CUMULATIVE SEMANTIC INHIBITION
12
contributions of these parameters, and their role in the cumulative inhibition effect.
The second important observation is the fact that more than one grouping parameter
(i.e. “category”) is needed to account for the performance. As shown by our second analy-
sis, a categorical representation of the items that relies solely on one level of abstraction -
be it the level of co-categories, or the level of supra-categories - would not capture the whole
pattern of cumulative inhibition. In short, more than one level of abstraction is required.
This conclusion argues in favor of a hierarchical representation model. In such a model,
individual items would belong to more than one semantic (or structural) nesting level. The
analysis we report was constrained by the categories available in the original study, yielding
only two nesting levels that were statistically tractable. Yet this was sufficient to indicate
that the cumulative inhibition effect is a useful tool for testing the structure of the repre-
sentational network involved in lexical access. If more than two levels of abstraction are
ultimately shown to modulate performance, featural or distributed representations (Lam-
bon Ralph, McClelland, Patterson, Galton, & Hodges, 2001; McClelland & Rogers, 2003;
McRae, de Sa, & Seidenberg, 1997) may provide the most appropriate description of the
representations driving lexical access.
Two final points should be raised about the nature of the cumulative inhibition effect.
Following Howard et al. (2006), we have modelled the effect with linear predictors (we did
not observe any non-linear components reaching significance). This provides reliable esti-
mates for groupings with a limited number of items. Obviously, a linear effect of magnitude
30 ms may become unrealistic for item groupings comprising ten or more items. Therefore,
the linear effect may prove to be non-linear after all. Also, as noted in the Introduction and
in the paragraphs above, we have described the cumulative inhibition effect mostly in se-
mantic terms. A thourough investigation should weight against one another possible visual
and semantic similarity contributions to the effect. Previous attempts of this kind in other
bare picture naming experiments have favoured a semantic interpretation independent of
visual similarity (Damian, Vigliocco, & Levelt, 2001).
In conclusion, we report an analysis of the cumulative inhibition effect discovered by
Howard et al. (2006). Where these authors reported a single regression line, we showed
that a richness of systematic variations can be observed and, more importantly, predicted.
Our results are better understood in terms of featural or distributed representations driving
lexical access.
References
Baayen, H. R. (2007). Analyzing linguistic data: A practical introduction to statistics. Cambridge
University Press.
Baayen, H. R., Davidson, D. J., & Bates, D. M. (in press). Mixed-effects modeling with crossed
random effects for subjects and items. Journal of Memory and Language.
Bates, D. M. (2005). Fitting linear mixed models in R. R News, 27-30.
Capitani, E., Laiacona, M., Mahon, B. Z., & Caramazza, A. (2003). What are the facts of category-
specific deficits? A critical review of the clinical evidence. Cognitive Neuropsychology, 20,
213-261.
Damian, M. F., Vigliocco, G., & Levelt, W. J. M. (2001). Effects of semantic context in the naming
of pictures and words. Cognition, 81 (3), B77-B86.
Efron, B. (1979). Bootstrap methods: Another look at the jackknife. The Annals of Statistics, 7,
1-26.