CUMULATIVE SEMANTIC INHIBITION
construct a working model of word selection involving a process of priming and competitive
selection in which the simulated naming performance exhibits the observed cumulative
effect. The purpose of this article is not the cumulative nature of this effect per se. Rather,
we focus on the relatively undiscussed structure of the semantic representations that may be
causing the effect. In doing so, we will refer to the effect as cumulative semantic inhibition
for consistency with the terminology used in the original study. A possible contribution of
visual factors will be addressed in the General Discussion.
Our main question is whether the reported cumulative inhibition effect provides any
information about the representation of the semantic categories that were tested. As we will
show, depending on the representational details one assumes, the linearity of the effect that
has been used to inform models of lexical access may or may not hold for a different choice
of categories or a different sequential arrangement of the items. This observation may lead
to important conclusions regarding semantic and lexical access processes (cf. Moss, Tyler,
& Taylor, 2007; Vigliocco & Vinson, 2007).
Howard et al. (2006) used a total of 24 categories (listed in their appendix). The use of
such a large number of ensures a high statistical power for the analysis in which the ordinal
position and distance parameters are contrasted. It also leads to considerable diversity in the
definitions, with categories ranging from rather general sets (e.g. buildings or furniture) to
more specific ensembles (e.g. computer equipment or farm animals). Despite this diversity
in the definition of categories, the reported semantic cumulative effect is remarkably strong
in the analysis “by categories”, suggesting that it is not driven by a subset of the materials.
Our investigation will pay special attention to the role played by these categories in
the observation of the effect. We will proceed in two steps, in which we will answer the
following related questions:
1. Is the magnitude of the effect similar across the 24 categories, or does it show a
systematic variability?
2. Is the effect better understood in terms of categorical representations of these
particular categories, or in terms of a by-product of some other representational structure?
To answer the first question, we reanalyzed Howard et al.’s (2006) original dataset
using a more sensitive technique. To answer the second question, we report an extension of
our analyses with newly defined categories replacing the original ones.
Our analyses used the mixed-effect modelling methodology (Bates, 2005) recently
introduced in psycholinguistics (Baayen, 2007; Baayen, Davidson, & Bates, in press). This
technique relies on single trial data, rather than on averages by participant, by category, or
by ordinal position. In this way, the fixed vs. random nature of these effects is explicitly
considered. This enables more detailed and robust analyses than it is possible with the
traditional analysis techniques.
Variability of the cumulative inhibition effect across categories
By focusing on single trial data we were able to test, in a single analysis, the potential
contributions of the different order and distance parameters that characterize every trial of
the experiment. Furthermore, this analysis enables the investigation of a question that was
not addressed in the original study, namely possible systematic variations in the magnitude
of the cumulative inhibition effect across categories.