Indicator Patterns as Features for Argument Mining

The paper discusses methods for creating dictionaries of indicators represented by specific speech constructions that signal the presence of an argument in a text written in Russian. The indicators are described using lexical-grammatical patterns. The initial filling of the dictionary is carried out by an expert based on the analysis of a non-annotated collection of popular science articles. The work of the expert is supported by software tools for multi-purpose study of lexical characteristics of text. Significant attention is paid to the replenishment of indicator dictionaries by forming samples with variables and implementing iterative search in collections of popular science and scientific texts, in order to enrich the lexical composition of the constructed patterns.