TRANSFORMING PROFESSIONAL TRANSLATOR COMPETENCIES IN THE CONTEXT OF WIDESPREAD ADOPTION OF AI TECHNOLOGIES AND CAT TOOLS

ABSTRACT

The rapid evolution of artificial intelligence technologies, primarily neural machine translation and large language models, in 2024–2025 has led to a qualitative restructuring of the language services industry. The study conducted within the framework of this work is focused on identifying and interpreting this shift through the lens of measurable changes in translation productivity and cognitive load, as well as on substantiating the need to update the professional profile in view of new technological and ethical constraints. Market conditions demonstrate a trajectory that is seemingly contradictory at first glance: amid a slowdown of the overall compound annual growth rate of the industry to 5,0%, an accelerated expansion of the segment of artificial intelligence application in translation is observed with a CAGR of 25,2%. A significant result is the empirically confirmed reduction in time expenditure for machine translation post-editing when using quality estimation systems: the average decrease is about 25%—from 1,27 to 0,95 sec/word, while a reduction in cognitive load is simultaneously recorded, including according to eye-tracking data. The recorded transformation of functionality determines a shift in the professional role from the predominant production of text toward technological integration, data curation, and the conduct of ethical audit. In this context, the UNESCO AI Competency Framework (2024) is considered as a supporting foundation of an updated pedagogical paradigm, in which priority is shifted to the ethical and normative dimension, the principles of systemic design, and critically validated human–machine interaction.

АННОТАЦИЯ

Быстрое развитие технологий искусственного интеллекта, прежде всего нейронного машинного перевода и больших языковых моделей, в 2024–2025 годах привело к качественной реструктуризации индустрии языковых услуг. Исследование, проведенное в рамках данной работы, направлено на выявление и интерпретацию этого сдвига через призму измеримых изменений в производительности перевода и когнитивной нагрузке, а также на обоснование необходимости обновления профессионального профиля с учетом новых технологических и этических ограничений. Рыночные условия демонстрируют траекторию, которая на первый взгляд кажется противоречивой: на фоне замедления общего среднегодового темпа роста отрасли до 5,0% наблюдается ускоренное расширение сегмента применения искусственного интеллекта в переводе со СГР в 25,2%. Важным результатом является эмпирически подтвержденное сокращение временных затрат на постредактирование машинного перевода при использовании систем оценки качества: среднее снижение составляет около 25% - с 1,27 до 0,95 сек/слово, при этом одновременно фиксируется снижение когнитивной нагрузки, в том числе по данным отслеживания движений глаз. Зафиксированная трансформация функциональности определяет сдвиг в профессиональной роли от преимущественного создания текста к технологической интеграции, обработке данных и проведению этического аудита. В этом контексте Рамочная программа компетенций ЮНЕСКО в области ИИ (2024) рассматривается как основа для обновленной педагогической парадигмы, в которой приоритет смещается на этический и нормативный аспект, принципы системного проектирования и критически обоснованное взаимодействие человека и машины.

Keywords: Artificial intelligence, machine translation, post-editing, translator competencies, cognitive load, UNESCO, technological literacy, LLM.

Ключевые слова: искусственный интеллект, машинный перевод, постредактирование, компетенции переводчика, когнитивная нагрузка, ЮНЕСКО, технологическая грамотность, магистр права.

Introduction

Globalization processes and the accelerating production of digital content have generated an unprecedented demand for prompt and simultaneously high-quality multilingual translation [6]. Over the past decade, the language services industry (Language Service Providers, LSP) has undergone a systemic reconfiguration: statistical machine translation (SMT) has been displaced by neural architectures (NMT), and in the most current phase an increasing integration of the potential of large language models (LLM) is observed [8]. These technological shifts have changed the very logic of translation labor, gradually reorienting practice from traditional translation (Human Translation, HT) toward machine translation post-editing (MTPE) as the dominant mode of work [7].

