Pragmatic Adaptation of Promotional Texts in Artificial Intelligence Chatbots: An Iterative Approach

DOI:

Alongside neural machine translation technologies, Artificial Intelligence-powered chatbots capable of handling a wide range of text processing and information retrieval tasks have become widespread today. Over the past two years, tools such as DeepSeek, ChatGPT, Perplexity, YandexGPT, etc. have assumed a significant role in the workflow of professional translators. According to scientific studies and feedback from language service specialists, these chatbots deliver commendable results, enabling optimisation of the translation process – particularly for large data volumes and tight project deadlines. The pace at which new AI technologies and systems applicable to translation practice emerge far outstrips the rate at which they are comprehended and mastered, fuelling heightened interest from the scholarly community. The translation services sector has entered a new AI-driven reality, where harnessing these capabilities represents an essential phase in advancing translation theory, practice, and pedagogy.

This paper examines the potential of AI chatbots for addressing the pressing challenge of developing multilingual online resources, which entails pragmatic adaptation of texts for foreign target audiences. The study objective is to devise an effective prompting algorithm within an iterative framework for translating and pragmatically adapting promotional texts in chatbots, thereby automating machine translation post-editing. Achieving this goal involves tackling several tasks: conceptualising the concept of iteration in the context of text processing via machine translation programmes; exploring diverse prompting algorithms for chatbot use; conducting a translation analysis of source and target texts; and evaluating the pragmatic potential of chatbot-generated texts.

Methodology and materials

The research methodology is shaped, on the one hand, by its focus on translation techniques and the need for textual analysis to meet the research objectives, and on the other, by the use of information technologies to address linguistic challenges.

A pivotal element of this methodology is the iterative approach, so let us substantiate its role in addressing the research tasks. Since AI training occurs solely through human interaction, each application of AI-based programs necessitates exploring novel research models and tools for managing these technologies. Given the specifics of cutting-edge AI systems (software development and customization, system-user communication), it is essential to grasp their operational logic and adopt approaches akin to those used in their design. The chatbot format inherently involves user communication, rendering feasible and expedient the construction of task-solving algorithms – including translation ones – via an iterative approach, which, according to N.V. Kirillina, entails adjusting user interaction settings based on interim results of such interaction [Kirillina, 2020, p. 176].

From a linguistic perspective, the value of this approach lies in enabling control over the written translation process in chatbots through diversification of prompt formulations aligned with interim translation outputs. Traditional neural machine translation programmes such as Yandex Translate, Google Translate, DeepL, and Prompt do not support user interactions and cannot respond to or accommodate specific requests. This distinction underscores the relevance and novelty of investigating the efficacy of state-of-the-art machine translation systems within the proposed iterative framework.

The research methodology also includes such methods as sampling and content analysis to compile the empirical corpus; pre-translation analysis to devise strategies for translation and pragmatic adaptation; comparative and translation analysis to evaluate the equivalence and adequacy of bot-generated target texts; definition analysis for refining specialised terminology; and discourse analysis to take into account the array of extralinguistic text parameters critical to pragmatic adaptation during translation.

The research material comprises promotional texts in Russian and English from museum websites. The translation of such texts has gained particular relevance today amid the flourishing multilingualism of Internet communication and the companies’ strive for international presence. The chatbot employed for the purpose of this study is represented by Perplexity – an AI-powered search engine designed to provide well-sourced answers in response to user queries .

The interaction between translators – as experts in intercultural communication and text generation – and AI-based translation systems ranks among the most pressing topics in modern translation studies, covered in numerous investigations. Alongside the issues of post-editing, pre-editing, legal regulation of AI use, and evaluation of machine translation systems, researchers increasingly address the issues of training these systems, which is becoming more transparent and controllable. German researcher O. Czulo discusses the operational principles of large language models (LLMs) underpinning contemporary neural translation technologies and emphasises that high-quality text generation is feasible only with meticulously hand-curated training corpora: “If humans are excluded from machine learning processes, allowing language models to train on self-generated texts, these systems will embark on a path of degradation, with information stripped of the intra- and extralinguistic nuances of human communication becoming amplified within the model” (cited in: [Aleksandrov, Chistova, 2025, p. 15]). The efficacy of machine translation systems, particularly their capacity to process entire texts rather than isolated utterances, is partly determined by the quality of their training data, including structural and stylistic text features, speech genres, and so forth [Klyachin, Khizhnyakova, 2024; Salimovsky et al., 2021].

