ARTIFICIAL INTELLEGENCE IN TRANSLATION: CURRENT STRUGGLES AND PROSPECTIVE SOLUTIONS

INTERNATIONAL SCIENTIFIC AND PRACTICAL CONFERENCE “INTEGRATION OF MODERN LINGUISTICS WITH SCIENCE, EDUCATION, AND PRACTICE IN THE PROCESS OF GLOBALIZATION: NEW APPROACHES, OPPORTUNITIES, AND CHALLENGES”, DECEMBER 4, 2025 938 ARTIFICIAL INTELLEGENCE IN TRANSLATION: CURRENT STRUGGLES AND PROSPECTIVE SOLUTIONS 1Urokova Dildorakhon Salimovna, 2S.S. Saydullayeva 1master’s degree student, Navoi State University, 2Scientific advisor: PhD. ORCID 0009-0008-2102-1827 https://doi.org/10.5281/zenodo.17837296 Abstract. The article identifies main challenges of AI-powered translation tools and presents prospective solutions to address them. Mistranslation of idiomatic expressions and misunderstanding contextual expressions are selected as main points of the discussion. Keywords: Artificial Intellegence (AI) translation, Neural machine translation, ChatGPT, Google translate, Microsoft translate, DeepL, contextual nuances, hybrid translation. Annotatsiya: Maqola sun’iy idrok tarjimasida jarayonidagi asosiy kamchiliklarni aniqlagan va ehtimoliy yechimlarni taklif etgan. Ideomatik birliklarni noto’g’ri tarjima qilish va kontekstni tushunishda muammolarga uchrashining muhokamasiga bag’ishlangan. Kalit so‘zlar: Sun’iy idrok (SI) tarjimasi, neyronli mashina tarjimasi, ChatGPT, Google tarjimon, Microsoft tarjimon, DeepL, kontekstdagi o’ziga xosliklar, gibrid tarjima. Аннотация: В статье выявлены основные недостатки, наблюдаемые в процессе машинного перевода, основанного на искусственном интеллекте, и предложены перспективные решения. В качестве ключевых аспектов обсуждения выделяются ошибки при переводе идиоматических выражений и неверное понимание контекстуальных значений. Ключевые слова: перевод с использованием искусственного интеллекта (ИИ), нейронный машинный перевод, ChatGPT, Google Translate, Microsoft Translator, DeepL, контекстуальные нюансы, гибридный перевод. Introduction. Translation has long been a crucial tool which builds mutual understanding between cultures and nationalities, and a great assistant in knowledge and information exchanging. From the beginning of the translation history the methods, tools and attitudes used in this process developed gradually. In contemporary years, after the introduction of artificial intelligence (AI) the practise of translation has met notable enhancements in speed and accessibility features. AI-powered tools have democratized translation services, making them widely available to individuals and organizations across the globe 1 . However, even some AI-powered tools like DeepL, Google translate and ChatGPT presents more accurate and efficient translations, the challenges in the interpretation of idiomatic expressions, metaphors and context-depended phrases are still remaining. By implementing hybrid translation models and expanding and diversifying the training data such gaps in AI translation process can be bridged. This article is devoted to study particular challenges which AI struggles with in the process of translation and proposing prospective solutions to them. 1 Freitag, Markus, and Yaser Al-Onaizan. “Beam Search Strategies for Neural Machine Translation.” Proceedings of the Second Conference on Machine Translation (WMT 2017), Association for Computational Linguistics, 2017, pp. 56–60. INTERNATIONAL SCIENTIFIC AND PRACTICAL CONFERENCE “INTEGRATION OF MODERN LINGUISTICS WITH SCIENCE, EDUCATION, AND PRACTICE IN THE PROCESS OF GLOBALIZATION: NEW APPROACHES, OPPORTUNITIES, AND CHALLENGES”, DECEMBER 4, 2025 939 Methodology. The survey method was used to collect data about the importance of artificial intelligence (AI) in translation, main problems in the practice of AI translation and to present potential approaches to address these issues. At first stage of survey the review of literature was conducted, some of the data gathered were filtered and then main ideas were selected to clarify throughout this work. Analysis. AI-powered translation tools like Google translate, Microsoft translator and DeepL are the most well-known neural machine translators of our time. Neural machine translators (NMT) use artificial neural networks which produce translation by analyzing massive amount of data gathered in a system from the previously queried requests. In this process NMT implements beam search method. The current beam search strategy generates the target sentence word by word from left-to-right while keeping a fixed number of active candidates at each time step 2 . However, despite its effectiveness, this method still gives rise to certain challenges and shortcomings during the translation process. Challenges of AI translation 1. Cultural and contextual nuances However, despite their remarkable technical progress, AI translation systems are not immune to limitations—especially when it comes to cultural nuance. The role of context cannot be overstated when discussing cultural nuance. Words and phrases acquire meaning based on the situation in which they are used. A single sentence can carry multiple interpretations depending on tone, setting, relationship between speakers, and cultural expectations. AI systems, which rely on large datasets and pattern recognition, often struggle to discern these subtle contextual cues. Without real-world experience or emotional intelligence, machines can misinterpret or flatten messages that require nuanced understanding. 