In 2024–2025, an understanding has become established that the professional value of the translator is determined to a lesser extent by creating text from scratch, and to a greater extent by verification procedures, quality control, and management of a technologically mediated process. Against the background of continuous improvement of algorithms, the role of the translator as the sole source of the target text loses sufficiency and becomes one component of a more complex production configuration. Under these conditions, demand shifts toward hybrid roles, including prompt engineering as optimization of input for LLM, data management, and the provision of clean data services suitable for training and fine-tuning models [1]. Consequently, research relevance is determined by the necessity of developing a new, empirically verifiable competency model capable of adequately describing and standardizing the resulting technological turning point.

The market dynamics of 2024–2025 reveal an asymmetry that is fundamentally important for interpreting professional transformation. On the one hand, the language services sector continues to expand: the market volume reached 71,7 billion US dollars in 2024, and for 2025 a value of 75,7 billion dollars is forecast [1]. At the same time, a slowdown in pace is recorded: prior to the widespread adoption of artificial intelligence, the compound annual growth rate (CAGR) was estimated at 7,0%, whereas the latest forecasts have been reduced to 5,0% [1]. This slowdown largely correlates with the cannibalization of routine components of translation labor by automated solutions, which displace low-margin operations and compress traditional employment segments.

On the other hand, the technological segment of artificial intelligence in translation demonstrates a different trajectory, close to exponential. The artificial intelligence translation market, valued at 2,34 billion US dollars in 2024, is forecast at 2,94 billion US dollars in 2025 with a compound annual growth rate (CAGR) of 25,2% [2]; it is expected that by 2029 it will reach 7,16 billion US dollars [2]. On the LSP side, revenue generation centers are concentrated around machine translation and post-editing, prompt engineering, as well as offerings related to clean technological services and data [1]. Such bifurcation indicates a structural redistribution of economic value: the need for human expertise does not disappear; however, the value created shifts into the domain of technological management, quality control, and work with data. As a result, the key figure becomes a specialist capable of integrating tools, optimizing production chains, and ensuring reproducible quality under conditions of intensifying competition and accelerated delivery cycles.

The identified technological changes objectively necessitate a revision of educational trajectories and professional standards. The central research problem is formulated as ensuring the correspondence of translators’ competencies to changed market requirements while simultaneously observing ethical responsibility and maintaining quality in specialized subject areas. Despite substantial progress in NMT, difficulties of domain mismatch remain, problems of translating rare words persist [10], as well as the unresolved nature of tasks of human-aligned evaluation as a criterion that aligns machine solutions with the expectations of professional communication [8].

The purpose of the study is to identify how the widespread implementation of NMT/LLM and CAT in recent years transforms the professional competencies of the translator through measurable changes in MTPE productivity and cognitive load, as well as to substantiate the necessity of updating the professional and educational profile with consideration of new technological and ethical constraints.

Scientific novelty is reduced to the fact that the work empirically links the effects of QE-supported post-editing (acceleration of MTPE and reduction of cognitive load according to eye-tracking data) with the author’s proposed reconfiguration of the translator competency model on the basis of the UNESCO AI Competency Framework (2024), fixing the role shift toward technological integration, data curation, and ethical audit.

The author’s hypothesis is based on the assumption that the implementation of QE-oriented MTPE in combination with LLM and CAT statistically reduces extraneous cognitive load and segment processing time, redistributing the translator’s efforts toward semantic validation, domain accuracy, and responsibility management, which requires targeted expansion of competencies (prompt engineering, QA/QE solutions, data curation, ethical and legal control).

The views expressed herein are those of the author and should not be attributed to the IMF, its Executive Board, or its management.

Materials and Methods

The methodological foundation of the study is constructed as a comprehensive analytical design combining the systematization of scholarly publications and the interpretation of statistical data arrays, which makes it possible to reconstruct the transformation of the translation profession under conditions of convergence of neural machine translation (NMT), large language models (LLM), and automation tools (CAT). As a basic premise, an interdisciplinary perspective is adopted, integrating the instruments of translation studies, cognitive psychology, and the analysis of industrial data of the language services sector (LSP).

