Computer-assisted approaches to semantic maps. A qualitative approach to large-scale lexical datasets

A semantic map is a way to visually represent the relationships between meanings based on patterns of co-expression across languages (Georgakopoulos & Polis 2018). In this talk, we will focus on the so-called ‘classical maps’ (Haspelmath 2003; van der Auwera, 2013), which typically take the form of a graph — with nodes standing for meanings and edges between nodes standing for relationships between meanings. Up until recently, the classical maps were plotted manually and based on relatively small datasets. The very possibility of plotting such maps automatically has been questioned, because the model has been considered “not mathematically well- defined or computationally tractable, making it impossible to use with large and highly variable crosslinguistic datasets” (Croft & Pool 2008: 1). However, Regier et al. (2013) showed that an efficient algorithm exists in order to infer semantic maps “that approximates the optimal solution nearly as well as is theoretically possible.” The first goal of this talk is to show that the algorithm of Regier et al. (2013) produces high quality maps, but that these need to be revised and emended based on qualitative semantic analyses (a) when the connection suggested between two meanings is possible from a mathematical point of view, but at the same time an alternative connection is equally possible and semantically more satisfying, and (b) when historical data show that there is an indirect relationship between two meanings that appear directly related in synchronic datasets. In order to do so, we resort to CLICS2 (List et al. 2018; https://clics.clld.org), an online database of synchronic lexical associations that provides information about 2638 distinct colexification patterns in 1220 language varieties. As case studies, we explore the structured network of two semantic fields, the one of PERCEPTION/COGNITION and the one of EMOTIONS/VALUES, taking as a point of departure concepts We conclude by arguing that classical semantic maps can be combined with the formal concept lattices introduced by Ryzhova & Obiedkov (2017). Formal concept analysis produces hierarchical graphs, which visualize in a principled way the mapping of polysemic items onto meanings. This approach therefore allows one to interpret the universal semantic networks of the classical semantic maps (a kind of black- box when considered in isolation) from both a genetic and an areal viewpoint, since the clustering of polysemic items according to parameters such as lineage and geography can be straightforwardly observed.