CUMULATIVE SEMANTIC INHIBITION
11
Original categories
Original categories and Supra-categories
ordinal position within categories 'A' and 'B'
Supra-categories
Figure 3. Predicted performance, with standard errors, for the members of two co-categories (A
and B) interleaved in the experimental lists, for different models instantiating different hypothesis.
Left panel: both co-categories are independent. Center panel: the co-categories depend on one
another such that both the ordinal position in the original categories and in the supra-category
produce independent significant effects. Right panel: the co-categories are in fact a single supra-
category such that only ordinal position within the supra-category produces a significant effect.
restricted dataset, a further cumulative effect was found for newly-defined supra-categories,
over and above the original ordinal position effect within co-categories.
Our analyses argue in favour of using single-trial information, instead of averaging
performance across levels. This is especially true in the case at hand, where non counter-
balanceable absolute and relative positions of individual trials have to be taken into consid-
eration simultaneously to understand the data. Thus, our first conclusion is methodological.
We have shown how the mixed effect methodology that Baayen et al. (in press) introduced
for language comprehension studies provides a powerful tool also for language production
studies.
From a theoretical perspective, the results of our analyses enrich the conclusions of
Howard et al. (2006) in several ways. The first important observation is the significantly
variable inhibition effect across categories. The fact that it is unrelated to overall speed
indicates that this variation does not reflect a mere performance effect. One plausible
explanation for this variation may come from some intrinsic property(ies) of the categories
in question. For example, the degree of similarity among the items of a category has been
previously suggested to influence the processes of lexical access. This hypothesis has been
operationalized by distinguishing categories with structurally similar or dissimilar objects
(Humphreys, Riddoch, & Quinlan, 1988). The degree of similarity can also be defined
on semantic grounds, by establishing semantic distances between related and unrelated
items. The picture-word interference paradigm has been used to investigate the effect of
this parameter. Some authors have reported worse performance when the target and the
distractor words were semantically closer (Vigliocco, Vinson, Damian, & Levelt, 2002).
Others, however, have observed that reducing semantic distance had - if anything - a
facilitatory effect (Mahon, Costa, Peterson, Vargas, & Caramazza, 2007). Further ad-
hoc studies using the simple picture naming task may contribute to clarify the relative