informИнформатикаInformatics1816-03012617-6963UIIP NASB10.37661/1816-0301-2026-23-1-88-104inform-1384Research ArticleОБРАБОТКА СИГНАЛОВ, ИЗОБРАЖЕНИЙ, РЕЧИ, ТЕКСТА И РАСПОЗНАВАНИЕ ОБРАЗОВSIGNAL, IMAGE, SPEECH, TEXT PROCESSING AND PATTERN RECOGNITIONТональные пространства векторных языковых моделейTonal spaces of vector language modelsЧерниковК. М.ChernikovKirill M.

Черников Кирилл Михайлович, студент факультета технологий искусственного интеллекта

Кронверкский пр., 49А, Санкт-Петербург, 197101

Kirill M. Chernikov, Undergraduate of Department of Artificial Intelligence Technologies

av. Kronverkskiy, 49A, Saint Petersburg, 197101

kmchernikov@itmo.ruСуровИ. А.SurovIlya A.

Суров Илья Алексеевич, кандидат физико-математических наук, доцент, старший научный сотрудник факультета технологий искусственного интеллекта

Кронверкский пр., 49А, Санкт-Петербург, 197101

Ilya A. Surov, Cand. Sci. (Phys.-Math.), Assoc. Prof., Senior Researcher of Department of Artificial Intelligence Technologies

av. Kronverkskiy, 49A, Saint Petersburg, 197101

ilya.a.surov@itmo.ruУниверситет ИТМОITMO University20262703202623188104Copyright © Черников К.М., Суров И.А., 20262026Черников К.М., Суров И.А.Chernikov K.M., Surov I.A.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://inf.grid.by/jour/article/view/1384Цели

Цели. Тональность как положительно-отрицательное настроение – ключевой параметр алгоритмов анализа и генерации текстов. Существующие методы машинного обучения кодируют ее вычислительно трудоемким и неинтерпретируемым образом, что затрудняет развитие этого направления. Целью работы является решение данных проблем для русского языка.

Методы

Методы. Используются предобученные векторные языковые модели, кодирующие слова векторами в многомерных пространствах. В таких пространствах тональность соответствует направлению, наилучшим образом разделяющему положительный и отрицательный прототипы. Тональность слова определяется его проекцией на это направление. Добавление тонального вектора к ключевому слову задает одномерное пространство, позволяющее находить в языковой модели его положительные и отрицательные ассоциации.

Результаты

Результаты. Алгоритм апробирован на машинных моделях семейств GloVe и FastText, кодирующих отдельные слова и морфемы русского языка векторами в 300-мерном пространстве. В качестве ключевых слов использовались частоупотребимые глаголы и существительные. Средняя достоверность найденных тональных ассоциаций составила 80 %.

Заключение

Заключение. Результаты свидетельствуют о применимости предобученных векторных языковых моделей для быстрой и интерпретируемой работы с тональной информацией. Разработанный подход востребован в задачах объектно-ориентированного сентимент-анализа, а также в задачах машинной генерации объектно-ориентированных текстов нужной тональности. Обобщение тональной оси до тройки семантических факторов Осгуда позволяет расширить представленный метод для работы с полным спектром эмоционально-смысловой информации.

Objectives

Objectives. Tonality as a positive-negative mood is a key parameter of text analysis and generation algorithms. Existing machine learning methods encode tonality in computationally extensive and uninterpretable ways which hampers the development of the corresponding applications. The work aims to solve this problem for Russian language.

Methods

Methods. Pre-trained vector language models are used to encode words as vectors in multidimensional spaces. In these spaces, the tonality corresponds to a specific direction which optimally discriminates the positive and negative prototypes. The tonality of word is then determined by its projection onto this direction. Adding a tonal vector to a key word defines a one-dimensional subspace, containing its positive and negative associations.

Results

Results. The algorithm is tested on GloVe and FastText machine language models, encoding individual Russian words and morphemes with vectors in 300-dimensional space. Commonly used verbs and nouns served as key words. The average reliability of the found tonal associations estimates as 80 %.

Conclusion

Conclusion. The results indicate the applicability of pre-trained vector language models for fast and interpretable working with tonal information. The developed approach is applicable for the tasks of aspect-based sentiment analysis, as well as for the machine generation of object-oriented texts with a required tonality. Generalization of the tonal axis to the triple of Osgood's semantic factors allows expanding the method to a full range of affectively-semantic information.

