Numerical scaling techniques for evaluating generative models of orthographies

Research presented introduces a new approach for empirically investigating generative or inductive-like systems. Two versions of a generative model of the uppercase English alphabet are examined. The approach is both formal and interdisciplinary, applying techniques developed by psychophysics to a problem that is typically linguistic. Experimental results presented illustrate empirical methods and analytic tools used, and demonstrate how the techniques advance the psychological study of the English alphabet. Two different theoretical models of the alphabetic system are evaluated strictly on the basis of empirically observed two-alternative forced-choice data. Scaling methods described produce a numerical scale for generated alphabetic items which permits informative comparisons between scale values and across independently derived scales. The scaling theory used is a variant of Thurstone's Case V model. The methods can also be utilized to further model construction. Implications of the findings for the existing body of writing-system research, and the generalizability of the approach to other domains of investigation are also discussed. PsyclNFO classification: 2100; 2300; 2340