Running Head Word Association Spaces 1 Word Association Spaces Semantic Spaces based on Free Association that Predict Memory Performance Send correspondence to
performance when predicting performance in various memory tasks. As with the SVD scaling
procedure, the number of dimensions was varied between 10 and 500.
Predicting Semantic Similarity Effects in Memory
Since Deese’s (1959b) classic study on intrusions in free recall, many studies have shown
that memory errors are in part based on semantic overlap between the response and the contents
of memory. In this research, we introduced WAS as a way of quantifying the semantic similarity
between words that might help in predicting these memory errors. The data from a recognition
experiment, Deese’s original free recall experiment and a cued recall experiment were taken as a
basis for testing various models that capture semantic similarity. We tested three WAS based
measures for semantic similarity. The first two were based on the SVD of S(1), the one step
symmetric associative strengths, and on the SVD of S(2), the one plus the two step associative
strengths involving indirect associative strengths. In these two semantic spaces (as well as in
LSA) the cosine of the angle between two words expresses the similarity between two words.
The last WAS measure was based on metric-MDS of the shortest path associative strengths. In
this space, the Euclidian distance between two word vectors is taken as a measure for the
dissimilarity between two words. These WAS scaling solutions were contrasted with the
(unscaled) associative strengths S(1) and S(2) that were taken as control comparisons. We also
tested two LSA based measures, one was based on a corpus of an encyclopedia and another on a
corpus called tasa that included reading material that an English reader might be exposed to from
3rd grade up to 1st year of college. The different methods that are compared are listed in Table 1. In a study by the first two authors (Steyvers & Shiffrin, submitted: Experiment 1), 89
subjects studied 144 words that contained 18 semantic categories of 5 words each. Based on a
热门文档