векторные языковые моделисентимент-анализтональностьсемантикапространствогенерация текстовvector language modelssentiment analysistonalitysemanticsspaceaffective text generationИсследование выполнено за счет гранта Российского научного фонда № 23-7101046, https://rscf.ru/project/23-71-01046/ (код ГРНТИ 28.23.29).The research was funded by the Russian Science Foundation grant No. 23-71-01046, https://rscf.ru/project/23-71-01046/ (GRNTI code 28.23.29).ReferencesClynes, M. The communication of emotion: Theory of sentics / M. Clynes ; ed.: R. Plutchik, H. Kellerman // Theories of Emotion. – Academic Press, 1980. – Р. 271–301. – https://doi.org/10.1016/B978-0-12-558701-3.50017-X.Clynes M. The communication of emotion: Theory of sentics. Theories of Emotion. In R. Plutchik, H. Kellerman (eds.). Academic Press, 1980, рр. 271–301. https://doi.org/10.1016/B978-0-12-558701-3.50017-X.Smetanin, S. The applications of sentiment analysis for Russian language texts: Current challenges and future perspectives / S. Smetanin // IEEE Access. – 2020. – Vol. 8. – Р. 110693–110719. – https://doi.org/10.1109/access.2020.3002215.Smetanin S. The applications of sentiment analysis for Russian language texts: Current challenges and future perspectives. IEEE Access, 2020, vol. 8, рр. 110693–110719. https://doi.org/10.1109/access.2020.3002215.Brauwers, G. A survey on aspect-based sentiment classification / G. Brauwers, F. Frasincar // ACM Computing Surveys. – 2021. – Vol. 55. – Р. 1–37. – https://doi.org/10.1145/3503044.Brauwers G., Frasincar F. A survey on aspect-based sentiment classification. ACM Computing Surveys, 2021, vol. 55, рр. 1–37. https://doi.org/10.1145/3503044.A summary of aspect-based sentiment analysis / S. Fan, J. Yao, Y. Sun, Y. Zhan // Journal of Physics: Conference Series. – 2020. – Vol. 1624, no. 2. – URL: https://iopscience.iop.org/article/10.1088/1742-6596/1624/2/022051 (date of access: 07.12.2025). – https://doi.org/10.1088/1742-6596/1624/2/022051.Fan S., Yao J., Sun Y., Zhan Y. A summary of aspect-based sentiment analysis. Journal of Physics: Conference Series, 2020, vol. 1624, no. 2. Available at: https://iopscience.iop.org/article/10.1088/1742-6596/1624/2/022051 (accessed 07.12.2025). https://doi.org/10.1088/1742-6596/1624/2/022051.Tang, D. Deep learning for sentiment analysis: successful approaches and future challenges / D. Tang, B. Qin, T. Liu // WIREs Data Mining and Knowledge Discovery. – 2015. – Vol. 5, no. 6. – Р. 292–303. – https://doi.org/10.1002/widm.1171.Tang D., Qin B., Liu T. Deep learning for sentiment analysis: successful approaches and future challenges. WIREs Data Mining and Knowledge Discovery, 2015, vol. 5, no. 6, pp. 292–303. https://doi.org/10.1002/widm.1171.van der Sluis, I. Affective text: Generation strategies and emotion measurement issues / I. van der Sluis, C. Mellish, G. Doherty // Proc. of the Twenty-Fourth Intern. Florida Artificial Intelligence Research Society Conf., Palm Beach, Florida, USA, 18–20 May 2011. – Palm Beach, 2011. – Р. 123–128.van der Sluis I., Mellish C., Doherty G. Affective Text: Generation Strategies and Emotion Measurement Issues. Proceedings of the Twenty-Fourth International Florida Artificial Intelligence Research Society Conference, Palm Beach, Florida, USA, 18–20 May 2011. Palm Beach, 2011, рр. 123–128.Adapting a language model for controlled affective text generation / I. Singh, A. Barkati, T. Goswamy, A. Modi // Proc. of the 28th Intern. Conf. on Computational Linguistics, Barcelona, Spain (Online), Dec. 2020. – Barcelona, 2020. – Р. 2787–2801. – https://doi.org/10.48550/arXiv.2011.04000.Singh I., Barkati A., Goswamy T., Modi A. Adapting a language model for controlled affective text generation. Proceedings