The theoretical and methodological framework is formed at the intersection of three complementary directions. The technological direction is focused on describing the evolution from classical NMT architectures to regimes in which NMT effects are strengthened through synergy with LLM. Within this framework, mechanisms are examined by which LLM reduce dependence on large parallel corpora [8], increase robustness to increases in the length of syntactic constructions (on the order of 80 words), and support coherence when expanding context to the document level (up to 512 words) [8]. A key point is the inclusion of human-aligned evaluation as a methodological response to the risk of hidden semantic defects and the phenomenon of AI hallucinations, when formal linguistic fluency does not guarantee the truth-conditional and referential correctness of content [10].

The normative and competency-based direction establishes a comparative framework uniting the requirements of international post-editing standards (ISO 18587) and current educational guidelines, including the UNESCO AI Competency Framework (2024). In this logic, levels of professional intervention are differentiated from light post-editing (LPE) to full post-editing (FPE) and further to adaptive models, where the translation function expands to data curation and the performance of the role of an arbiter of meaning in situations of probabilistic generation [13, 14]. Thus, the competency profile is described as a transition from performing an operation to managing the quality and normative acceptability of the outcome within a hybrid system.

Cognitive direction relies on Cognitive Load Theory (CLT) as an explanatory model of MTPE efficiency. Within CLT, the redistribution of extraneous, intrinsic, and germane components of load is emphasized; the automation of low-level operations is considered a factor that frees working memory resources, the volume of which is typically described by the range of 5–9 units [17], for solving higher-order tasks—stylistic adjustment, cultural and pragmatic adaptation, and the resolution of ambiguities [18]. This interpretation provides a theoretical linkage between the technological characteristics of systems and changes in the profile of mental effort in translation activity.

The empirical base is formed on the basis of secondary data extracted from specialized studies and industry statistics from recent years. The selection of the specified time frame serves as a chronological filter that makes it possible to capture the stage of mass integration of LLM into production chains and the related changes in organizational and cognitive parameters of work.

Source selection was carried out according to an inclusion and exclusion protocol ensuring data comparability and reproducibility of analytical conclusions. Priority was given to peer-reviewed publications (Scopus/WoS), conference proceedings on NLP and HCI, as well as reports by leading analytical structures specializing in the LSP/AI segments. Studies with methodological transparency were included in the sample, that is, with an explicit description of measurement procedures (sample size and characteristics, MTPE/QE protocols, eye-tracking parameters and modes). Sources without dating, without a verifiable calculation methodology, or without explication of the metrics used were excluded. In the presence of duplicate data, preference was given to more recent materials and works with stricter procedural specification.

The toolkit for assessing the transformation of labor costs and market dynamics was operationalized through several groups of indicators. The productivity block included time-per-word indicators in MTPE, as well as the impact of Quality Estimation (QE) tools on decision-making speed and the economic feasibility of task segmentation and routing [19]. Objective cognitive indicators included eye-tracking metrics, primarily parameters of saccades and fixations, used as proxy indicators of mental effort during verification of NMT/LLM output and attention management in the editing process [4]. At the macro level, the volumes of the language services market, the dynamics of segmentation of AI solutions, and the compound annual growth rate (CAGR) in the automated translation sector were analyzed.

Such a methodological configuration ensures simultaneous fixation of the state of the industry and testing of the hypothesis of a structural change in the professional role: translation activity in hybrid human–machine systems shifts from linear execution to a high-level expert function of validation, technological management, and semantic control.

Results and Discussion

Empirical materials of recent years demonstrate that machine translation post-editing (MTPE) transforms not only the operational structure of translation activity but also provides statistically observable gains in productivity and cognitive efficiency compared with traditional human translation (HT). The observed effect has a dual nature: acceleration of task performance is combined with a change in the profile of mental effort, redistributed among components of cognitive load.

The results of MTPE studies (2025) indicate a pronounced influence of quality estimation (QE) systems applied at the sentence level on the temporal effectiveness of post-editing. A significant reduction in average time costs has been recorded: when performing MTPE without QE, the indicator was 1,27 seconds per word, whereas with activated QE it decreased to 0,95 seconds per word [3]. Thus, this represents a reduction in time expenditure of approximately 25%, which confirms the functional role of QE as a tool that optimizes decision making and attention routing in the editing process.

