Mining culture from professional discourse: a lexicon-based hybrid method

Abstract

Professional discourse is a culturally situated construct. However, due to methodological constrains, culture as a discourse determinant revealed by analytical traditions cannot easily be generalized. By integrating culture representative word stems generated by a bag-of-word (BoW) model into a latent Dirichlet allocation (LDA) model, this paper explored a lexicon-based hybrid approach to mining corporate cultures from MD&As. Two separate corpora, the corporation culture corpus sized 801,219 tokens and the MD&A corpus with 2,199,109 tokens, were used. Key word stems drawn from the corporate culture corpus by the BoW showed a strong capacity of distinguishing implicit corporate values between texts, while the LDA model applied to the MD&A corpus succeeded in identifying three types of cultures in MD&As. Case studies confirmed the credibility of the classification. The hybrid method extends the literature by providing a quantitative tool for the qualitative study of culture in professional discourse, and shows implications for what interdiscursivity in MD&As looks like and how it works to formulate a renewable cycle in discourse realization.