The controllability of neural network training – including achieving specific short-term practical outcomes such as text translation – becomes feasible through the chatbot format, which is inherently designed for user communication. Chatbots’ responsiveness to linguistic prompts has rendered them indispensable across domains, including translation. However, their effectiveness hinges substantially on iterative approach algorithms and prompt formulation. The concept of “iteration” (from Latin iteratio , “repetition”) entered information technology and communication-related fields from mathematics, denoting the repetition of a typical action, step, or sequence of operations within an overarching process, wherein outputs from prior steps are used in subsequent ones [Iteratsiya…, 2026]. As N.V. Kirillina observes, each subsequent stage involves analysing interim results with subsequent refinements to the system’s core parameters and user interaction requirements based on that analysis [Kirillina, 2020, p. 176]. According to this approach, each successive iteration enhances the system’s functionality through user feedback.

The tool for implementing the iterative approach is represented with prompting, encompassing user queries (prompts). A prompt is considered to be a query, command, or set of instructions that a user conveys to an AI programme to accomplish a specific task efficiently [Bashkirov, 2025]. The literal meaning of this word is hint, command, or reminder. Among extant prompt types (textual, visual, programmatic, mathematical, interactive, etc.), textual prompts hold particular relevance in the present study, as they are geared towards text creation, such as requests for article writing, text translation, comment generation, etc. [Bashkirov, 2025].

A number of studies are devoted to prompt engineering, wherein authors evaluate the efficacy of diverse prompt formats across a broad spectrum of tasks, including translation [Jiao et al., 2024; Karachun, Smakotina, Astakhova, 2024; Reynolds, McDonell, 2021; Lyu et al., 2023]. A recent study by Chinese researchers substantiates the construction of multi-level prompts to enhance translation quality in ChatGPT for the English-Chinese language pair [Jiao et al., 2024]. The authors demonstrate that chain-of-thought prompting improves the quality of ChatGPT-generated translations. American researchers Reynolds and McDonell conduct a comparative evaluation of zero-shot and few-shot prompts, illustrating – via metalinguistic tasks – how prompt variations can yield substantial shifts in system outputs [Reynolds, McDonell, 2021]. The authors further draw an analogy between prompt engineering and natural language programming, which is undoubtedly more complex and culturally determined than formal programming languages [Reynolds, McDonell, 2021, p. 4].

The effectiveness of chatbots in performing translation tasks largely hinges on a creative approach to prompt design, which steers AI’s generative activity and mitigates the issue of divergent outputs from identical inputs. Beyond this variability tendency, researchers from Wake Forest University have identified ChatGPT’s propensity to simplify texts across iterations, resulting in the loss of critical information [Lyu et al., 2023]. In the scholars’ view, a solution here also lies in detailing the task with respect to the target text structure, semantic content, and even the number of paragraphs.

Important aspects of prompt engineering, including various techniques and strategies for optimizing prompts and increasing their effectiveness, are highlighted in the study by Russian researchers titled “Optimizing Prompts for Effective Communication with the ChatGPT” [Karachun, Smakotina, Astakhova, 2024]. The authors conclude that a prompt, or a sequence of prompts, should include such mandatory characteristics as a description of the goal and result (searching for information, translating a text, etc.), assigning a specific role to the bot (linguist, programmer, professional translator), and language design in the instructional genre (use of imperative constructions and formal style) [Karachun, Smakotina, Astakhova, 2024, p. 537]. In addition to the above characteristics, a pragmatic function plays a significant role in generating bot queries. This involves identifying the target audience for which specific content is created. It seems advisable to define potential recipients by age, gender, profession, and cultural background, since the translation result depends on these characteristics.