Additional analytical examination shows that the positive effect of QE remains stable regardless of the initial level of machine translation quality—both under medium and under high characteristics of MT output [3]. This result indicates broad applicability of QE as a universal accelerator of the production cycle, rather than a narrowly specialized means of compensating for weak NMT quality. At the same time, the temporal efficiency of QE is determined by a critically important mediator—technological trust. With inaccurate or unstable predictions, the system provokes distrust and, as a consequence, compensatory behavior in the form of double checking, which reduces work speed and worsens user experience [3]. Consequently, maximizing the effect of QE presupposes not only an increase in its predictive accuracy, but also the formation in the specialist of a meta-evaluation competency: a critically calibrated attitude toward technological output, taking into account its probabilistic nature and limitations of applicability [26, 27].

The transition to MTPE is accompanied by a noticeable change in the distribution of cognitive resources, which is confirmed in objective psychophysiological metrics. Eye-tracking studies show that in the human–machine interaction mode there is a decrease in indicators correlating with the intensity of search, attention switching, and the volume of online information processing compared with HT. In particular, when performing the HT task, the average number of saccades was 18, whereas in MTPE implementation it decreased to 15 [4]. Similarly, the number of fixations decreased, reflecting moments of sustained concentration and active processing in working memory: from 29 to 23 [4]. The reduction in the frequency of saccades and fixations is interpreted as a direct indicator that a significant part of the extraneous cognitive load associated with routine comparison of source and target material and primary generation of structure is transferred to the artificial intelligence system. As a result, human cognitive capacity is freed for higher-level tasks—strategic analysis, cultural and pragmatic adaptation, and stylistic refinement, that is, for those components of germane load that form the added value of professional expertise in the final product.

At the same time, it has been established that the reduction of subjective and objective complexity of MTPE compared with HT is most pronounced not under all conditions, but predominantly in a configuration where high quality of NMT output and increased complexity of the source text are combined [20]. This result indicates the contextual conditionality of cognitive advantages: technological support manifests maximal effectiveness where automation removes the greatest volume of low-level operations, while not eliminating the need for expert semantic and pragmatic validation.

Below, Table 1 will be presented, containing key indicators of productivity and cognitive effort obtained in studies [3-5; 20].

Table 1.

Comparative analysis of productivity and labor costs: Traditional translation (HT) vs. MTPE with consideration of quality estimation (QE) systems (compiled by the author on the basis of [3-5; 20]).

The presentation of the dynamics of the language services market in Figure 1 serves as confirmation of the strategic shift in the industry’s focus.

Figure 1. Growth dynamics of the artificial intelligence translation market (CAGR) (2024–2029) (compiled by the author on the basis of [3-5; 20]).

A conceptual figure demonstrating the slowdown of the overall CAGR of the LSP industry (5.0%) against the background of the exponential growth of the artificial intelligence segment in translation (25.2%), which confirms market bifurcation and the redistribution of investments toward technological solutions [1].

As artificial intelligence systems take over routine operations, the profile of the translator’s professional competencies shifts toward an integrated model that includes three interrelated dimensions: deepening linguistic expertise, developing technological literacy, and strengthening capabilities for strategic management of the production process. Within this configuration, human value is determined not by the volume of mechanically performed work, but by the quality of expert decisions made at the intersection of meaning, context, and technological constraints.

Despite the impressive achievements of automated solutions, the human role remains indispensable in tasks that require a high level of interpretation, cultural and pragmatic awareness, and emotional intelligence [9]. This is especially evident in high-risk domains—legal, medical, financial, and literary translation—where the cost of an error goes beyond stylistic inaccuracy and can transform into legal, clinical, reputational, or economic consequences. In these areas, critical thinking and contextual understanding remain critically significant, and they are not fully reproduced by artificial intelligence, even with high formal fluency of the generated text [7, 12]. In such a context, the translator functionally acts as a filter of responsibility, providing expert verification not only of linguistic correctness, but also of the normative, genre, and procedural adequacy of the result: in legal materials, correlation with jurisdiction-conditioned terminology and established formulations is required, while in medical texts the priority is clarity and compliance with international regulatory expectations [9]. The implementation of this function presupposes pronounced bilingual competence, reliance on developed skills of semantic reconstruction, and deep cultural awareness as a factor that distinguishes an acceptable result from a high-quality one [21, 25]. In practical terms, human intervention is concentrated on high-level errors—terminological inaccuracy, stylistic dissonances, and cultural inappropriateness—that is, on components directly correlating with the categories Accuracy, Terminology, and Style [16].

