The influence of the place-value structure of the Arabic number ...

single node in an ordered sequence of input nodes. In several studies by different authors and
model architectures it was observed that empirical data from different numerical tasks was
associated reliably with the modelled data (number comparison: e.g., Dehaene & Changeux,
1993; Zorzi & Butterworth, 1999; number naming: e.g., Grossberg & Repin, 2003; Verguts et
al., 2005; number priming: Verguts, Stevens, & Fias, 2005; Zorzi, Stoianov, & Umiltà, 2005).
However, almost all of these models were focused on small numbers and numerosities up to
15 (Verguts et al., 2005; up to 5 in Dehaene & Changeux, 1993; 9 in Zorzi & Butterworth,
1999; but see the discussion of the Grossberg & Repin, 2003 model below). Moreover,
considering the relevant literature revealed that most of the papers on computational models
of number representation were concerned with questions regarding the actual coding
characteristics upon the assumed and implemented number line representation. Recently,
summation or numerosity coding (i.e., representing magnitude by the number of nodes
activated, e.g., Zorzi & Butterworth, 1999) and place coding (i.e., representing a specific
number by the activation of a node actually reflecting its position on the number line as well
as the preceding and subsequent one, e.g., Verguts & Fias, 2004; Verguts et al., 2005) were
evaluated. Neither of these coding schemes differentiates between the representations of
single-digit and two-digit numbers: in the first case all nodes along the number line up to the
i th node are activated to code magnitude i and in the latter case only the i th node is activated to
do so. In both cases no differentiation between e.g., tens and units for two-digit numbers is
assumed (but see Grossberg & Repin, 2003). Taken together, to date computational modelling
was primarily employed to clarify coding and scaling aspects of numerical magnitude along
the mental number line rather than distinguishing between different processing models (see
Verguts et al., 2005 for a more detailed discussion of this point).
However, the model by Grossberg and Repin (2003) presents an exception to these
considerations. In this model, representations of single- and multi-digit numbers are
discerned. The authors assume multi-digit numbers to be represented by an interaction of the
239