The dynamics of Pr ignanz
Psychologische Forschung, Peter A. van der Helm · 1994
chunking contains seven different elements: x, z, (xz), t, p, q, and (pq). The reason to count all different elements in the abstract chunking that results from a code is twofold. First, the number of different symbols in the abstract chunking equals the number of pattern symbols in the code and can therefore be said to reflect the amount of irregularity (residual nonidentity) in the code (cf. Collard & Buffart, 1983). Second, the chunks in the abstract chunking reflect the hierarchy in the code, and therefore the number of different chunks can be said to quantify hierarchy in terms of irregularity at higher hierarchical levels. So, the In~w load measures code complexity in terms of irregularity at all hierarchical levels. Two further examples may be illustrative. In both Figures 2 and 3, the hierarchical chunking is also the abstract chunking. So, the code S[2*((a)(b))] in Figure 2 gets Inew = 3, since the abstract chunking ((a)(b))((a)(b)) (b)(a)(b)(a) contains three different elements: a, b, and ((a)(b)). Similarly, the code / in Figure 3 gets Inew = 5, since the abstract chunking ((k)(x))((k)(y))((k)(x)) contains five different elements: k, x, ((k)(x)), y, and ((k)(y)). The Inew load is more plausible than the Iold load since, unlike the Iold load, the In~w load does not depend on syntactical artifacts of the coding model. Moreover, the theoretical significance of the Inew load is enhanced by its strong relation with the ISA rules. For, the Inew load can be applied only to transparent coding rules, since it requires hierarchical chunking. So, it cannot be applied to, e. g., the D or the T rule, as these coding rules are not transparent (nor holographic, by the way). In that sense, the I~w load differs essentially from all other measures proposed so far, and provides a possible way out of an empirical deadlock. For it has been difficult to decide empirically which set of coding rules is most appropriate, since the complexity measure seems more decisive than the coding rules (cf. Simon, 1972). Now, however, empirical support for the Inew load automatically implies support for (the transparent character of) the ISA rules.