Pegi1727/Decolonizing-the-Algorithm-DAR-AI-Assisted-L2-Academic-Writing: Decolonizing the Algorithm (DAR) in AI-Assisted L2 Academic Writing: Epistemic Agency, Authorial Voice, and Pre–Post Research Data

Abstract

Decolonizing the Algorithm (DAR) in AI-Assisted L2 Academic Writing: Epistemic Agency, Authorial Voice, and Pre–Post Research Data Description This research project investigates the Decolonizing the Algorithm (DAR) framework, a critical pedagogical approach designed to support negotiated epistemic agency and authorial voice in generative artificial intelligence (GenAI)-assisted second-language (L2) academic writing. As large language models become increasingly integrated into academic writing and language education, they offer opportunities to improve linguistic accuracy, facilitate revision, and expand access to writing support. However, their use also raises important questions concerning linguistic standardization, authorship, cultural representation, epistemic authority, and the potential reproduction of dominant academic norms. The DAR framework approaches AI-assisted writing as a process of critical engagement and negotiation rather than the passive acceptance of AI-generated text. It consists of four pedagogical stages: (1) Critical Deconstruction, in which writers examine AI-generated language, assumptions, and possible changes in meaning; (2) Epistemic Disobedience, in which writers question or reject suggestions that conflict with their knowledge, intentions, or contextual commitments; (3) Prompt Negotiation, in which writers refine instructions and articulate their rhetorical and contextual requirements; and (4) Reflective Authorial Review, in which writers evaluate whether the resulting text preserves their intended stance, meaning, and authorial position. The study uses a pre–post research design involving 70 participants to examine changes associated with participation in a DAR-informed pedagogical intervention. Seven measures are investigated: prompting density, mean prompt length, AI-suggestion rejection rate, stance marking, lexical agency, voice consistency, and critical dialogue. The first three measures represent selected aspects of participants' interaction with AI, while the remaining four operationalize dimensions of authorial voice assessed in the study. The quantitative analysis uses descriptive statistics and Wilcoxon signed-rank tests to compare pre-intervention and post-intervention measurements. The reported results indicate statistically significant pre–post differences across all seven measures (p < .001), with higher post-intervention scores. Reported effect sizes are large (r = .75–.87). All 70 participants showed increases in prompting density, mean prompt length, and AI-suggestion rejection rate, whereas individual patterns of change varied across the four authorial-voice dimensions. These findings are consistent with the proposition that a structured critical approach to AI-assisted writing can be associated with changes in interactional practices and assessed dimensions of authorial voice. Nevertheless, the measured indicators are proxies rather than direct or exhaustive measures of epistemic agency. Increased rejection of AI suggestions does not automatically demonstrate critical reasoning, and changes in rubric scores do not independently establish deeper epistemic awareness or resistance to structural linguistic inequalities. The project contributes to research at the intersection of applied linguistics, L2 academic writing, critical AI literacy, decolonial perspectives, and educational technology. Its central contribution is the integration of a decolonial pedagogical rationale with multiple interactional and authorial-voice measures in a pre–post empirical study. It also offers a basis for further investigation into how writers evaluate AI-generated language, negotiate rhetorical choices, and retain responsibility for the knowledge expressed in academic texts. The findings should be interpreted in light of the study design. Because the research uses a pre–post design without a comparison group, the observed differences do not independently establish that the DAR intervention caused the changes. Further research should incorporate comparison groups, qualitative analysis of writing and AI-interaction processes, longitudinal designs, and replication across different linguistic, educational, and institutional contexts. This Zenodo record is intended to support research transparency, scholarly discussion, and future investigation of critical pedagogical approaches to generative AI in academic writing. The framework may be relevant to researchers and educators in applied linguistics, English for Academic Purposes (EAP), second-language writing, AI literacy, educational technology, and decolonial approaches to knowledge production. Keywords: Decolonizing the Algorithm; DAR framework; generative artificial intelligence; large language models; L2 academic writing; English for Academic Purposes; authorial voice; epistemic agency; critical AI literacy; linguistic justice; AI-assisted writing; academic discourse; decolonial pedagogy; prompting behavior; higher education. Research design: Pre–post study Sample size: 70 participants Analytical approach: Descriptive statistics and Wilcoxon signed-rank tests Primary focus: Critical engagement with AI, negotiated epistemic agency, and authorial voice in L2 academic writing.