The paper considers systems of automatic verification of open answers in natural language and describes the requirements for such systems. The article proposes a functional scheme of such a system and discusses its implementation. The Tomita-parser by Yandex for extracting information from texts was chosen as a linguistic processor of the developed system. Grammatical rules for extracting entities from texts in Russian are written and an algorithm for analyzing the answers is proposed. Weights can be set for each entity and for each of the available reference responses when evaluating each response, allowing to customize the system. The system is implemented and tested by checking students’ answers.