Regarding the implementation of the pragmatic function in solving translation tasks while using chatbots, it should be noted that issues of pragmatic text adaptation, which first attracted the attention of Russian and foreign translation scholars in the 1970s and 1980s, are particularly important in the context of modern institutional Internet communication, particularly in the sphere of public and commercial services. The multilingual Internet space has become a phenomenon of deglobalization, which has manifested itself as a trend of the latest decade, and translation practitioners and theorists were the first to respond to the need of website creators to create multilingual web pages [Mityagina, Novikova, 2025, p. 139]. As translation scholar P. Sandrini rightly notes, creating multilingual content on the Internet using machine translation alone is ineffective because machine translation systems operate only on data from performed translations and cannot take into account the purpose of the translation [Sandrini, 2021, S. 41].

Informing a foreign target audience is accomplished through the creation of understandable, accessible texts that take into account the recipient’s pragmatics and culture, as well as the sender’s pragmatics, while pursuing specific goals. According to A.V. Achkasov, pragmatic adaptation is the transformation of a text during translation, when it is accompanied by taking into account the cultural, social, marketing, and economic parameters of the new locale [Achkasov, 2016, p. 12]. From this perspective, written translation of texts on various topics, including the content of websites of various organizations, as an interlingual and intercultural transfer of information that requires consideration of the communicative-and-cultural code of the text and the author’s intentions, remains inaccessible to large language models, including chatbots, without post-editing or the use of a prompting algorithm relevant to each specific case.

Results and discussion

National and cultural vocabulary used in promotional texts is always hard for comprehension by foreign readers, so pragmatic adaptation of the texts containing various types of exotic words is an important step in the translation process. A large number of culturally marked lexical units are often used in the content of museum websites whose collections reflect the cultural heritage of a particular people.

To verify the statements above we offer to examine the capabilities of the Perplexity neural network when translating a promotional text of the Museum of Russian Fairytales (Volgograd Region), which contains numerous fairy tale references, including the names of fairy tale characters, place names, and allusions to various elements of Russian folklore. The objective of this stage of the study is to compare the effectiveness of two prompts – a simple one and a more complex one – while refining additional translation parameters based on the results of the first iteration. For the first iteration, we’ll take a prompt specifying the bot role and the translation task, and perform a translatological analysis of the source and target texts (see Table 1). The prompt is worded as follows: “You are a professional translator. Translate this text from Russian to English.”

A key distinction between chatbots and traditional machine translation systems lies in the trend of the former to provide comments to translations performed. In the given example, the bot’s comments are as follows:

• (1)    “As a professional translator, I’ve provided a precise, natural English rendition of the text. It preserves the original’s whimsical, inviting tone while ensuring fluency for native speakers. Fairy-tale elements use standard English adaptations for accessibility”;

• (2)    “Would you prefer a more literal translation, additional cultural notes, or adjustments for a specific audience (e.g., children)?”

Nesterova and Soboleva emphasize that transfer of cultural units into another cultural environment demands, on the one hand, profound knowledge of the original linguoculture and, on the other, professional mastery in selecting expressive means in the target language [Nesterova, Soboleva, 2012, p. 76]. To evaluate the adequacy of the translation from a pragmatic viewpoint it is necessary to find equivalents for the clearest English translations of the Russian fairytales, i.e. to find out what names might be appropriate in catching cultural background of Russian fairy tale characters by native English speakers. However,

Table 1. The source text and the neural translation of the Museum of Russian Fairytales promotional text (Iteration 1)

Source (The Museum of Russian Fairytales) Одним из таких волшебных мест является «Музей русской сказки» в посёлке Кировец Среднеахтубинского района Волгоградской области. Необычный музей, а скорее парк развлечений, начал принимать гостей в 2008 году. Все экспозиции выполнены в натуральную величину и стилизованы под любимые русские сказки , повсюду ощущается атмосфера волшебства. Зоны оформлены тематически: Полянка, Царские палаты, Тридевятое царство, пещера Велеса, Глухомань, Лукоморье и Емелина трапезная . Гостей встречают и сопровождают сказочные персонажи: Дядюшка Чур, Баба-Яга, Лель, Балда, Иванушка, Емеля, Несмеяна и другие герои русских сказок. В музее повсюду стоят артефакты знакомые с пеленок : сапоги-скороходы, ковер-самолет, меч-кладенец, скатерть-самобранка .

