Effectiveness of combining adaptive prompts and peer assessment in developing sociolinguistic competence

Abstract

The article examines the effectiveness of combining adaptive prompts produced by generative models with peer assessment in developing sociolinguistic competence in academic writing among linguistics students. The focus is on sociolinguistic appropriateness of written discourse, understood as alignment with genre and addressee, observance of academic etiquette, informed register choice, management of categorical statements, and appropriate expression of disagreement. The empirical basis was a semester-long implementation of a hybrid feedback model in a face-to-face course: the experimental group (Year 2, B2) worked with AI-generated prompts and rubric-based peer assessment, while the control group (Year 1, B1) completed the same tasks without AI prompts. Changes were assessed using pre- and post-tests on a 25-point scale and an analysis of revisions across text versions. The average gain in the total score reached 3.6 points in the experimental group and 1.4 points in the control group; for the sociolinguistic appropriateness component the gains were 1.2 and 0.4 points, respectively. In the experimental group, around 40% of revisions targeted appropriateness, and the share of automatically accepted AI suggestions decreased from 78% to 56%. A comparison of YandexGPT and GigaChat showed a higher proportion of rubric-aligned recommendations for YandexGPT (74% vs. 61%). The findings support the practical value of combining AI prompts with peer assessment when a clear criterion framework and instructor moderation are maintained. Further research should expand the sample, align group proficiency levels, and test the durability of the effect across academic writing genres.