Using Text Vectorization to Analyze Students' Answers Formulated in Different Languages

The paper is devoted to studying the results of automated assessment of students' short answers (responses). Vector representations of the answers are obtained using language model. To evaluate the answers, cosine similarity was determined first between the student answer vector and the vector of correct (template) answer, then between the student answer vector and the vector that is the average of the vectors of right students' answers. The students' answers were translated into Tatar and evaluated using cosine similarity. The results of evaluating the answers formulated in different languages were studied.