In everyday life, human beings can report memories of past events that did not occur or that occurred differently from the way they remember them because memory is an imperfect process of reconstruction and is prone to distortion and errors. In this recognition study using word stimuli, we investigated whether a specific operationalization of semantic similarity among concepts can modulate false memories while controlling for the possible effect of associative strength and word co-occurrence in an old-new recognition task. The semantic similarity value of each new concept was calculated as the mean cosine similarity between pairs of vectors representing that new concept and each old concept belonging to the same semantic category. Results showed that, compared with (new) low-similarity concepts, (new) high-similarity concepts had significantly higher probability of being falsely recognized as old, even after partialling out the effect of confounding variables, including associative relatedness and lexical co-occurrence. This finding supports the feature-based view of semantic memory, suggesting that meaning overlap and sharing of semantic features (which are greater when more similar semantic concepts are being processed) have an influence on recognition performance, resulting in more false alarms for new high-similarity concepts. We propose that the associative strength and word co-occurrence among concepts are not sufficient to explain illusory memories but is important to take into account also the effects of feature-based semantic relations, and, in particular, the semantic similarity among concepts.

Current feature-based semantic memory models assume that the semantic representations of concepts differ systematically across living and nonliving categories and that such differences account for the emergence of category-specific semantic deficits in brain-damaged people. To assess some of the different models' main assumptions about structural differences at the semantic feature level in the two major semantic domains, we administrated a feature-listing task to normal young volunteers on 64 concepts drawn from living and nonliving semantic categories. We investigated whether feature correlation, a variable with a crucial role in the emergence of category-specific deficits, should be computed as a concept-dependent or as a concept-independent measure, and we chose the former. We also addressed the issue of a psychological counterpart of feature production frequency. Finally, we analysed the database obtained from the feature-listing tasks, looked at cross-domain differences for correlation, feature frequency, distinctiveness, and feature type, and discussed the implications of these findings for current semantic memory models.

No abstract