D.A. Bakhmatov1, V.V. Vakulenko2, I.M. Zatsman3

1 Lomonosov Moscow State University (Moscow, Russia)
2, 3 FRC «Computer Science and Control» of the Russian Academy of Sciences (Moscow, Russia)
1 bahdan2013@yandex.ru, 2 v@vvak.ru, 3 izatsman@yandex.ru

We propose an approach to solving word sense disambiguation problems using large language models. These problems fall under the category of semantic complexity problems. It is difficult to define such problems formally (e.g., creating a system of rules for semantic interpretation of a text). The proposed approach is described using the example of disambiguating polysemantic modal verbs in German. Disambiguation involves choosing one of several meanings of a polysemantic word that corresponds to the semantic content of its use in the analysed context. For this task, it is difficult to formally define the concept of «meaning» or the boundaries between different meanings of a polysemantic word, that explains its classification as a problem of semantic complexity. Traditionally, such problems have been solved by experts during the creation of dictionaries, thesauri, and other knowledge organization systems. It was associated with significant time and labour costs. With the advent of artificial intelligence systems (AI systems) based on large language models, some expert work can be transferred to them.

The purpose of this article is to describe an approach to solving word sense disambiguation problems for polysemantic words based on hybrid intelligence, combining the operation of an AI system with subsequent expert analysis of its results. Practical relevance is that significant reduction time of word sense disambiguation.

Bakhmatov D.A., Vakulenko V.V., Zatsman I.M. Word sense disambiguation by large language models // Highly Available Systems. 2026. V. 22. № 2. P. 33−39. DOI: https://doi.org/10.18127/j20729472-202602-03

1. Akata Z., Balliet D., De Rijke M., Dignum F., Dignum V. et al. A research agenda for hybrid intelligence: Augmenting human intellect with collaborative, adaptive, responsible, and explainable artificial intelligence. Computer. 2020. V. 53. № 8. P. 18–28.
2. Bredeweg B., and Kragten M. Requirements and challenges for hybrid intelligence: A case-study in education. Frontiers in Artificial Intelligence. 2022. V. 5. № 8. P. 1–11.
3. Weber F. Wambsganß T., and Soellner M. Enhancing legal writing skills: The impact of formative feedback in a hybrid intelligence learning environment. British Journal of Educational Technology. 2025. V. 56. № 2. P. 650–677.
4. Dobrovol'skiy D.O. Nemetskiye modal’nyye glagoly v parallel’nom korpuse i zadachi dvuyazychnoy leksikografii. Germanskiye yazyki: tekst, korpus, perevod. Moscow: Vinogradov Russian Language Institute. 2020. P. 103–116.
5. Dobrovol'skiy D., and Zalizniak Anna A. Ob osobom tipe modal’nosti neobkhodimosti: semantika nemetskogo glagola sollen po dannym parallel’nykh korpusov. Voprosy yazykoznaniya. 2021. № 6. P. 51–68.
6. Dobrovol’skiy D., and Zatsman I. Model’ izvlecheniya znaniya iz parallel’nykh tekstov leksikograficheskoy informatsionnoy sistemy. Informatika i eyë primeneniya. 2024. V. 18. № 3. P. 97–105.
7. Postanovleniye Pravitel'stva RF ot 25 dekabrya 2024 g. № 1892 «O Natsional'nom slovarnom fonde». http://government.ru/ docs/53894/
8. DWDS (Digitales Wörterbuch der deutschen Sprache). https://www.dwds.de
9. Klein W., and Geyken A. Das Digitale Wörterbuch der Deutschen Sprache (DWDS). Lexicographica. 2010. V. 26. № 2010. P. 79–96.
10. Geyken A., Wiegand F., and Würzner K.-M. On-the-fly Generation of Dictionary Articles for the DWDS Website. Proceedings of eLex 2017 conference. Brno: Lexical Computing CZ s. r. o. 2017. P. 560–570.
11. What is AI inference? https://www.ibm.com/think/topics/ai-inference
12. LLM Inference. https://www.deepchecks.com/glossary/llm-inference/
13. Wörterbücher im DWDS:sollen. https://www.dwds.de/wb/sollen
14. Duden Online:sollen. https://www.duden.de/rechtschreibung/sollen
15. Parallel'nyj nemeckij korpus v sostave Nacional'nogo korpusa russkogo jazyka. https://ruscorpora.ru/search?search=CiUqFwoICAAQ ChgyIAogADIFZ3JzdGRACngBMgcIBRIDZGV1OgEB