Innovation in Language Learning

Critical Discourse Analysis (CDA) was developed to read texts as sites where power and ideology play out, on the implicit assumption that every text has a locatable human author; generative AI breaks this assumption, since much of the text we now read in classrooms, workplaces, and policy documents has no single author, no clear intention, and no transparent provenance. Scholars have framed this problem in two inadequate ways: some treat large language models (LLMs) as speakers, which is misleading because a model matches statistical patterns rather than communicating, while others treat LLMs as passive tools, which hides the chain of corporate decisions shaping what the model produces. This study proposes a third reading: an LLM is best understood as a Mediated Discursive Operator (MDO), a corporately owned statistical engine whose outputs intervene in human discourse at scale on terms set by three upstream choices, namely training data, alignment through human feedback, and platform deployment. Placed in dialogue with platform studies, actor-network theory, critical AI ethics, and data-colonialism theory, the thesis yields three points: agency is real but unevenly distributed; bias claims require a specified scope to be testable; and serious analysis must follow the corporate value chain rather than stop at the prompt-output interface. These are operationalized as a five-axis reporting framework, the CDA-GenAI Critical Agenda (CGCA), illustrated through a worked example and grounded in Systemic Functional Linguistics, the Discourse-Historical Approach, and multimodal CDA. For language teaching and learning, the framework gives educators and learners a concrete tool for reading AI-generated texts critically and is offered as a starting point for field testing in applied linguistics, with particular attention to Arabic and MENA contexts, not as a finished standard.

Keywords: algorithmic mediation; applied linguistics; critical discourse analysis; generative AI; large language models; reporting framework.

REFERENCES

[1] Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots. In Proceedings of FAccT ’21 (pp. 610–623). ACM.

[2] Crawford, K. (2021). Atlas of AI: Power, politics, and the planetary costs of artificial intelligence. Yale University Press.

[3] Couldry, N., & Mejias, U. A. (2019). The costs of connection: How data is colonizing human life. Stanford University Press.