A Frequency-Based Algorithm for Argument Extraction from Russian Treebanks

Arguments, unlike adjuncts, are typically understood as verb-specific dependents, which includes the fact that the morphosyntactic devices used for argument encoding are determined by individual verbs. Building on this observation, we operationalize arguments as dependents whose encoding device occurs with a given verb at a significantly higher-than-average frequency. We apply an argument extraction algorithm to a dataset of 132,221 verb dependents from Russian treebanks available in the Universal Dependencies (UD) platform. To evaluate the algorithm ’ s performance, we compare its results to a manually annotated subset, informed by The Active Dictionary and a detailed semantic understanding of argumenthood. The frequency-based algorithm achieves acceptable precision (approx. 0.83), with particularly few false positives, making it a promising tool for cross-linguistic applications in typologically diverse languages with UD treebanks. Theoretically, we argue that a quantitative distributional approach to valency—originally proposed in Ju. D. Apresjan ’ s early pioneering work—broadly aligns with the in-depth semantic analyses of individual verbs and their meanings found in his later works, including The Active Dictionary.