We investigated the effect of semantic information on word memory, using imageability and sensorimotor strength as predictors, with data from a mega-study of word recognition memory (Cortese et al., 2010; 2015), as well as from an online memory task. Memory performance was analysed in hierarchical linear regressions. Both sensorimotor strength and imageability had an effect on word memory performance, but not as strong as reported in previous literature. However, the effects were smaller when the memory task was unexpected, suggesting that the semantic effects are dependent on memory strategies (or context). Most importantly, different types of sensorimotor strength had a variable effect on memory, which was not in line with the prediction of the semantic richness effect, and highlighted the importance of a multi-dimensional approach to measuring and testing semantic experience, and its effect on cognitive processing. The findings have implications for the use of semantic variables in memory research.

The linguistic-simulation approach to cognition predicts that language can enable more efficient conceptual processing than sensorimotor-affective simulations of concepts. We proposed that this has implications for working memory, whereby use of linguistic labels enables more efficient representation of concepts in a limited-capacity store than representation via full sensorimotor simulation. In two pre-registered experiments, we asked participants to remember sequences of real-world objects, and used articulatory suppression to selectively block access to linguistic information, which we predicted would impair accuracy and latency of performance in an object memory recognition task. We found that blocking access to language at encoding impaired memory performance, but blocking access at retrieval unexpectedly facilitated speed of responding. These results suggest that working memory for object concepts normally relies on language but people can flexibly adapt their memory strategies when language is unavailable. Moreover, our data suggest that a sequence of up to 10 object concepts can be held in working memory when relying on sensorimotor information alone, but this capacity increases when linguistic labels are available.