Cascaded lexical processing to semantics in spoken embedded word

Abstract

Previous research has revealed that visually presented subset words embedded within a carrier word are processed to the level of semantics and can interfere with performance in a semantic categorisation task (Bowers, Davis, & Hanley, 2005). For example, a participant is presented with a category (e.g. “Is this a vehicle?”) followed by words belonging (e.g. truck), or not belonging (e.g. tree), to that category. The critical words in the study did not belong to the stated category but contained embedded words that did belong to the category, e.g. the word scar (containing the embedded word car). The carrier word (scar) demands a ‘no’ response as to whether this word belonged to the vehicle category, while the subset word (car) would demand a ‘yes’ response. Bowers et al. found increased reaction times (RTs) and error rates to visually presented conflicting items (i.e. carrier words that contained embedded words belonging to the specified category) relative to non-conflicting items (when both the carrier word and the embedded words were unrelated to the category). This indicated some interference from the embedded word during semantic categorisation of the carrier, and was thought to indicate that lexical processing of the subset words had reached the level of semantics.
Method: In order to establish whether lexical activation of embedded words occurs in spoken words 30 participants were presented with 8 categories followed by 119 auditorily presented words requiring either a “yes” or “no” response as to whether they belonged to that category. As above, critical words were either conflicting (the carrier demanded a “no” response while the subset word demanded a “yes”) or non-conflicting (both the carrier and the subset word demanded a “no”). The embedded word occupied either an initial, middle, or final position within the carrier word, and crucially had either the same or different pronunciation as if it had been heard within the carrier word, e.g. ram within ramp (same) as opposed to ear within earn (different; see Table 1 for more stimuli examples).
Results: Conflicting stimuli yielded longer RTs (907ms versus 893ms, t(29)=3.95, p < .001), and higher error rates (1.6% versus .7%, t(29)=5.19, p < .001) than non-conflicting stimuli. RTs and error rates from conflicting trials (e.g. Is ‘ramp’ an animal?) were then subtracted from non-conflicting trials (e.g. Is ‘ramp’ a vehicle?) to produce a conflict score that revealed the amount of semantic interference per stimulus type. These data were submitted to individual paired t-tests and revealed a significant difference in RTs and errors for only initially embedded subset words with the same pronunciation (see Table 1 for full statistics). Results from this paradigm provide evidence that when words are presented auditorily subset word processing cascades to the level of semantics to produce interference effects when attempting to semantically categorise the larger carrier word, but only in cases of onset-embedded words. We discuss our finding in the context of recent ERP research that has found effects for final-embedded spoken words, but only following a supporting sentential context (van Alphen & van Berkum, 2009).