Alongside this, within the contemporary professional standard, technological literacy acquires the status of a basic requirement and extends beyond skills of working with CAT tools. Prompt engineering is identified as a critically significant element, understood as the purposeful optimization of input for LLM in order to increase the quality and controllability of output [1]. At the same time, the logic of workflow management changes: central becomes not the linear execution of an order, but strategic decision making based on diagnosing MT quality and selecting an adequate intervention model. The practice of MTPE presupposes rapid assessment of the feasibility and depth of editing, followed by the selection of the level—light (LPE), full (FPE), or adaptive—depending on the quality of the initial machine output, requirements for the final product, and the economic constraints of the project [14]. This capability relies on confident command of a wide set of digital tools and methods, which, as noted in studies, increases overall effectiveness both of professional training and of applied activity in a production environment [22].

Table 2 presents a classification and quality requirements for machine translation post-editing.

Table 2.

Classification and quality requirements for machine translation post-editing (MTPE) (compiled by the author on the basis of [4, 9, 15, 16, 28]).

A radical restructuring of professional expectations regarding translation activity necessitates the prompt adjustment of educational programs. The conceptual basis for such an adjustment is the AI Competency Framework for Students, published by UNESCO in September 2024, which enshrines the human-centered principle of integrating artificial intelligence into the educational environment [5]. Within this approach, artificial intelligence is considered not as an autonomous substitute for human expertise, but as a sociotechnical tool, the use of which must be correlated with human needs, risks, and norms of responsibility.

The UNESCO 2024 architecture is described through a two-dimensional model: four interrelated competency domains and three levels of mastery. Applied to the training of Master of Translation and Interpreting (MTI) students, this means the purposeful formation of a competency profile that includes, first, human-centered thinking regarding artificial intelligence—development of critical analysis, ethical sensitivity, and understanding of the connections between technological solutions, human needs, and dimensions of social justice [5]. Second, artificial intelligence ethics is positioned as an independent core of training: mastery of principles of responsible use, procedures for accounting for and reducing bias, and practices for ensuring fairness and a non-discriminatory effect of technological solutions [5]. Third, a block titled AI Technologies and Applications is specified, which presupposes confident command of applied tools, including prompt engineering and management of technologically mediated workflows [1]. Fourth, it includes competencies related to the design of artificial intelligence systems: understanding of principles of development, configuration, and optimization of systems, as well as the logic of their implementation within organizational and production contours [5, 11].

A substantial emphasis in UNESCO 2024 is placed on the gradation of mastery levels: Understanding, Application, and the highest level—Creation [5]. Achieving the Creation level in the professional and educational projection means a transition from consumption of technological solutions to active participation in their improvement and contextual optimization. In the translation domain, this correlates with the formation of the role of a model curator, implying contribution to the configuration of feedback loops and the development of adaptive post-editing (Adaptive PE), as well as participation in ensuring that training and fine-tuning data are ethically correct, representative, and domain-relevant [5]. Such a shift institutionalizes the translator’s responsibility not only for the final text, but also for the quality of the sociotechnical infrastructure through which this text is produced.

Table 3 demonstrates the adapted structure of translator competencies in accordance with the UNESCO artificial intelligence foundation.

Table 3.

Adapted structure of translator competencies in accordance with the UNESCO AI Framework (compiled by the author on the basis of [1, 5, 23]).

The large-scale integration of artificial intelligence into translation practice, especially in sensitive and high-risk domains, raises a complex of ethical and legal issues that extend beyond purely technological efficiency. Among the key ethical problems are ensuring data confidentiality, observing the principles of academic integrity, and maintaining fairness in the application of artificial intelligence [23]. Under conditions in which large language models (LLM) operate with data and probabilistic regularities embedded in training corpora, the significance of the risk of algorithmic bias increases. The presence of such biases requires institutionalization of the translator’s role as an ethical auditor, capable of identifying biased patterns, assessing their consequences for the meaning and pragmatics of the message, and initiating corrective actions within the framework of quality control.

