CUMULATIVE SEMANTIC INHIBITION
17
22
16
114
23
69
10
151
2
512
20
14
24
1819 13
21
-100 -50 0 50 100 150
-100 -50 0 50 100 150
Main random effect of Category (in ms)
Figure 1. Estimates of the random intercept of Category (x-axis) and the mixed interaction between
Category and Ordinal position (y-axis) for each of the original categories.Larger values in the x-axis
indicate slower categories. Larger values in the y-axis indicate stronger coefficients for the linear
semantic effect (i.e. stronger inhibition). Each data point is a category, numbered according to
the appendix in Howard et al. (2006). Left: complete data set (H-model 2). Right: subset used
in the second analysis (N-model 1). The magnitudes and relative orderings of these random effect
estimates have been validated using a bootstrapping technique (N = 100; Efron, 1979)
Contrastive hypothesis about the cumulative inhibition effect
The second analysis we report investigates the representational status of some of the
categories used in the original study. Our first analysis showed that all of these categories
produce a sizeable cumulative inhibition effect. Does this mean the categories should be
understood, in any strong sense, as representational categories in the speakers’ cognitive
system? Although a positive answer to that question would be rather surprising (e.g.zoo
and farm animals were distinguished; white goods is a category in itself), the question is a
useful anchor point for our rationale.
An inspection of the list of materials shows that at least 10 categories allow a natural
regrouping in pairs of “co-categories” under a “supra-category” (Table 2). This property
allows manipulating the structure of the categories used to analyze the data, and hence
to provide answers to the question above. If two co-categories are representationally inde-
pendent, then naming the members of one of them should not affect how the members of
the other are named anymore than having named any other items. For example, the speed
of naming of farm animals should be independent of whether zoo animals (or buildings)
were named before or not. By contrast, if the items of co-categories share part of their
representation, such influence should be apparent in the naming performance. In the latter
case the relative contributions of supra- and co-category groupings might be clarified.
We report a series of analysis on a subset of the original data in which only the ma-