This is a brainstorm.
Using measures of first and second order information, it is possible to get valuable distributional information regarding the meanings of words from a small corpus.
Potentially apply these measures to books, poems, or ask human subjects to talk about a given topic for a while. Measure first and second order statistics on small corpus. Yields two nxn matrices, where n=number of words. Maybe preprocessing, maybe not. If we want to use the phrase 'semantic network', this is what I am referring to from now on.
Compare matrices from one corpus to matrices from another corpus using subgraph isomorphism techniques. Overlap may be in the form of a given click - how well matched are two clicks. Overlap may be in the number of clicks. Overlap may be in the amount of the network subsumed in clicks. Use set theoretic measures for guidance and various applications.
Potential outcomes. May be able to identify metaphors in literature and poetry. May be able to identify where students lack understanding on a given topic. May be able to predict the degree to which two human subjects agree or are able to communicate on given topics.