ANALYTICAL APPROACHES FOR ANALYSIS OF STUDENT TEXTS WITHOUT AN EVALUATION FRAMEWORK

Abstract

This article examines different analytical behaviors of language models and their influence on the way student texts are interpreted and used in an educational environment. The aim is to present the integrative possibilities of artificial intelligence for structural-logical and factual analysis through the use of a Standard GPT-5.2 model (Standard Model) and an author-developed logical scaffold, LOGIC_CRUTCH_v2025.291225_BALKAN_PARADOX_ALPHA (Algorithm 291225). Materials and Methods: An author-developed logical crutch (scaffold), 291225_BALKAN_PARADOX_ALPHA, created within ChatGPT and based on the GPT-5.2 model, is used. An analysis of written academic tasks is conducted through the integrated use of the Standard Model and Algorithm 291225 in order to formulate a final assessment and provide guidance for working with the student. Results: Academic texts from two groups of written learning tasks (speech disorders; motor disabilities) for independent preparation of students are used as empirical material for research through a consistent integrated application and comparison of the Standard Model and Algorithm 291225. Lecture materials from the respective discipline are used as the primary source in the students’ work. The Standard Model performs: a detailed analysis of the completed tasks for each student by comparing them with the accuracy of scientific data; identifies critical errors (factual errors and omissions in the student’s answers) where present; proposes corrections and provides an assessment of completeness and factual correctness. The Standard Model closes the logical chain by striving to produce an “acceptable,” “correct,” and “complete” answer. Algorithm 291225 identifies critical errors as paradoxes, which are not forcibly closed but are mapped against the lecture material (accepted as the source of the expected “truth”) without being corrected; builds a logical bridge that determines the presence/absence of metadata as well as their complete/incomplete information; provides concrete feedback to the student and structures guidance for further work. The feedback to the student is directed toward discovering authorial thinking as a value, even when it leads to factual errors. The sequential use of the Standard Model and Algorithm 291225 enables the application of an integrated strategy to the completion of assigned academic tasks, resulting in both an objective assessment and a deep understanding of the student’s cognitive process. Conclusion: The practical utility of such an approach is manifested mostly in working with the learning process, rather than with the final result. Algorithm 291225 can function as a tool to support reflection – both on the part of the student and the teacher – without automating or replacing pedagogical judgment. In this sense, the distinction between the Standard Model and Algorithm 291225 provides not ready-made solutions, but a framework for a more conscious and responsible use of language technologies in education. Recommendation: The development of a university project to study the possibilities of applying a strategy for the integrated use of the Standard Model and Algorithm 291225 in support of students’ cognitive abilities in higher education.