One of the most common method of eliciting false memories in the laboratory is the Deese-Roediger-McDermott paradigm (Deese, 1959; Roediger \ McDermott, 1995), where participants study a set of items that are all similar to a non-presented critical lure. A common finding is that false recognition to critical lures is much higher than to other non-presented items and in some cases is even comparable to true recognition, regardless of whether similarity is semantic or structural (e.g., phonological or orthographic) relations. While there exists a handful of computational models of this paradigm, they have only been applied to semantic but not structural false recognition, they have not been fit at the level of individual participants, and they have not been applied to response times (RTs). We present a global matching model that addresses all three of these current gaps. Global similarity of semantic and structural representations drives a pair of linear ballistic accumulators, which are used to produce decisions as well as complete RT distributions. In addition to being able to account for heightened false recognition of critical lures, the model was able to account for differences across both individual participants and items, lower correlations between semantic and structural false recognition than true recognition, differences in false recognition across levels of processing, improved true recognition but not false recognition with higher study time, and heightened false recognition under speed emphasis. The model suggests that semantic and structural false recognition can be explained using only a single retrieval mechanism.
Osth et al. (Fri,) studied this question.