Running Head Word Association Spaces 1 Word Association Spaces Semantic Spaces based on Free Association that Predict Memory Performance Send correspondence to
We applied SVD separately on these two measures of associative strength. The result of
each SVD is the placement of words in a high dimensional space, so that words that have similar
associative structures are represented by similar vectors. Because of the SVD method, and based
on work in LSA (see Derweester et al., 1990), a suitable measure for the similarity between two
vwords is the cosine of the angle between two word vectors. Let Xirepresent the vector in WAS
for word i. The similarity between words i and j is calculated by:
vvXi Xjsimilarity(i,j)=cos(α)= XiXj
vvvX Xjrepresents the inner product between where X is the length of the vector and i
vectors i and j. Two words that are similar in meaning or that have similar associative structures
are expected to have high similarity as defined by the cosine of the angle between the two word
vectors. The SVD of the associative strengths can uncover the latent relationships between
words. In the SVD of S(1), words that are not direct associates of each other can still be
represented by similar vectors if their associates are related. In the SVD of S(2), words that not
directly associated or indirectly associated through one intermediate associate, can still be
represented by similar vectors if the associates of the associates of the words are related. In other
words, the whole pattern of direct and indirect correlations between associations is taken into
account when placing words in the semantic space.
An important variable (which we will call k) is the number of dimensions of the space.
One can think of k as the number of feature values for the words. We varied k between 10 and
500. The number of dimensions will determine how much the information of the free association
database is compressed4. With too few dimensions, the similarity structure of the resulting
vectors does not capture enough detail of the original associative structure in the database. With
热门文档