Automatic grading of answers in knowledge control for “Definition” and “Description” question types

The progress in developing an effective automatic knowledge control system is directly associated with creating and implementing a software module for grading answers to test questions formulated in natural language. Previously, an experimental prototype of such a system was designed, and a study was performed where short answers to basic question types provided by students were examined via a pragmatically oriented question-answer text processing algorithm, considering its outputs and exposing flaws. This article introduces the next iteration of the algorithm tailored to handle more complex question types that requires the identification of relations such as “Definition” and “Description.” The key features of the enhanced algorithm were outlined, with a particular focus on the problem of segmenting answers into meaningful chunks, a task for which processing methods have already been found. The results of an experiment based on the developed prototype with obtaining answers from students and a thorough analysis of the instances of the system’s incorrect behavior were discussed.