Translation by Perplexity

One such enchanting spot is the “Museum of Russian Fairytales” in the village of Kirovets, Sredneakhtubinsky District, Volgograd Oblast. This unconventional museum – more like an amusement park – opened to visitors in 2008. All exhibits are built to life size and styled after beloved Russian fairy tales , with a pervasive sense of magic in the air. The zones are themed: the Fairy Glade, the Tsar’s Chambers, the Thrice-Ninth Kingdom, Veles’s Cave, the Backwoods, Lukomorye, and Emelya’s Banquet Hall . Fairytale characters greet and guide visitors: Uncle Chur, Baba Yaga, Lel, Balda, Ivanushka, Emelya, Princess Nesmeyana , and other icons of Russian folklore. The grounds brim with childhood-familiar artifacts: seven-league boots, flying carpet, treasure sword, and self-spreading tablecloth .

even at the preliminary stage of evaluating the AI-generated translation, it is obvious that many proper names require adaptation, as the transliterated form suggested by the bot is insufficient in fulfilling the informative function and requires clarification of the semantic content (e.g. Emelya , Lel , Balda , Lukomorye , etc.). The study by [Barinova, Nesterova, Sergutina, 2016] provides a detailed translation analysis of the most common names from Russian folk tales being rendered into English, and the authors conclude that English usage of these names is characterized by variability [Barinova, Nesterova, Sergutina, 2016, p. 51]. For example, the name Баба-Яга can be translated as Baba Yaga , a witch Baba Yaga , the old Baba Yaga ; За тридевять земель – beyond the Thrice-Nine Land ; in the thirtieth kingdom ; in the Thrice-Ten Kingdom ; beyond twenty-seven lands , etc. [Barinova, Nesterova, Sergutina, 2016, pp. 51-52]. Online collections of Russian fairytales translated into English by foreign authors likewise offer numerous examples of rendering Russian fairy-tale names: Иванушка – Ivanoushka the Simpleton , Ivan the Fool ; Емеля – Emelyan the Fool (Russian Fairytales). In the absence of an “ideal recipe” for translating culture-specific units, one should focus on the combined use of literal and semantic translation methods, which enables, on the one hand, preserving the national flavor in the context of promotional discourse and, on the other hand, explaining the main meaning of words through a brief description (e.g. Лукоморье – Lukomorye Land ; Баба-Яга – The Witch Baba Yaga ; Лель – Lel, the Spirit of Spring , etc.).

It is worth noting that the example under study contains certain expressions that appeal to representatives of the Russian linguistic culture, being brought up with the Russian fairytales, specifically, from the cradle :

• (3)    Все экспозиции выполнены в натуральную величину и стилизованы под любимые русские сказки ;

• (4)    В музее повсюду стоят артефакты знакомые с пеленок.

When the focus shifts from a Russianspeaking addressee to a foreign one, these expressions lose their significance, as representatives of other linguistic cultures may have only a superficial knowledge of the Russian folklore or be unfamiliar with it altogether. Therefore, a pragmatic adaptation is required to offer an outside perspective and neutralize the linguistic and cultural specificity that is inappropriate in this context.

Based on the translatological analysis of the presented neural network translation, we formulate a second prompt aimed at adapting the text: “Perform a pragmatic adaptation of the translation, taking into account that foreign readers are unfamiliar with Russian fairytales.” In our opinion, in this case, specifying additional addressee characteristics is inappropriate, as the museum promotional text is intended for a wide range of readers of various ages. Table 2 presents fragments of the source text and the target text generated by Perplexity for the specified query.

To illustrate the changes the bot made to the text after the second prompt, Table 2 presents the results of two iterations, with the text broken down into individual sentences. As the comparative analysis shows, the adapted text is semantically deepened due to the inclusion of parenthetical explanations for almost every cultural item. These explanations provide a brief overview of the fairytale location or character and add clarity to the translation: Лукоморье → Lukomorye → Lukomorye (a mythical oak grove from folklore) ; Емеля → Emelya → Emelya (the lucky lazybones who catches a magic fish) ; Дядюшка Чур → Uncle Chur → Uncle Chur (a quirky guardian spirit) , etc.

The bot also clarifies the concept of sevenleague boots , which is an element of the European folklore and serves as a functional analogue of the Russian sapogi-skorokhody (“boots for super-speedy travel”). In our opinion, a descriptive translation is also justified in this case, given the continuing role of English as a lingua franca: in addition to being the national language in many countries, it also serves as a lingua franca for residents of many other countries.