of the 28th International Conference on Computational Linguistics, Barcelona, Spain (Online), December 2020. Barcelona, 2020, рр. 2787–2801. https://doi.org/10.48550/arXiv.2011.04000.Nie, G. A review of affective generation models / G. Nie, Y. Zhan. – 2022. – URL: https://arxiv.org/pdf/2202.10763 (date of access: 07.12.2025). – https://doi.org/10.48550/arXiv.2202.10763.Nie G., Zhan Y. A review of affective generation models, 2022. Available at: https://arxiv.org/pdf/2202.10763 (accessed 07.12.2025). https://doi.org/10.48550/arXiv.2202.10763.Explainable AI / V. C. Storey, R. Lukyanenko, W. Maass, J. Parsons // Communications of the ACM. – 2022. – Vol. 65, no. 4. – Р. 27–29. – https://doi.org/10.1145/3490699.Storey V. C., Lukyanenko R., Maass W., Parsons J. Explainable AI. Communications of the ACM, 2022, vol. 65, no. 4, рр. 27–29. https://doi.org/10.1145/3490699.Tanaka, Y. Cross-culture, cross-concept, and cross-subject generality of affective meaning systems / Y. Tanaka, C. E. Osgood // Journal of Personality and Social Psychology. – 1965. – Vol. 2, no. 2. – Р. 143–153. – https://doi.org/10.1037/h0022392.Tanaka Y., Osgood C. E. Cross-culture, cross-concept, and cross-subject generality of affective meaning systems. Journal of Personality and Social Psychology, 1965, vol. 2, no. 2, рр. 143–153. https://doi.org/10.1037/h0022392.Osgood, C. E. On the whys and wherefores of E, P, and A / C. E. Osgood // Journal of Personality and Social Psychology. – 1969. – Vol. 12, no. 3. – Р. 194–199. – https://doi.org/10.1037/h0027715.Osgood C. E. On the whys and wherefores of E, P, and A. Journal of Personality and Social Psychology, 1969, vol. 12, no. 3, рр. 194–199. https://doi.org/10.1037/h0027715.Суров, И. А. Открытие черного ящика: извлечение семантических факторов Осгуда из языковой модели word2vec / И. А. Суров // Информатика и автоматизация. – 2022. – Т. 21, № 5. – С. 916–936. – https://doi.org/10.15622/ia.21.5.3.Surov I. A. Opening the black box: finding Osgood's semantic factors in word2vec space. Informatika i avtomatizacija [Informatics and Automation], 2022, vol. 21, no. 5, рр. 916–936 (In Russ.). https://doi.org/10.15622/ia.21.5.3.Груздева, А. С. Машинно-семантический дифференциал: картирование эмоций посредством векторных языковых моделей / А. С. Груздева, И. А. Суров // Ученые записки Института психологии РАН. – 2025. – Т. 5, № 4. – С. 86–99. – https://doi.org/10.38098/proceedings_2025_05_04_10.Gruzdeva A. S., Surov I. A. Machine-semantic differential: emotion mapping through vector language models. Uchenye zapiski Instituta psihologii Rossijskoj akademii nauk [Proceedings of the Institute of Psychology of the Russian Academy of Sciences], 2025, vol. 5, no. 4, рр. 86–99 (In Russ.). https://doi.org/10.38098/proceedings_2025_05_04_10.Surov, I. A. Process-semantic analysis of words and texts / I. A. Surov // Artificial Intelligence in Models, Methods and Applications / ed.: O. Dolinina [et al.]. – Cham : Springer, 2023. – С. 247–260. – https://doi.org/10.1007/978-3-031-22938-1_17.Surov I. A. Process-semantic analysis of words and texts. Artificial Intelligence in Models, Methods and Applications. In O. Dolinina, I. Bessmertny, A. Brovko, V. Kreinovich, V. Pechenkin, …, V. Zhmud (eds.). Cham, Springer, 2023, рр. 247–260. https://doi.org/10.1007/978-3-031-22938-1_17.Mikolov, T. Linguistic regularities in continuous space word representations / T. Mikolov, W. Yih, G. Zweig // Proc. of the 2013 Conf. of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Atlanta, Georgia, 9–14 June 2013. – Atlanta, 2013. – Р. 746–751.Mikolov T., Yih W., Zweig G. Linguistic regularities in continuous space word representations. Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Atlanta, Georgia, 9–14 June 2013. Atlanta, 2013, рр. 746–751.Distributed representations of words and phrases and their compositionality / T. Mikolov, I. Sutskever, K. Chen [et al.] // NIPS’13: Proc. of the 26th Intern. Conf. on Neural Information Processing Systems, Lake Tahoe, Nevada, 5–10 Dec. 2013. – Lake Tahoe, 2013. – Vol. 2. – Р. 3111–3119.Mikolov T., Sutskever I., Chen K., Corrado G., Dean J. Distributed representations of words and phrases and their compositionality. NIPS’13: Proceedings of the 26th International Conference on Neural Information Processing Systems, Lake Tahoe, Nevada, 5–10 December 2013. Lake Tahoe, 2013, vol. 2, рр. 3111–3119.McLachlan, G. J. Discriminant Analysis and Statistical Pattern Recognition / G. J. McLachlan. – N. Y. : Wiley, 2004. – 526 р.McLachlan G. J. Discriminant Analysis and Statistical Pattern Recognition. New York, Wiley, 2004, 526 р.Pennington, J. Glove: Global vectors for word representation / J. Pennington, R. Socher, C. D. Manning // Proc. of the 2014 Conf. on Empirical Methods in Natural Language Processing (EMNLP), Doha, Qatar, 25–29 Oct. 2014. – Doha, 2014. – Р. 1532–1543.Pennington J., Socher R., Manning C. D. Glove: Global vectors for word representation. Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), Doha, Qatar, 25–29 October 2014. Doha, 2014, рр. 1532–1543.Kukushkin, A. Navec_hudlit_v1_12B_500K_300d_100q.tar / A. Kukushkin. – 2023. – URL: https://github.com/natasha/navec (date of access: 07.12.2025).Kukushkin A. Navec_hudlit_v1_12B_500K_300d_100q.tar, 2023. Available at: https://github.com/natasha/navec (accessed 07.12.2025).Enriching word vectors with subword information / P. Bojanowski, E. Grave, A. Joulin, T. Mikolov // Transactions of the Association for Computational Linguistics. – 2017. – Vol. 5. – Р. 135–146. – https://doi.org/10.48550/ARXIV.1607.04606.Bojanowski P., Grave E., Joulin A., Mikolov T. Enriching word vectors with subword information. Transactions of the Association for Computational Linguistics, 2017, vol. 5, рр. 135–146. https://doi.org/10.48550/ARXIV.1607.04606.Koltsova, O. Y. An opinion word lexicon and a training dataset for Russian sentiment analysis of social media / O. Y. Koltsova, S. V. Alexeeva, S. N. Kolcov // Computational Linguistics and Intellectual Technologies. – 2016. – Vol. 2016. – Р. 277–287.Koltsova O. Y., Alexeeva S. V., Kolcov S. N. An opinion word lexicon and a training dataset for Russian sentiment analysis of social media. Computational Linguistics and Intellectual Technologies, 2016, vol. 2016, рр. 277–287.From extraction to generation: Multimodal emotion-cause pair generation in conversations / H. Ma, J. Yu, F. Wang [et al.] // IEEE Transactions on Affective Computing. – 2025. – Vol. 16, no. 2. – Р. 586–597. – https://doi.org/10.1109/TAFFC.2024.3446646.Ma H., Yu J., Wang F., Cao H., Xia R. From extraction to generation: Multimodal emotion-cause pair generation in conversations. IEEE Transactions on Affective Computing, 2025, vol. 16, no. 2, рр. 586–597. https://doi.org/10.1109/TAFFC.2024.3446646.Text-based fine-grained emotion prediction / G. Singh, D. Brahma, P. Rai, A. Modi // IEEE Transactions on Affective Computing. – 2024. – Vol. 15, no. 2. – Р. 405–416. – https://doi.org/10.1109/TAFFC.2023.3298405.Singh G., Brachma D., Rai P., Modi A. Text-based fine-grained emotion prediction. IEEE Transactions on Affective Computing, 2024, vol. 15, no. 2, рр. 405–416. https://doi.org/10.1109/TAFFC.2023.3298405.Surov, I. A. Quantum core affect. Color-emotion structure of semantic atom / I. A. Surov // Frontiers in Psychology. – 2022. – Vol. 13. – https://doi.org/10.3389/fpsyg.2022.838029.Surov I. A. Quantum core affect. Color-emotion structure of semantic