.
Peer feedback is widely used in English as a foreign language (EFL) higher education, yet its benefits are often limited by uneven feedback quality and learners’ difficulty in interpreting and using comments. This theoretical paper synthesizes research on peer feedback, student feedback literacy, and recent developments in generative artificial intelligence (GenAI) to propose a theory-informed design framework that positions GenAI as Trainer and Synthesizer in L2 speaking peer feedback. Building on feedback literacy as a set of capacities (appreciating feedback, making judgments, managing affect, and taking action), the paper argues that speaking tasks create distinct constraints, including time pressure, fleeting performance, and heightened affect, which make real-time peer feedback promising but pedagogically challenging. To address these challenges, here we introduce two complementary roles for GenAI in peer feedback workflows: a Trainer that supports feedback quality through calibration with exemplars, rubric-guided practice, and feedback-on-feedback; and a Synthesizer that aggregates peer input into concise, actionable guidance linked to criteria and learning goals. The conceptual proposal specifies key design principles (e.g., transparency, learner agency, teacher-in-the-loop oversight, and privacy-conscious data practices) and outlines researchable propositions for evaluating learning, engagement, and equity outcomes. The paper concludes with implications for task design, training sequences, and responsible classroom implementation.
Counterfactual explanations (CEs) enhance the interpretability of machine learning models by identifying the smallest change to an input required to obtain a desired output. Although CEs are conventionally formulated as a distance-minimization problem, the theoretical basis of this formulation has received limited attention. We show that a distance-minimization-based CE is mathematically equivalent to the maximum a posteriori (MAP) estimate of a Gibbs posterior within the generalized Bayes framework, specifically when a distance-based prior is used. We call this formulation the Distance-Prior Generalized Bayes CE (DP-GBCE). Building on this posterior perspective, we introduce two decision rules beyond MAP within a unified framework: a Bayes decision that minimizes expected decision loss and CVaR-CE, a risk-averse decision rule. We also propose an extension that uses Bayesian model weights to mix the posterior distributions of multiple models, thereby accounting for model multiplicity, where several models have comparable predictive performance. Finally, we define metrics for evaluating both individual CEs and the posterior distribution as a whole, and use experiments on simulated data and Google Trends data to quantify the trade-offs among the decision rules.
