Integrating Feature Frequency with Orthographic Representations in Global Matching Models of Recognition Memory

Recent studies have integrated realistic orthographic representations of words into global matching models of memory in order to capture similarity effects on an item-by-item basis. However, these models have also made the simplified assumption that all features are equally weighted when determining the similarity between letter strings. Previous work has instead found evidence that words containing rare letter features exhibit better recognition memory performance than words with more common features. The current study aimed to explore the consequences of integrating orthographic feature frequency in global matching models of recognition memory. We conducted three recognition memory experiments with feature frequency manipulations. Participants studied words containing low or high frequency letters (Experiment 1) or bigrams (Experiment 2), or received a factorial manipulation of letter and bigram frequency (Experiment 3). Results showed feature frequency mirror effects – words comprised of rare letters and bigrams showed higher hit rates and lower false alarm rates than words comprised of common letters and bigrams, and showed evidence for a stronger bigram frequency effect compared to letter frequency. These data were fit with global matching models containing orthographic representations where the underlying features were either equally weighted or with weights proportional to their frequency in language. Results consistently favoured the feature frequency models over the equal weight models. The necessity of differential feature weighting was further validated through fitting with a recognition memory mega-study where differential feature weighting improved the model’s ability to capture variability in hit rates across individual words.