Underlying dimensions in conceptual change

KU Leuven (Belgium)

Objectives: Distributional semantic methods offer a powerful tool for analysing changes in the textual representation of lexically identifiable topics: an automated analysis of the changing contexts in which words like ‘migrant’ or ‘democracy’ appear reveals shifts in the conceptual associations of these concepts. Such a distributional method can be greatly enhanced by doubling it up upon itself, i.e. by distributionally identifying the similarity among context words associated with the target items, and subsequently analysing the latter in terms of the emergent underlying dimensions. Bách will: 1) apply and refine state-of-the-art vector space modelling and other distributional semantic methods to diachronic datasets, with a focus on higher-order collocates; 2) document examples of semantic change and their distributional dimensions; 3) form a general hypothesis about the distributional dimensions of conceptual change; 4) recommend best practice workflow for similar applications.  Bách will have access to the same datasets as Ángela (ESR 3), and to EEBO-TCP. 

Expected Results:
1) Innovation in (and evaluation of) distributional methods that ‘double up on themselves’ as they measure conceptual change.
2) A new computational workflow for analysing underlying distributional dimensions in relation to conceptual change. They will present research at D7.2, D7.4. 

Planned secondments: University of Sheffield (2 months, Year 3; to refine visualisation and workflow design, and to collaborate on lexicalisation article, D5.2).