Providing the target text with a large number of descriptions raises the issue of excessive explication. However, this hurdle can be solved by restructuring and creolizing the text: presenting information concisely; breaking it down into short points and subpoints, which are considered the most readable form of web content; and using images with brief descriptions.

As for the appeal to representatives of Russian linguoculture, as a result of the second iteration, the expression artifacts familiar from the cradle was eliminated in the text: артефакты, знакомые с пеленок → childhoodfamiliar artifacts → magical artifacts from the tales.

The second prompt proved effective, and the chatbot identified this expression as inappropriate for foreign recipients. However, the expression beloved Russian fairytales , which also marks linguocultural identity to a certain extent, remained unchanged in the translation.

Now we will test the effectiveness of the prompts used on the example of AI translation of an English-language promotional text of The Story Museum (Oxford) into Russian. Taking into account the change in translation direction, the prompt sequence looks like this:

Iteration 1 : “You are a professional translator. Translate the text from English to Russian.”

Iteration 2: “Perform a pragmatic adaptation of the translation, taking into account that Russian readers are not familiar with English culture.”

Table 3 presents the source text and the target text generated by Perplexity for the above specified query:

The source text contains an allusion to a British children’s book We’re Going on a Bear Hunt – a piece of classical children’s literature written in 1989: You can go on a Bear Hunt. The translation of this expression, based on the first prompt, is the following phrase: отправьтесь на охоту за медведем. A further prompt for pragmatic adaptation leads to a clarification of the plot: отправьтесь на поиски медведя (по книге «Охота за медведем» Майкла Роуза) . Other characters from the English children’s literature have been similarly described:

Table 2. Fragments of the source and the neural translation of the Museum of Russian Fairytales promotional text (Iterations 1 and 2)

Source (The Museum of Russian Fairytales)	Translation by Perplexity (iteration 1)	Translation + pragmatic adaptation by Perplexity (iteration 2)
Все экспозиции выполнены в натуральную величину и стилизованы под любимые русские сказки .	All exhibits are built to life size and styled after beloved Russian fairytales , with a pervasive sense of magic in the air.	All exhibits are built to life size and styled after beloved Russian fairytales , with a pervasive sense of magic in the air.
Зоны оформлены тематически: Полянка, Царские палаты, Тридевятое царство, пещера Велеса, Глухомань, Лукоморье и Емелина трапезная .	The zones are themed: the Fairy Glade, the Tsar’s Chambers, the Thrice-Ninth Kingdom, Veles's Cave, the Backwoods, Lukomorye, and Emelya’s Banquet Hall .	The zones are themed: the Fairy Glade, the Tsar’s Chambers, the Thrice-Ninth Kingdom (a faraway magical realm), Veles’s Cave (named for the Slavic god of the underworld), the Backwoods, Lukomorye (a mythical oak grove from folklore), and Emelya’s Banquet Hall.
Гостей встречают и сопровождают сказочные персонажи: Дядюшка Чур, Баба-Яга, Лель, Балда, Иванушка, Емеля, Несмеяна и другие герои русских сказок.	Fairy-tale characters greet and guide visitors: Uncle Chur, Baba Yaga, Lel, Balda, Ivanushka, Emelya, Princess Nesmeyana , and other icons of Russian folklore.	Fairytale characters greet and guide visitors: Uncle Chur (a quirky guardian spirit), Baba Yaga (the famous witch in her hut on chicken legs), Lel (a playful spring spirit), Balda (the clever farmhand), Ivanushka (the mischievous little brother), Emelya (the lucky lazybones who catches a magic fish), Princess Nesmeyana (who laughs only after endless tears) , and other Russian folklore icons.
В музее повсюду стоят артефакты знакомые с пеленок: сапоги-скороходы, ковер-самолет, меч-кладенец, скатерть-самобранка.	The grounds brim with childhood-familiar artifacts: seven-league boots, flying carpet, treasure sword, and self-spreading tablecloth .	The grounds brim with magical artifacts from the tales : seven-league boots (for super-speedy travel), a flying carpet, a gleaming treasure sword, and a selfspreading tablecloth (that instantly serves a feast).

