TY - GEN

T1 - Russian Neural Morpheme Segmentation: From Lemmata to Wordforms

AU - Morozov, Dmitry

AU - Shcherbakova, Olga

AU - Glazkova, Anna

N1 - Conference code: XXVIII

PY - 2026

Y1 - 2026

N2 - Among the algorithms for automated morpheme segmentation for the Russian language, the best-performing ones are based on machine learning. The quality of these algorithms is currently close to expert-level. However, most research focuses on the analysis of lemmata rather than word forms. In this study, we compare state-of-the-art methods for morpheme segmentation using a prepared dataset of word forms. We evaluate three approaches: an ensemble of convolutional neural networks, a subword Transformer model DeepSPIN-3, and RuRoberta-based Morphberta models. To assess the robustness of these models, we employed multiple strategies for splitting the dataset into training and test sets, specifically to examine how performance degrades when handling out-ofvocabulary lemmata and roots. The best results were achieved using Morphberta models (over 99.5% completely accurate segmentations). However, our findings also demonstrate that random dataset splitting does not provide a comprehensive understanding of quality of the algorithm. Specifically, when dealing without-ofvocabulary morphemes, segmentation accuracy significantly declines, with the extent and nature of the decline varying across algorithms.

AB - Among the algorithms for automated morpheme segmentation for the Russian language, the best-performing ones are based on machine learning. The quality of these algorithms is currently close to expert-level. However, most research focuses on the analysis of lemmata rather than word forms. In this study, we compare state-of-the-art methods for morpheme segmentation using a prepared dataset of word forms. We evaluate three approaches: an ensemble of convolutional neural networks, a subword Transformer model DeepSPIN-3, and RuRoberta-based Morphberta models. To assess the robustness of these models, we employed multiple strategies for splitting the dataset into training and test sets, specifically to examine how performance degrades when handling out-ofvocabulary lemmata and roots. The best results were achieved using Morphberta models (over 99.5% completely accurate segmentations). However, our findings also demonstrate that random dataset splitting does not provide a comprehensive understanding of quality of the algorithm. Specifically, when dealing without-ofvocabulary morphemes, segmentation accuracy significantly declines, with the extent and nature of the decline varying across algorithms.

KW - Deep Learning

KW - Morpheme Segmentation

KW - Russian Language

UR - https://www.scopus.com/pages/publications/105035222502

UR - https://www.mendeley.com/catalogue/20d51f40-d8a4-3d49-abde-cfda9454077c/

U2 - 10.1007/978-3-032-04958-2_12

DO - 10.1007/978-3-032-04958-2_12

M3 - Conference contribution

SN - 9783032049575

T3 - Communications in Computer and Information Science

SP - 157

EP - 167

BT - Internet and Modern Society

PB - Springer

T2 - Международная объединённая научная конференция «Интернет и современное общество» (Internet and Modern Society – IMS-2025)

Y2 - 23 June 2025 through 25 June 2025

ER -