Using Topic Salience and Connotational Drifts to Detect Candidates to Semantic Change

Semantic change has mostly been studied by historical linguists and typically at the scale of centuries. Here we study semantic change at a finer-grained level, the decade, making use of recent newspaper corpora. We detect semantic change candidates by observing context shifts which can be triggered by topic salience or may be independent from it. To discriminate these phenomena with accuracy, we combine variation filters with a series of indices which enable building a coherent and flexible semantic change detection model. The indices include widely adaptable tools such as frequency counts, co-occurrence patterns and networks, ranks, as well as model-specific items such as a variability and cohesion measure and graphical representations. The research uses ACOM, a co-occurrence based geometrical model, which is an extension of the Semantic Atlas. Compared to other models of semantic representation, it allows for extremely detailed analysis and provides insight as to how connotational drift processes unfold.