Responsible application of artificial intelligence technologies presupposes the development of legal frameworks and ethical guidelines that possess both normative definiteness and adaptability sufficient to regulate rapidly evolving systems while maintaining a balance of the rights and interests of all involved parties [24]. In the hybrid production model, the question of the distribution of responsibility remains fundamental: despite the participation of machine suggestions and automated generations, final responsibility for the textual product remains with the human. This circumstance intensifies the need for a clear legislative definition of the boundaries and conditions of responsibility when using machine suggestions in high-risk domains, where an error can have legal, clinical, or financial consequences, as well as in situations related to the processing of personal and confidential data.

Figure 2 describes a model of hybrid human–artificial intelligence interaction, recording the transition from a linear work scheme oriented toward direct generation to a cycle of managed production, including strategic decision making, data curation, quality assurance, elements of model training, and ethical audit (Strategy, Curation, QA, Training) [9].

Figure 2. Description of the hybrid human–artificial intelligence interaction model (compiled by the author on the basis of [9]).

As can be seen from Figure 2, in the presented model translation activity is redefined as a set of managerial and expert-evaluative functions that ensure reproducible quality and normative acceptability of the result under conditions of a technologically saturated environment.

Thus, MTPE becomes not merely an accelerating instrument, but a systemic transformation of translation labor: with sentence-level QE enabled, a stable reduction of working time by approximately one quarter is recorded (1,27 → 0,95 sec/word), and the effect is maintained under different initial levels of NMT quality, but depends on technological trust (erroneous QE predictions provoke double checking and neutralize the gain). At the same time, the structure of cognitive load changes: according to eye-tracking, indicators of saccades and fixations decrease (18→15; 29→23), which is interpreted as the transfer of low-level operations to artificial intelligence and the freeing of human resources for semantic, pragmatic, and stylistic expertise; at the same time, maximal cognitive benefit manifests in the combination of high MT quality plus a complex source text. At the level of the industry and competencies, this leads to market bifurcation (growth of the artificial intelligence segment with slower growth of LSP overall) and to a redefinition of the translator’s role: from a text executor to a technological integrator, a quality strategist, and a filter of responsibility in high-risk domains, where contextual understanding, cultural validity, and ethical and legal control are irreplaceable. However, the stability and scaling of MTPE advantages require not only improvement of NMT and QE, but also an institutional update of translator training (in the spirit of UNESCO 2024) with an emphasis on critical thinking, artificial intelligence ethics, process management, and participation in the system improvement loop (feedback, data curation, QA), since final responsibility for the result remains with the human.

Conclusion

Artificial intelligence has redefined the professional contour of translation activity, shifting it from predominantly linguistic generation toward technologically mediated management, quality control, and critical verification. Empirical results of 2024–2025 confirm the applied and economic validity of this shift: a substantial increase in productivity is recorded due to a reduction of MTPE time by approximately 25% when using quality estimation (QE) systems, as well as an objective reduction of cognitive load expressed by a decrease in saccade and fixation indicators at the level of 17–20%. At the same time, a fundamental conditionality of the achieved advantages is revealed: maximal effectiveness is realized under a combination of high quality of machine output and strategically calibrated human control. Within this configuration, the core of professional value is concentrated around competencies that are not amenable to full technological replication: cultural and pragmatic adaptation, reconstruction of emotional context, domain expertise, and, first and foremost, ethical and legal responsibility for the final result. Accordingly, the expert function of the translator is increasingly determined by ensuring semantic accuracy and contextual adequacy, whereas formal syntactic correctness ceases to be the leading quality criterion.

