CUMULATIVE SEMANTIC INHIBITION
checked. The last column of Table 2 summarizes the relative position of each of the two
co-categories. The order in which they were presented in the experimental lists is relatively
well balanced across participants. Any effect observed for co-categories presented second
can therefore be attributed to their position in the experimental list, rather than to the
specifics of the items that compose them.
Secondly, the absolute position of each co-category within each experimental list has
to be controlled. Obviously, the second co-category will always be presented later in the
experiment. If there is a systematic variation of performance with trial position (see, for
example, the suggestive effect of trial position in N-model 1 above), it should be taken
into account when interpreting relative category position. This was done by considering
three related predictors. The first one, used previously, was the position of the trials in
the list. The second predictor measured the absolute position of the category within the
experimental list. It is defined as a constant number for all the members of a co-category,
namely the trial number of the first item of the category. Finally, we introduced the critical
predictor capturing the relative position of twp co-categories, with values 1 (vs. 0) when
the co-category had (vs. had not) been named before. We also included the predictor of
theoretical interest, ordinal position within the co-category.
N-model 2 shows that the effect of ordinal position within the category was significant.
More importantly, the relative position of the co-category had a strong and significant effect,
whereas the absolute position of the trial, or the absolute position of the category were not
significant. This result indicates that the relative position of the co-category is indeed the
best predictor of naming latencies (better than trial position or category position). This
finding also clarifies why trial position effect may have been significant in the restricted
dataset (N-model 1 ) and not in the complete dataset (e.g. H-model 2). The restricted
dataset includes only categories whose second occurrence is slowed down by their previously
presented co-category.By contrast, the complete dataset includes many categories that did
not have a co-category named earlier.
Summing up, this second analysis reveals a clear dependency between co-categories
which is not due to the absolute position of the trials in the experiment, and which is
observed over and above the effect of ordinal position within category. This is summarized
in N-model 3, where only significant variables were included. The occurrence of the second
co-category is slower than the first one, by an order of magnitude that is similar to the effect
of ordinal position within category. The performance estimated by this model is plotted in
the left-center panel of Figure 2.
It should be noted that the variable magnitude of the cumulative inhibition effect
does not reach significance in these two models. Whether this is a mere consequence of the
reduction in power of the analysis or a reflection of a theoretically meaningful observation
remains an open question.
A very simple explanation for the dependency between co-categories would state that
co-categories are in fact one single category, namely the supra-category listed in Table 2. In
this view, going from the end of one co-category (e.g. 5th farm animal) to the beginning of
its co-category (e.g. 1st zoo animal) is equivalent to increasing by one the ordinal position
of the item in the corresponding supra-category. If this were an appropriate description
of the data, then the cumulative inhibition effect should make trials at the beginning of
the second co-category be slower than trials at the end of the first co-category. A quick