Abstract:

This study exhaustively examines “actually” in the spoken data from the BNC in terms of its collocation variability across syntax, gender, domain and interactivity, based on which pragmatic functions are classified by applying the Recurrent Neural Network (RNN) model. It reveals which pragmatic classifier controls the collocation variability of pragmatic functions. It is found that the significance of syntax and domain controls is higher than that of gender and interactivity. The specific patterns of convergence between pragmatic classifiers and collocation variants revise the Subjectivity Intersubjectivity Peripheries Hypothesis and marginalizes the efficacy of interactional exigency. These enlighten a corpus-based solution to parameterizing the functionality of pragmatic inferences from units of meaning under the canopy of data-driven research to combine digital humanities with information technologies.

Key words: pragmatic particle, collocational variability, corpus-based, Recurrent Neural Network