Theoretically Grounded AI Feedback for Adult Swedish Second Language Writers : Design, Implementation, and Preliminary Evaluation of a Language Examination Preparation Platform

Abstract

This thesis investigates to what extent a purpose-built artificial intelligence platform can provide theoretically grounded feedback mechanisms to support adults learning writing in Swedish as a second language for the National Certificate of Language Proficiency examination in Finland. It presents the design, implementation, and preliminary evaluation of a web-based platform that provides AI-generated written corrective feedback on examination-aligned writing tasks.

The platform was designed around seven theoretically grounded principles derived from second language acquisition research, feedback theory, and motivational psychology, and implements a structured feedback interaction requiring learners to attempt independent error correction before receiving metalinguistic explanations. The platform was evaluated through a heuristic usability inspection conducted by the researcher and two user experience design professionals, and an AI feedback validation procedure comparing the platform's corrections against expert human assessment.

The heuristic evaluation confirmed that the design principles were implemented without major usability barriers. The validation procedure achieved 92.1% category-level agreement and 83.3% instance-level agreement with expert human corrections, both exceeding the pre-set 80% threshold. Register detection was identified as the primary area requiring further calibration. The findings demonstrate that theoretically grounded AI feedback of sufficient quality for this examination context is technically achievable. The natural next step is to bring the platform to YKI candidates and evaluate whether the feedback translates into measurable improvement in writing performance.