” and towards “recognise”. Once that path is taken, the semantic
association is reinforced and inference overcomes the correct reading where the phonemes can be
conflated, indicated by the double-line arrows . As before, solid arrows indicate strong likelihoods,
and broken ones are relatively weak
It also allows us to better understand the functions of the system independent of their
implementation. Thus, we are better placed to improve future versions of the model.
7.6 Creativity
Finally, having discussed an extended example of how IDyOT perceives and analyses
sentences, we turn to its potential for creativity and other general reasoning.
The key here lies in the idea of predictive parsing laid out in the previous sec-
tions. It is necessary to drop the more conventional notion of parsing as the input
and processing of known symbols, and instead to view the parser as a prediction
machine which continually attempts to match its predictions with what is perceptu-
ally encountered. There is considerable flexibility in this process, which can result,
as demonstrated above, in substantial changes in semantics. The benefits conferred
by this approach are efficiency and robustness: prediction allows the listener to get
ahead of the speaker, and also to reconstruct obscured or unclear parts of the input