Abstract
In a verification task of simple additions composed of Arabic or Roman numerals, Gonzalez and Kolers (1982) reported data that were interpreted as supporting the idea that cognitive operations are not independent of the symbols that instigate them. We propose an alternative interpretation of these results and argue that the effects reported may have been produced by a peculiarity of the Roman code for which the encoding time would not be constant for all numerals. We hypothesize that three different “structures” can be distinguished in the Roman code, and that the time necessary to encode a numeral would vary according to its structure, with the analogical (numerals I, II, and HI) and the symbolic (V, X) structures being processed faster than the complex structures (IV, VI, VII, VIII, IX, XI,…). This structure effect is tested in two experiments: a verification of transcoded forms and a parity judgement. Data repeatedly showed support for this hypothesis. Moreover, a verification task for additions showed that the presentation format of the addends played a role in the encoding stage but did not interact with variables relative to the size of the addition problems. These data could thus sustain the hypothesis of a “translation model”, according to which numerals would be translated into a specific code to which the calculation process would be applied.
