New Method of Automated Terminology Extraction: Case Study of Russian-Language Textbooks

Abstract

AbstractIn this article, we report the initial results of our research into the ways different terminological subsystems are organised and function in Russian secondary school textbooks. Traditionally, extraction of terms from specialised texts has been done manually, by reading the texts through and compiling lists of terms manually. While reliable and clear with regard to selection criteria, the procedure is ill-suited for large volumes of data and does not yield additional metadata, e.g. frequency of terms’ occurrence and co-occurrence, their syntagmatic connections and systematic relations. This project involves (1) creating a full-text corpus of Russian school textbooks for years of study from 5 to 11 that have been included in the Federal List of the Ministry of Education of the Russian Federation, (2) automatic term extraction and stratification using computational linguistics tools aimed to analyse distributional semantics, and (3) training a deep neural network capable of identifying the academic subject, year of study and teaching topic given a particular set of terms. Our research results may be of significant interest for terminology studies and can also be used as an assistance in creating various types of teaching literature for schools. The reported study was funded by RFBR, project number 19-29-14032 mk.KeywordsTermTerminologyVector representationTextbookGeneral educationRussianCo-occurrenceNeural networkDeep learningModelWord2VecCBOWSkip-gramNgram