Network measures: A new paradigm towards reliable novel word sense detection

Abstract

In this era of digitization, with the fast flow of information on the web, words are being used to denote newer meanings. Thus novel sense detection becomes a crucial and challenging task in order to build any natural language processing application which depends on the efficient semantic representation of words. With the recent availability of large amounts of digitized texts, automated analysis of language evolution has become possible. Given corpus from two different time periods, the main focus of our work is to detect the words evolved with a novel sense precisely. We pose this problem as a binary classification task to detect whether a new sense of a target word has emerged. This paper presents a unique proposal based on network features to improve the precision of this task of detecting emerged new sense of a target word. For a candidate word where a new sense has been detected by comparing the sense clusters induced at two different time periods, we further compare the network properties of the subgraphs induced from novel sense clusters across these two time periods. Using the mean fractional change in edge density, structural similarity and average path length as features in a Support Vector Machine (SVM) classifier, manual evaluation gives precision values of 0.86 and 0.74 for the task of new sense detection, when tested on 2 distinct time-point pairs, in comparison to the precision values in the range of 0.23-0.32, when the proposed scheme is not used. The outlined method can, therefore, be used as a new post-hoc step to improve the precision of novel word sense detection in a robust and reliable way where the underlying framework uses a graph structure. Another important observation is that even though our proposal is a post-hoc step, it can be used in isolation and that itself results in a very decent performance achieving a precision of 0.54-0.62. Finally, we also show that our method is able to detect well-known historical shifts in 80% cases.