Both the path lengths and the clustering coefficients of the LSA networks show a decreasing
trend as the dimensionality d is increased. It is possible that at dimensionalities
higher than 400, these statistics will come closer to the values observed for the word association
network. We were not able to investigate this possibility, as only 400 LSA dimensions
were available to us. However, it is unlikely that increasing the dimensionality
would threaten our main argument, because the lack of scale-free degree
distributions is the primary feature distinguishing the LSA networks from naturally occurring
semantic networks and our growing network models. Based on Fig. 7, it seems
doubtful that these distributions would match significantly better at higher dimensionalities
(unless perhaps d was increased far beyond the typical value of 300).