Visualising Conceptual Change through Discourse: A graph – based strategy
Bach Phan-Tat
Abstract
Lexical Concept Modelling (LCM), developed within the Linguistic DNA project, identifies statistically ranked constellations of non-adjacent co-occurring lemmas across large discursive windows. However, the scale and complexity of its output present challenges for exploration and interpretation. This technical report proposes a graph-based visualisation strategy for examining LCM constellations and their variation across historical periods. Using plane as a demonstration case in the Corpus of Historical American English (COHA) from the 1890s to the 1960s, we first examine embedding-based visualisations and identify their limitations in representing the co-occurrence relations captured by LCM. We then introduce a weighted graph representation in which context lemmas form nodes and their co-occurrence within selected trios forms weighted edges. Frequency and pointwise mutual information (PMI) bands support flexible filtering of these relations. The approach combines global network visualisation with Seed Evolution, a local diachronic view centred on a selected context lemma. Using angle as a seed, we illustrate variation in co-occurrence configurations across geometric, scientific and aircraft-related contexts. Rather than automatically identifying discursive concepts or establishing conceptual change, the proposed strategy provides an exploratory framework for generating hypotheses, inspecting candidate discursive relationships and tracing them back to their original textual contexts.