Vector Space Models as Non-narrative Descriptions of Time

• In Event: 038 - Section on Methodology Paper Session. Advancement in Observing and Modeling Social Processes

Abstract

Narrative conceptions of time and history are subject to the problem of indeterminacy of the past (Danto 1985). Past work in event strucutre analysis has dealt with the consequences of this problem for the question of casing a given historical sequence (Bearman, Faris and Moody, 1999). This paper proposes a complementary method that requires no prior knowledge of an historical sequence or its narrative representations, challenging our reliance on interpretation in the production of data about social processes. I argue that the proposed method provides a strategy for the modeling of social processes that is not subject to the limitations of narrative knowledge, particularly those related to the ad-hoc nature of its results and the inherent bias introduced by eventful conceptions of time. This is accomplished through the deployment of vectorial semantic models and the construction of second order units of analysis from the induction of semantic networks for the representations of different points in time. The paper presents a thorough technical description of the process and its theoretical underpinnings, and it demonstrates it using data from a text corpus covering a 240 year period.

Author

• Jose Tomas Atria, Columbia University