Running Head Word Association Spaces 1 Word Association Spaces Semantic Spaces based on Free Association that Predict Memory Performance Send correspondence to
any other word in the graph (in fact, the maximum number of steps between any pair of words is of the associative strengths along the shortest path in the network defined by S(1). This is
equivalent to the (negative) sum of the logs of the associative strengths along the shortest path:
(1)(1)(1)(1)(1)(1))= logSikTij= log(SikSkl Sqj+logSkl+...+logSqj[]
Here, the shortest path between words i and j is from i to k to l through other words to q
and finally j. With this distance measure, word pairs with weak or long associative paths are
assigned large distances whereas word pairs with short or strong associative paths are assigned
small distances. The distances Tij were calculated for all word pairs in the word association
database. Then, these distances were scaled by metric-MDS. The result is that the words are
placed in a multidimensional space and the dissimilarity or distance between two words is
expressed by the Euclidian distance between the two corresponding word vectors:
2 distance(i,j)= ∑(Xik Xjk) k 1/2
Because of computational constraints, it was not possible to apply metric-MDS to the full
matrix T containing the distances for all word pairs. Instead, we chose 2500 words from the
original 5018 words in the word association database. The words in this smaller set included
words appearing in various experiments listed in the next section and included a selection of
randomly chosen words from the original set. The SVD procedure was applied on both the
smaller matrix of 2500 words as well as the set of 5018 words.
As with the SVD procedure, the choice of the number of dimensions in the space is
important. Having too few dimensions or too many dimensions might lead to suboptimal
热门文档