Predicting False Memories with Data-Driven Computational Models: The Role of Visual and Linguistic Similarity in the DRM Paradigm

The Deese–Roediger–McDermott paradigm (DRM) requires presenting lists of words to individuals and then assessing their ability to recall these words. Research has found that when novel words are semantically related to the memorised ones they tend to be erroneously recognised as part of the memorised list. Although the role of linguistic similarity in generating false memories is well established, less is known about whether visual similarity contributes to this phenomenon. To investigate this issue, we adopt a data-driven computational approach to independently quantify visual and linguistic similarities for the referents of words and image stimuli to be presented in two DRM variants and tested their effects on false memory generation. Our results show that false memories consistently increase for novel words and images that are visually and linguistically similar to those presented in the memorised lists, with a larger degree of visual and linguistic similarity in the DRM with images and words, respectively. These findings indicate that visual and linguistic processes are both involved in memory distortions, regardless of whether the information is presented in a visual or linguistic format.