The prospective trajectory of the profession is formed as hybrid and human-centered. Its foundation is critically calibrated human–machine interaction (HMI), in which the professional retains an active role not only in the exploitation of tools, but also in the loop of their configuration, training, and audit. The highest levels of competence specified by UNESCO 2024 set a normative benchmark that presupposes mastering technological means simultaneously with understanding the ethical foundations and social consequences of their application. Such a framework logic supports the preservation of human subjectivity under conditions of widespread artificial intelligence adoption, institutionalizing new professional niches: data curation, prompt engineering, and consulting on issues of ethics and responsibility in translation processes.

On the basis of the conducted analysis, a number of practice-oriented recommendations are formulated for key stakeholders—educational organizations, technology developers, and regulatory institutions. The priority direction is the modernization of MTI training: integration of the UNESCO AI Competency Framework (2024) into curricula is required, with a shift of emphasis from exclusively traditional translation to MTPE, prompt engineering, and, especially, competencies in the field of artificial intelligence ethics and systemic design; particular attention is required for developing the ability for meta-evaluation of technological output as a condition of sustainable technological trust. In the technological track, it is advisable to concentrate efforts on increasing QE accuracy, since unreliable assessments provoke compensatory rechecking and can slow down rather than accelerate the workflow; in parallel, the development of unified human-aligned evaluation metrics is timely, allowing adequate evaluation of LLM output quality in narrowly specialized domains. In the regulatory dimension, comprehensive and adaptive legal and regulatory solutions are necessary, defining the boundaries of responsibility within the hybrid production loop, ensuring data protection and minimization of algorithmic bias, including in adaptive translation scenarios.

Prospects for further research are associated with the analysis of long-term effects of reducing extraneous cognitive load on the translator’s creative and strategic potential, as well as with the development of valid methodologies for assessing the effectiveness of forming Creation-level competencies within MTI programs, including measurement of learners’ contribution to improving feedback loops and the quality of data used in educational and production scenarios.

References:

