Analysing Speech Acts in Organisational Communication

Abstract

The research presented in this paper addresses the challenges of analysing organisational stakeholder communication using speech act theory, supported by a novel analytical model. We present a large, automatically annotated dataset focused on present- and future-oriented speech acts in Swedish organisational communication, specifically with regard to the information security standard ISO/IEC 27001. To construct and evaluate the dataset, we employ a hybrid annotation pipeline combining quantised large language models, retrieval-augmented prompting, and a finetuned multilingual XLM-RoBERTa classifier, which is evaluated against a manually annotated gold standard. The classifier is subsequently applied to a large corpus of corporate texts from Swedish companies’ websites, enabling comparative analysis between general corporate communication and ISO-focused discourse. The results show that ISO/IEC 27001 communication is dominated by assertive speech acts, with commissives playing a secondary role, reflecting an emphasis on factual compliance rather than aspirational or promotional claims. Beyond its empirical findings, the study contributes a reusable dataset and classifier to the CLARIN infrastructure, demonstrates how CLARIN K-centres can support social science research, and provides insights into the use of quantised large language models for speech act classification in mid-resource language settings.