It has been suggested and it is widely believed that the experimental technique of crossmodal semantic priming with partial primes, combined with lexical decision as a subjects’ task, is a suitable method for obtaining information on the temporal course of multiple lexical activation and selection during spoken-word recognition (Zwitserlood, 1989; Frauenfelder & Floccia, 1999). We will report on such an experiment that was in many ways a replication of Zwitserlood (1989), although some changes were made in linguistic material and design. Specifically, a within-subjects design was used instead of a balanced-incomplete-blocks design. We failed to produce consistent priming effects with partial primes. This was reason for a computer simulation comparing the statistical analysis of a balanced-incomplete-blocks design and a within-subjects design, using fictitious data sets. The balanced-incomplete-blocks design together with normalization as applied by Zwitserlood, was shown to easily produce spurious significance. We argue that, unfortunately, crossmodal semantic priming with partial primes is not a reliable method to study the time course of multiple activation and selection.

In order to study the time course of lexical activation as a function of sensory information and context, we must apply a different technique. In another experiment we used rhyme monitoring. Admittedly, rhyme monitoring may bias the results because the rhyme cue very likely causes phonological priming and thus pre-activation of the word candidate. However, if in the presence of such pre-activation no early context effects are found, one can make a strong case for context only affecting the late stages of word recognition. If, on the other hand, one does obtain early context effects in a rhyme monitoring experiment, this would show that at least in some conditions context may affect lexical activation before sensory information has led to a unique choice. Our experiment shows clear context effects at a moment in time that disambiguating sensory information cannot possibly have arrived. We conclude that the early stages of word recognition are not always impenetrable for context effects.