atom. Frontiers in Psychology, 2022, vol. 13. https://doi.org/10.3389/fpsyg.2022.838029.Widdows, D. Should semantic vector composition be explicit? Can it be linear? / D. Widdows, K. Howell, T. Cohen // Proc. of the Workshop on Semantic Spaces at the Intersection of NLP, Physics, and Cognitive Science, Groningen, The Netherlands, 14–18 June 2021. – Groningen, 2021. – Р. 76–86. – https://doi.org/10.48550/arXiv.2104.06555.Widdows D., Howell K., Cohen T. Should semantic vector composition be explicit? Can it be linear? Proceedings of the Workshop on Semantic Spaces at the Intersection of NLP, Physics, and Cognitive Science, Groningen, The Netherlands, 14–18 June 2021. Groningen, 2021, рр. 76–86. https://doi.org/10.48550/arXiv.2104.06555.Schlegel, K. A comparison of vector symbolic architectures / K. Schlegel, P. Neubert, P. Protzel // Artificial Intelligence Review. – 2022. – Vol. 55, no. 6. – Р. 4523–4555. – https://doi.org/10.1007/s10462-021-10110-3.Schlegel K., Neubert P., Protzel P. A comparison of vector symbolic architectures. Artificial Intelligence Review, 2022, vol. 55, no. 6, рр. 4523–4555. https://doi.org/10.1007/s10462-021-10110-3.Vector symbolic architectures as a computing framework for emerging hardware / D. Kleyko, M. Davies, E. P. Frady [et al.] // Proceedings of the IEEE. – 2022. – Vol. 110, no. 10. – Р. 1538–1571. – https://doi.org/10.1109/JPROC.2022.3209104.Kleyko D., Davies M., Frady E. P., Kanerva P., Kent S. J., Olshausen B. A. Vector symbolic architectures as a computing framework for emerging hardware. Proceedings of the IEEE, 2022, vol. 110, no. 10, рр. 1538–1571. https://doi.org/10.1109/JPROC.2022.3209104.Суров, И. А. Геометрическая семиотика падежей русского языка / И. А. Суров // Вестник Московского государственного университета. Гуманитарные науки. – 2026. (В печати)Surov I. A. Geometrical semiotics of Russian cases. Vestnik Moskovskogo gosudarstvennogo universiteta. Gumanitarnye nauki [Vestnik of Moscow State Linguistic University. Humanities], 2026. (In press). (In Russ.).Predicting survey responses: How and why semantics shape survey statistics on Organizational Behaviour / J. K. Arnulf, K. R. Larsen, Ø. L. Martinsen, C. H. Bong // PLoS ONE. – 2014. – Vol. 9, no. 9. – https://doi.org/10.1371/journal.pone.0106361.Arnulf J. K., Larsen K. R., Martinsen Ø. L., Bong C. H. Predicting survey responses: How and why semantics shape survey statistics on Organizational Behaviour. PLoS ONE, 2014, vol. 9, no. 9. https://doi.org/10.1371/journal.pone.0106361.Arnulf, J. K. Measuring the menu, not the food: “psychometric” data may instead measure “lingometrics” (and miss its greatest potential) / J. K. Arnulf, U. H. Olsson, K. Nimon // Frontiers in Psychology. – 2024. – Vol. 15. – Р. 1308098. – https://doi.org/10.3389/fpsyg.2024.1308098.Arnulf J. K., Olsson U. H., Nimon K. Measuring the menu, not the food: “psychometric” data may instead measure “lingometrics” (and miss its greatest potential). Frontiers in Psychology, 2024, vol. 15, р. 1308098. https://doi.org/10.3389/fpsyg.2024.1308098.Lukyanenko, R. Integrating LLMs and psychometrics: Global construct validity / R. Lukyanenko, K. R. Larsen // Forty-Fifth Intern. Conf. on Information Systems, Bangkok, Thailand, 15–18 Dec. 2024. – Bangkok, 2024. – URL: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5341306 (date of access: 07.12.2025).Lukyanenko R., Larsen K. R. Integrating LLMs and Psychometrics: Global Construct Validity. Forty-Fifth International Conference on Information Systems, Bangkok, Thailand, 15–18 December 2024. Bangkok, 2024. Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5341306 (accessed 07.12.2025).

The authors declare that there are no conflicts of interest present.