Corpora paper academic stance regarding AI

<b>Cautiously optimistic: A multidimensional Kullback-Leibler Divergence-based keyness analysis of academic stance on artificial intelligence in </b><b><i>The Conversation </i></b><b>[UNDER REVIEW]</b>Since artificial intelligence (AI) increasingly impacts educational policy, it is worth investigating how experts communicate its risks and potentials to a wider audience. This corpus-assisted discourse study (CADS) explores how academic authors construe a stance towards AI in articles published in <i>The Conversation</i>. Grounded in Systemic Functional Linguistics (SFL), the analysis focuses on the lexicogrammatical realisation of stance, particularly in relation to opportunities and ethical dilemmas associated with AI applications in education. Relevant stance markers, retrieved through semi-supervised automated retrieval and syntactic parsing with SpaCy, are identified employing a multidimensional keyword analysis that combines Kullback-Leibler Divergence (KLD) and Gries’ normalised dispersion measure (DPnorm). The findings reveal a frequent use of epistemic markers (e.g., <i>likely</i>, <i>could</i>) and attitudinal resources (e.g., <i>useful</i>, <i>difficult</i>) in academic news discourse, indicating a cautiously optimistic stance that recognises both the promise and pitfalls of AI tools and systems. Stance marker density was significantly higher in the target corpus than in a comparable general news dataset, which supports the interpretation of academic news as a hybrid sub-register that blends expert authority with public engagement. The results underscore the advisory function of academic news in influencing public understanding of emerging technologies.