Running Head Word Association Spaces 1 Word Association Spaces Semantic Spaces based on Free Association that Predict Memory Performance Send correspondence to
too many dimensions or the number of dimensions approaching the number of cues, the
information in the norms is not compressed enough so that we might expect that the similarity
structure of the vectors does not capture enough of the indirect relationships in the associations
between words. In the analyses of predicting performance in a variety of tasks (recognition, free
and cued recall), we will show that although the optimal value of k depends on the specific task,
intermediate values of k between 200 and 500 are appropriate for this method. An interesting comparison for the two WAS spaces based on SVD would be to construct
a metric space in which the distance between two words, i.e., their dissimilarity, can be measured
by the Euclidian distance between their vectors. Metric MDS is a classic method for placing
stimuli in a space such that the Euclidian distance between points in the space approximates the
Euclidian distances in the dissimilarity matrix (see Torgeson, 1952; Schiffman, Reynolds &
Young, 1981). In order to apply metric MDS, estimates are needed for the distance between any
two words. In fact, all non-diagonal entries in the matrix have to be filled with some estimate for
the distance between words since no missing values are allowed in the method5. This raises the
problem how to estimate the distance between i and j when the associative strength as measured
by Sij(1) is zero6.
In our solution of this problem, we were inspired by network models for proximity data
(e.g. Cooke, Durso, & Schvaneveldt, 1986; Klauer, & Carroll, 1995). In these network models,
dissimilarity between two stimuli is calculated by the shortest path between two nodes in a
graph. In this research, we can use the word association norms as defining a graph: two words
are linked by an edge if they have nonzero associative strengths. We will use the symmetric S(1)
associative strengths because in the graph defined by S(1), it is possible to reach any word from
热门文档