Diachronic discourse structure and typology with time series models

There is long-standing interest in how discourse fonns, functions, and use change over time. However, different approaches to time-based or diachronic discourse analysis face collective issues with i) connecting temporal to discourse variables in clear and replicable ways, ii) handling complex change patterns in more spontaneous discourse contexts over shorter time frames, and iii) limiting the analysis of change to a small number of time intervals, without considering possible interdependences or autocorrelations between them. This project applies and evaluates how Time Series Analysis methods (TSA) can address these issues and develop novel conceptualizations of discourse across time. TSA is widely used in finance and engineering to understand naturally occurring variables like stock prices in terms of their key components (trends, seasons, cycles, and irregular fluctuations) and autocorrelation structure. This information is used to build a suitable time series model that reveals how successive observations are mutually related, and forecasts future values based on historical values. This project will i) show that the frequencies of discourse phenomena across contexts and time scales (e. g. metaphors in psychotherapy sessions, thematic keywords in newspapers, or pronouns in social media) resemble canonical time series, and can thus be analyzed with TSA methodology; ii) apply the Box-Jenkins TSA methodology to multiple discourse datasets and critically evaluate how well the models fit the data; iii) interpret the models in ways that draw upon and enrich qualitative analyses of discourse content and structure, and iv) develop a novel discourse typology using time series model types as a basis of categorization. The project advances discourse analytic methodology through an approach established in empirical social sciences but less familiar to discourse analysts, directly addressing the methodological issues outlined above. It also makes theoretical contributions by combining quantitative modelling with qualitative interpretation of discourse and developing an innovative classification of diachronic discourse types for future research purposes. The specific discourse datasets to be analyzed, covering a range of contexts (e. g. psychotherapy, classroom discourse, newspaper discourse, political speeches, social media) and phenomena (e. g. metaphors, pronouns, non-inforrnational markers, thematic keywords, rhetorical questions) will furtherrnore address specific issues of interest in their respective areas.