Large language models as a tool for translator's practical activities: a review of foreign research projects

Immediately after its market launch, ChatGPT rapidly captured the attention of users worldwide and became a sought-after tool among professionals engaged in text-related activities. Despite competition from numerous specialized online services, ChatGPT is gradually expanding its presence in the field of machine translation. The capabilities and limitations of this program have triggered discussions in various professional communities, including the academic circles. Research concerning ChatGPT and other LLM in terms of translation performance is currently limited and predominantly published abroad. However, the rapid advancement and dissemination of this technology necessitate the synthesis of such works, which will help crystallize the major trends in the development of the translation industry in the near future. The aim of the review is to show the results of research projects conducted by foreign scholars regarding the application of LLM in translation practices. The novelty of the study lies in identifying a spectrum of directions for the integration of artificial intelligence technologies into the translation process and in synthesizing the applied potential of neural network technologies, exemplified by the use of ChatGPT. The analysis is based on 12 printed and electronic sources in English and German, primarily dated 2023 and 2024, which focus on testing the aforementioned chatbot in translating texts of various genres into languages with both low and high resource bases, as well as utilizing different prompt designs. The synthesis of this material reveals universal trends and characteristics in the application of LLM as innovative, multifunctional, and continuously evolving tools for practicing translators. This paper reviews the approaches of specialists from China, Iran, the United Kingdom, the United States, and several European countries regarding the application of ChatGPT and other LLM in translation activities. These works include practical studies, methodological recommendations, and expert evaluation that reveal trends in the development of LLM. The coverage of a wide range of countries and genres of the reviewed scientific texts allows for the formation of a comprehensive and multifaceted understanding of the role of LLM in the market for digital products aimed at professional translation activities. The study employs conceptual-comparative and evaluative-critical types of analysis. Due to them the information particularly relevant to Russian digital translation studies is highlighted and synthesized. The review indicates that specialists confirm the efficiency of such LLM as ChatGPT in translation practice. They are capable of automating tasks related to terminology extraction from texts and performing high-quality machine translation, competing with well-known services of Google, DeepL and other large companies. It is noted that ChatGPT successfully copes with inter-stylistic translation. According to experts, the ability of LLM to transform the style and genre of the original text while preserving the essential key semantics in the secondary text will find wide application across various sectors of the economy. Furthermore, ChatGPT is applicable not only for addressing immediate translation tasks but also for automating pre-translation text analysis, including the extraction of terminology from the text, providing this lexicon with necessary linguistic information, and systematization. However, analysts stress the tendency of neural network systems to generate incorrect information, as well as the dependence of translation quality on the genealogical relationship and resource capacity of the national languages processed by these systems. To overcome these limitations, specialists are implementing projects to optimize the architectures of LLM and are developing prompt engineering within the translation industry. The enhancement of prompt bases should, as experts claim, follow the track of developing comprehensive prompts that contain exhaustive contextual information, including requirements regarding the genre-stylistic affiliation of the secondary text, its structural-organizational properties. The modifiers and add-ons of prompts should include reference translation patterns and requirements for machine post-editing of the obtained primary results. Thus, ChatGPT and other LLM pose competition to professional translators but cannot yet replace them. Nevertheless, specialized commercial and scientific-educational organizations are actively working to improve these systems. The rapid inhancement in quality within the field of natural language processing, along with the multifunctionality and adaptability of digital products, already requires translators to obtain new skills and demands that researchers identify promising study domains focused on understanding various aspects of artificial intelligence from a linguistic perspective. The authors declare no conflicts of interests.

Keywords

Authors

References