3 2. Idiomatic expression AI translation tools often struggle to interpret idiomatic expressions, metaphors, and context-dependent phrases. These linguistic features, deeply embedded in cultural and social contexts, frequently extend beyond their literal meanings, posing difficulties for AI systems reliant on statistical patterns and training data. 4 NMT often translates idioms word by word, which leads to loss of meaning and awkward translation. For instance, idiomatic expression like “It is raining cats and dogs” can be translated by AI verbatim and expresses the meaning of “It is raining animals” which is actually not about animals but intensity of rain. The reason can be explained as cultural unawareness of NMT. Because idioms cannot be translated precisely but the meaning depends on the culture and its mentality. Discussion. The challenges like cultural and contextual misinterpretation and misunderstanding of idiomatic expressions in AI-based translation requires complex approaches. Below some potential solutions will be cited which is aim to asses to those difficulties in AIpowered translation process. Proposed solutions. 1. Hybrid translation models 2 Freitag, Markus, and Yaser Al-Onaizan. “Beam Search Strategies for Neural Machine Translation.” Proceedings of the Second Conference on Machine Translation (WMT 2017), Association for Computational Linguistics, 2017, pp. 56–60. 3 Boluwatife, Oni Samuel. “Cultural Nuances in Translation: AI vs Human Translators.” ResearchGate, Apr. 2025. 4 Vaswani, Ashish, et al. “Attention Is All You Need.” Advances in Neural Information Processing Systems, vol. 30, 2017, pp. 5998–6008. INTERNATIONAL SCIENTIFIC AND PRACTICAL CONFERENCE “INTEGRATION OF MODERN LINGUISTICS WITH SCIENCE, EDUCATION, AND PRACTICE IN THE PROCESS OF GLOBALIZATION: NEW APPROACHES, OPPORTUNITIES, AND CHALLENGES”, DECEMBER 4, 2025 940 Hybrid translation method (HMT) is a combination of rule-based (RBMT), statistical (SMT), neural machine translation (NMT) and sometimes human post-editing approaches which aims to produce more accurate translations and overcome limitations of each system. For instance, while RBMT has the limitation on understanding idioms, SMT struggles with grammatical problems and NMT is weak in interpretation of context. In order to overcome the translation problems in different Machine Translation (MT) systems and to receive the accurate and better results of the translation process, Hybrid Machine Translation (HMT) is used. HMT incorporates the full advantages of the MT systems that are used in the creation of HMT systems. 5 2. Expanding and diversifying training data The quality of AI translation tools is inherently tied to the diversity and accuracy of the data on which they are trained. To reduce biases and improve cultural sensitivity, AI systems must be trained on datasets that are representative of a wide range of languages, dialects, and cultural contexts. Including minority languages and culturally specific expressions in training datasets ensures more accurate and inclusive translations. 6 Diversifying data is about to create wider range of information from different registers, multiple styles in order to make AI aware of different contexts. Conclusion. The process of translation met notable enhancements after the integration of AI to the field. Machines offer speed, better accessibility chances and cost efficiency which can be stated as revolutionary effects of AI in translation. However, there are a number of challenges which AI deals with in translating idioms, contextual and cultural nuances. Such gaps lead to misinterpretation, inaccuracy and loss of meaning in texts. In order to address problems above, hybrid models of translation should be implemented to the process. Consequently, each individual approach (HBT, RBMT, NMT, SMT) fulfill the gap of the other system. Furthermore, training data of the system also should be extended and enriched. By bridging such gaps in AI-powered translation, opportunities also will be broadened and AI can become more powerful tool which serves humanity to break linguistic barriers. References 1. Anugu, Anusha, and Gajula Ramesh. “A Survey on Hybrid Machine Translation.” ResearchGate, 2020. 2. 2.Bahri, M. “Machine Translation: Challenges and Opportunities in the Age of Artificial Intelligence.” Journal of Linguistic Studies, vol. 45, no. 2, 2020, pp. 123–137. 3. 3.Boluwatife, Oni Samuel. “Cultural Nuances in Translation: AI vs Human Translators.” ResearchGate, Apr. 2025. 4. 4.Freitag, Markus, and Yaser Al-Onaizan. “Beam Search Strategies for Neural Machine Translation.” Proceedings of the Second Conference on Machine Translation (WMT 2017), Association for Computational Linguistics, 2017, pp. 56–60. 5. Shahmerdanova, Roya. “Artificial Intelligence in Translation: Challenges and Opportunities.” Acta Globalis Humanitatis et Linguarum, vol. 2, no. 1, 2025. 5. Vaswani, Ashish, “Attention Is All You Need.” Advances in Neural Information Processing Systems, vol. 30, 2017, pp. 5998–6008. 5 Anugu, Anusha, and Gajula Ramesh. “A Survey on Hybrid Machine Translation.” ResearchGate, 2020 6 Bahri, M. “Machine Translation: Challenges and Opportunities in the Age of Artificial Intelligence.” Journal of Linguistic Studies, vol. 45, no. 2, 2020, pp. 123–137. INTERNATIONAL SCIENTIFIC AND PRACTICAL CONFERENCE “INTEGRATION OF MODERN LINGUISTICS WITH SCIENCE, EDUCATION, AND PRACTICE IN THE PROCESS OF GLOBALIZATION: NEW APPROACHES, OPPORTUNITIES, AND CHALLENGES”, DECEMBER 4, 2025 941 6. Raximov, Ulugbek Amirbekovich. “A Contrastive Study of Artificial Intelligence and Human Translation Based on Legal Texts.” International Journal of Scientific Researchers, vol. 14, no. 2, 2025. 7. https://www.researchgate.net 8. https://www.doi.org 9. https://aclanthology.org