ABSTRACT
We investigate if and how the model of hypothesis generation and probability judgment HyGene can be implemented in ACT-R. We ground our endeavour on the formal comparison of the memory theories behind ACT-R and HyGene, whereby we contrast the predictions of the two as a function of prior history and current context. After demonstrating the convergence of the two memory theories, we provide a 3-step guide of how to translate a memory representation from HyGene into ACT-R. We also outline how HyGene’s processing steps can be translated into ACT-R. We finish with a discussion of points of divergence between the two theories.
Acknowledgements
The author would like to thank Julian Marewski for the provided encouragement.
Disclosure statement
No potential conflict of interest was reported by the author.
ORCID
Cvetomir M. Dimov http://orcid.org/0000-0002-8127-6724
Notes
* This research includes and extends work first published elsewhere (Dimov, Citation2016).
1 Dimov (Citation2016) demonstrated that the optimised learning equation is an unbiased approximation of the base-level learning equation even when there is some noise in this periodicity. Specifically, the noise is allowed to increase superlinearly as a function of the number of periods since the event or object encounter.
2 MINERVA2’s typically uses S as a notation for similarity. Since this overlaps with ACT-R’s associative strength, here we will use SM for similarity in MINERVA2. Similarly, we will use AM to indicate activation of a trace and keep A as an activation of a chunk in ACT-R.
3 In ACT-R there are default values for many of the parameters, departure from which is discouraged unless justified.