Abstract:

In the conditions of the development of the knowledge economy, it seems relevant and significant to us the research aimed at creating a service of complex machine analysis of educational texts, which involves the identification of their essential semantic-thematic and lexico-grammatical characteristics, the assessment of complexity for the student and subsequent adaptation on the basis of artificial intelligence algorithms, taking into account the factor of the addressee - a child-foreigner. The realization of the present scientific project is ultimately aimed at solving a significant social problem of teaching in educational institutions of different levels of native speakers of Russian and foreignophones, whose level of Russian language proficiency often varies from zero to sufficient for everyday communication on everyday topics, but insufficient for education. In our opinion, the study of disciplines of the social and humanitarian cycle by inophones is particularly difficult. It is connected with three reasons. The first two are formal in nature: the need to master a large amount of textual material and the absence of the possibility to convey information non-verbally, in the form of formulas, schemes and other things. The third reason is of a substantive nature: unlike the exact and natural sciences, the disciplines of the social and humanitarian cycle are not international, their content, as well as the interpretation and evaluation of facts are conditioned by cultural and historical specifics, and, accordingly, their successful mastering requires additional background knowledge related to the country of residence and study.

The mismatch of educational texts with the cognitive and linguistic abilities of part of the target audience leads to the complication of the learning process, which concerns all parties involved in it: both the learner, whose level of language proficiency does not correspond to the complexity of the texts studied, and the teacher, who has to adapt texts intuitively, based on approximate ideas about the linguistic and cognitive abilities of the learner.

This requires text adaptation both from the content point of view (extracting the most valuable information from the text, allowing to get a sufficient understanding of the text content in a short time) and from the linguistic point of view (transformation of grammatical constructions difficult for perception, synonymic replacement of unfamiliar vocabulary, interpretation of words that cannot be replaced, for example, terms). These processes can be automated (which, however, does not exclude human participation), but there are no tools for complex analysis and adaptation of texts taking into account the addressee-nonophone factor, which is the scientific problem the project is aimed at solving.

The fundamental scientific task of this research project is to study the possibilities of automatic processing of educational texts from the point of view of:
a) identifying difficulties in the perception of Russian-language educational text by foreign-speaking children associated with the use of unfamiliar vocabulary and grammatical constructions;
b) adapting educational texts for use in teaching foreign-speaking children;
c) further use in specialized literature.

Keywords:

Artificial intelligence, recurrent neural networks, generative-adversarial neural networks (GANs), transformers, text analysis, text adaptation, machine learning linguodidactics.