Traction Man → Тракшн Мэн → Тракшн Мэн (герой приключенческих книжек о супергерое из игрушек) ; Elmer → Элмер → Элмер (лоскутный слоник из книжек Дэвида Макки) . In some cases, epithets are added, which makes the text more expressive and contributes to its better understanding: The Story Bus → Автобус историй → волшебный Автобус историй ; The Night Safe → безопасная ночная зона → уютная «Ночная зона безопасности».

The chatbot’s comment on the adapted translation is as follows: Here’s a pragmatically adapted Russian translation, taking into account that Russian readers may not be familiar with English children’s books and characters. I’ve added subtle parentheses for key titles and contexts to ensure the text remains natural, engaging, and understandable without overwhelming the reader.

Despite the adaptation, the text contains some violations of the translation language norms. For example, instead of the phrase игрушечный супергерой , the bot used the phrase супергерой из игрушек ; the phrase подводный мир под раковиной is divided by the parenthetical comment – откройте подводный мир Тракшн Мэна (героя приключенческих книжек о супергерое из игрушек) под раковиной ; the expression безопасная ночная зона does not comply with the text style.

Conclusions

Thus, studying the potential of AI-powered chatbots for translating folklore museum promotional texts suggests that the effectiveness of an iterative approach to interacting with the bot as this allows for the evaluation of interim results and the adjustment and supplementation of instructions at subsequent stages of the translation process. Pragmatic adaptation of texts for foreign target audiences is not part of the bot’s functionality for professional translation requests. However, subsequent requests result in adjustments to the translation results, including descriptive translation of culture-specific lexical units, explication of allusions, addition of epithets, and the use of functional equivalents. Despite a number of violations of the linguistic norms of the translation language and complex syntactic structures with a large number of lists and comments, as a result of adaptation, the text becomes more understandable for representatives of other cultures through the explication of the semantics of foreign-language inclusions and implements the informative, appellative and expressive functionality.

It was found that an effective prompting algorithm for translation and pragmatic adaptation of museum promotional texts in the Perplexity chatbot includes a sequence of two prompts: firstly, an indication of the bot’s role and the formulation of the translation task, and, secondly, a command to perform pragmatic adaptation and

Table 3. The source text and neural translation of the Story Museum promotional text (Iterations 1 and 2)

Source (The Story Museum) Translation by Perplexity (iteration 1) Translation + pragmatic adaptation by Perplexity (iteration 2) Explore five story-themed zones where you can go on a Bear Hunt, discover Traction Man’s under-sink world, relax in the Night Safe, ride on the Story Bus and enjoy a bedtime story our giant bed. Исследуйте пять зон, оформленных в стиле любимых историй: отправьтесь на охоту за медведем, откройте подводный мир Тракшн Мэна под раковиной, отдохните в безопасной ночной зоне, прокатитесь на Автобусе историй и послушайте сказку на ночь в нашей гигантской кровати. Исследуйте пять зон, оформленных в стиле знаменитых историй: отправьтесь на поиски медведя (по книге «Охота за медведем» Майкла Роуза), откройте подводный мир Тракшн Мэна (героя приключенческих книжек о супергерое из игрушек) под раковиной, отдохните в уютной «Ночной зоне безопасности», прокатитесь на волшебном Автобусе историй и послушайте сказку на ночь в нашей гигантской кровати. Our most recent addition is an Elmer storytelling space, created in partnership with Andersen Press to mark the 35th anniversary of the beloved picture books by David McKee. Наша новейшая зона – пространство для рассказов про Элмера, созданное в партнерстве с издательством Andersen Press в честь 35-летия любимых книжек с картинками Дэвида Макки. Наша новейшая зона – пространство для рассказов про Элмера (лоскутного слоника из книжек Дэвида Макки), созданное в партнерстве с издательством Andersen Press в честь 35-летия этих любимых детских книг. take into account the limited knowledge of the foreign target audience.

A promising study is a comparative assessment of the effectiveness of a single-level prompt, which includes all the necessary information, and a sequence of prompts that allow for the evaluation of intermediate results and adjustment of requests.

NOTE

1 The research was supported by Russian Science Foundation (project No. 25-28-00223) “Translatological substantiation of multilinguality of presentational content in the AI paradigm” .