Weaving the fabric of science: Dynamic network models of science's unfolding structure

This paper examines the structure of linguistic predications in English text. Identified by the copular “is-a” form, predications assert category membership (hypernymy) or equivalence (synonymy) between two words. Because predication expresses ontological structure, we hypothesize that networks of predications will form modular groups. To measure this, we introduce a semantically motivated measure of predication strength to weight relevant predications observed in text. Results show that predications do indeed form modular structures without any weighting (Q _ 0.6) and that using predication strength increases this modularity (Q _ 0.9) without discarding low-frequency items. This high level of modularity supports the networkbased analysis and the use of predication strength as a way to extract dense semantic clusters. Additionally, words’ centrality within communities exhibits slight correlation with hypernym depths in WordNet, underscoring the ontological organization of predication.

For more: Gerow, A., & Evans, J. (2014, October). The modular community structure of linguistic predication networks. In Proceedings of TextGraphs-9: the workshop on Graph-based Methods for Natural Language Processing (pp. 48-55).