Computing Semantic Change Using Context Volatility

In this paper we describe problems occurring while modeling the dynamics of terminology context in diachronic corpora and how to solve them. The foundation for this work is context volatility, a measurement for contextual change [4], that we use as our main measuring unit. The computation of context volatility for a word relies on the significance-values of its co-occurrent terms and the corresponding co-occurrence ranks. In the ongoing process its necessary to evaluate these ranks precisely. After applying context volatility according to [4] it can be shown that there are many cases for which a co-occurrence of two words at a specific point of time has no joint occurrence of its terms in other time stamps. This leaves gaps in the assigned ranks for the co-occurrences which must be handled accordingly. Just setting an arbitrary rank is not accompanied by a reasonable model. In this paper we present solutions and ideas to overcome this problem. We show that the use of very sparse term-term matrices leads to undesired results. We use corpus level statistics and a recommender system to recalculate frequencies and co-occurrence statistics on the time slices. Within such a setting we can use the measure context volatility in a way more consistent manner and we propose the idea for a well-defined statistical model based on the presented results.