1. The 2025 Nimdzi 100 | Nimdzi Insights. Retrieved from: https://www.nimdzi.com/nimdzi-100-2025/ (date accessed: September 1, 2025).
2. AI in Language Translation Global Market Report 2025 | The Business Research Company. Retrieved from: https://www.thebusinessresearchcompany.com/report/ai-in-language-translation-global-market-report (date accessed: September 3, 2025).
3. Béchara, H., Ladhak, F., & Jamshid Lou, J. (2021). The role of translation technologies in improving the process of teaching translation: Toward a more effective approach. Informatics, 8(3), 61. https://doi.org/10.3390/informatics8030061
4. Chen, L. (2025). Examining cognitive load in human-machine collaborative translation: Insights from eye-tracking experiments of Chinese-English translation. Frontiers in Psychology, 16, 1570929. https://doi.org/10.3389/fpsyg.2025.1570929
5. Tian, Y. (2025). Enhancing MTI translation competence in the era of AI. Cultural and Religious Studies, 13(4), 201–208. https://doi.org/10.17265/2328-2177/2025.04.007. Retrieved from: https://www.davidpublisher.com/Public/uploads/Contribute/6811c41cefab6.pdf (date accessed: September 7, 2025).
6. 2025 ELIS Report | European Language Industry Survey. Retrieved from: https://elis-survey.org/wp-content/uploads/2025/04/ELIS_2025_Report_EN.pdf (date accessed: September 9, 2025).
7. Jiang, L., Wei, B., & Al-Shaibani, G. K. S. (2025). Effective neural machine translation with human post-editing of Chinese intangible cultural heritage corpus into English. SAGE Open, 15(4). https://doi.org/10.1177/21582440251386954
8. Pang, J., Ye, F., Wong, D. F., Yu, D., Shi, S., Tu, Z., & Wang, L. (2025). Salute the classic: Revisiting challenges of machine translation in the age of large language models. Transactions of the Association for Computational Linguistics, 13, 73–95. https://doi.org/10.1162/tacl_a_00730. Retrieved from: https://aclanthology.org/2025.tacl-1.4.pdf (date accessed: September 11, 2025).
9. Bartlett, R. (2025, October 11). The future of AI translation: A world without language barriers? Financial Times. Retrieved from: https://www.ft.com/content/25a36cc3-9691-4fbb-96d2-916993161dae (date accessed: October 15, 2025).
10. Centrone, S., Perini Brogi, C., & Reifberger, S. (2024, December). Machine Translation: Early Criticisms Revisited. In Proceedings of the 2024 Conference on Human Centred Artificial Intelligence-Education and Practice (pp. 51-55).
11. ISO 18587:2017 — Translation services — Post-editing of machine translation output — Requirements | ISO. Retrieved from: https://www.iso.org/standard/62970.html (date accessed: September 15, 2025).
12. Lei, T., & van de Weijer, J. (2025). Evaluating different translation methods: A case study of Chinese graduate students in an MTI program. Cogent Arts & Humanities, 12(1), 2511386. https://doi.org/10.1080/23311983.2025.2511386
13. Post-editing | Wikipedia. Retrieved from: https://en.wikipedia.org/wiki/Postediting (date accessed: September 19, 2025).
14. MT Post-Editing Guidelines | TAUS. Retrieved from: https://www.taus.net/resources/reports/mt-post-editing-guidelines (date accessed: November 2, 2025).
15. Kenny, D., & Doherty, S. (2014). Statistical machine translation in the translation curriculum: Overcoming obstacles and empowering translators. The Interpreter and Translator Trainer, 8(2), 276–294. https://doi.org/10.1080/1750399X.2014.936112
16. Vieira, L. N. (2014). Indices of cognitive effort in machine translation post-editing. Machine Translation, 28(3–4), 187–216. https://doi.org/10.1007/s10590-014-9156-x
17. Cognitive Load Theory | Medical College of Wisconsin. Retrieved from: https://www.mcw.edu/-/media/MCW/Education/Academic-Affairs/OEI/Faculty-Quick-Guides/Cognitive-Load-Theory.pdf (date accessed: September 27, 2025).
18. Baxter, K. A., Sachdeva, N., & Baker, S. (2025). The application of cognitive load theory to the design of health and behavior change programs: Principles and recommendations. Health Education & Behavior, 52(4), 469–477. https://doi.org/10.1177/10901981251386406
19. Shenoy, R., Herbig, N., Krüger, A., & van Genabith, J. (2021). Investigating the helpfulness of word-level quality estimation for post-editing machine translation output. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (pp. 10173–10185). Association for Computational Linguistics. https://doi.org/10.18653/v1/2021.emnlp-main.799
20. Jia, Y., & Sun, S. (2023). Man or machine? Comparing the difficulty of human translation versus neural machine translation post-editing. Perspectives, 31(5), 1–19. https://doi.org/10.1080/0907676X.2022.2129028
21. García, I. (2011). Translating by post-editing: Is it the way forward? Machine Translation, 25(3), 217–237. https://doi.org/10.1007/s10590-011-9115-8
22. Chan, V. (2025). Enhancing translation technology skills through blended learning: A study of student perceptions and satisfaction with bMOOC in higher education. Frontiers in Education, 10, 1613880. https://doi.org/10.3389/feduc.2025.1613880
23. Rismani, E., & Köprülü, F. (2025). Ethical dimensions of artificial intelligence in educational technology and policy: A global bibliometric analysis (2020–2025). Qeios. https://doi.org/10.32388/MFJSM4.2
24. Al-kfairy, M., Mustafa, D., Kshetri, N., Insiew, M., & Alfandi, O. (2024). Ethical challenges and solutions of generative AI: An interdisciplinary perspective. Informatics, 11(3), 58. https://doi.org/10.3390/informatics11030058
25. Aboudaif, S. A. (2025). Translation education in the artificial intelligence era. Journal of Languages and Translation, 2(1), 1–8. https://doi.org/10.21608/cpijlt.2025.460071
26. Machine Translation Market Growth Analysis — Size and Forecast 2025–2029 | Technavio. Retrieved from: https://www.technavio.com/report/machine-translation-market-industry-analysis (date accessed: October 13, 2025).
27. Sun, S., Wang, H., & Jia, Y. (2025). Direction matters: Comparing post-editing and human translation effort and quality. PLOS ONE, 20(7), e0328511. https://doi.org/10.1371/journal.pone.0328511