Contextual Code Switching for Machine Translation using Language Models (In the Example of Uzbek- English Languages)

• Xonqulova Nilufar Ravshanovna

  English applied translation department, translation faculty Uzbekistan state university of world languages

This study investigates the challenges and prospects of handling contextual code switching (CCS) in machine translation (MT) between Uzbek and English using large language models (LLMs). The objective is to explore how CCS affects translation quality and to evaluate the effectiveness of transformer-based models in adapting to bilingual utterances. Using a manually curated corpus of code-switched Uzbek-English texts, we test the translation capabilities of multilingual models like MBART and NLLB. Our findings reveal that while LLMs show promise in semantic retention and contextual awareness, they often struggle with preserving syntactic fidelity across mixed-language input. Recommendations for future research include the development of CCS-sensitive pretraining and fine-tuning strategies.

Poplack, S. (1980). "Sometimes I'll start a sentence in Spanish y termino en español": Toward a typology of code switching. Linguistics, 18(7–8), 581–618.

2. Conneau, A., et al. (2020). Unsupervised Cross-lingual Representation Learning. ACL.

3. Fan, A., et al. (2022). Building Machine Translation Systems for the Next Thousand Languages. ArXiv.

4. Bojar, O., et al. (2018). Findings of the WMT18 Shared Task on Machine Translation.

5. Tiedemann, J., & Thottingal, S. (2020). OPUS-MT – Building Open Translation Services for the World. EACL.

6. Bakirova H.B. (2022). Terminology is a Tool in Development of Lexical Competence. Eurasian Scientific Herald, 4, 69–73.

7. Bakirova Kh.B. (2021). Selection of lexic material for terminological dictionary minimum of energy specialty. "Online – conferences & Quot; platform, 108–109.

8. Bakirova H. (2021). Typology of methodological and linguistic difficulties in the formation of lexical competence. ACTA NUUz. 1/5/1. 44p.