Transforming Translation Models Driven by Technology in the AI Era

In the era of artificial intelligence, the rapid advancement of translation technology has greatly promoted the development of the translation industry, leading to profound changes in translation models. This article first outlines the evolution of technology-driven translation models, namely Computer-Aided Translation—Machine Translation Post-Editing—Interactive Machine Translation. It then explores the driving forces behind the evolution of translation models from three aspects: language service demand, technological innovation, and socio-cultural factors. Finally, it proposes suggestions for establishing a harmonious human-machine ecosystem, promoting human-centered technological innovation, emphasizing translation data security and technological ethics, and enhancing translators’ technical skills and cultural literacy.

Keywords: Translation Models; Evolution; Computer-Aided Translation; Post-Editing; Interactive Machine Translation

Entering the 21st century, human society has gradually stepped into the era of artificial intelligence (AI). New technologies such as natural language processing, pattern recognition, machine learning, and visual perception have surged, increasingly permeating various fields of the economy and society. The rapid advancement of artificial intelligence technology has greatly propelled the development of the translation industry, giving rise to a host of new technologies such as translation memory, terminology management, and neural machine translation, resulting in a revolutionary change in translation production models. The traditional pure manual translation model can no longer meet the ever-growing industry and market demands, while the technology-driven translation model has gradually been accepted and recognized by the industry and widely applied in practice. The tremendous transformation of translation models has overturned people’s traditional understanding of the translation industry while injecting fresh blood into its development. What are the forms of technology-driven translation model evolution in the AI era? What are the driving forces behind it? What implications does the evolution of translation models have for the development of the translation industry and translation education? Exploring and answering these questions not only contributes to the theoretical construction of translation models but also has important guiding and leading roles for the development of the translation industry and translation practice.

In translation studies, the concept of models is widely used but difficult to define. There is often ambiguity and inconsistency in the objects referred to by the term “translation models” in academia. Overall, research on translation models can be broadly categorized into four types:

2.1 Research on Translation Models Based on Strategies and Methods

In early translation theory research, translation models mainly referred to theoretical summaries based on translation practices regarding strategies and methods. Research on this type of translation model includes classifications of translation strategies or methods, such as domestication and foreignization (Venuti, 1995), communicative and semantic translation (Newmark, 2001); translation methods and strategies for specific research objects, such as source concept translation models (Yuan Yanling, Ge Lingling, 2019), metaphor translation models (Wen Bing, Wang Binhua, 2021), and literary translation models (Lü Shisheng, 2013); as well as studies on translation strategies of specific translators, such as Lu Xun’s translation model (Wang Yougui, 2003), and Ge Haowen’s translation model (Liu Yunhong, Xu Jun, 2014). The focus or core of these model studies is the strategies or methods used in the translation process, which, despite differing emphases, are of significant guiding importance for translation practice.

2.2 Research on Translation Models Based on Cognitive Processes

Translation models centered on the translation process mainly provide formalized descriptions of the translation process or the cognitive thinking processes of translators. Researchers typically use flowcharts and other schematic methods to diagram certain processes and relationships in translation based on specific disciplinary theories. For example, Nida & Taber (1969) proposed a four-stage translation model “analysis-conversion-reconstruction-verification” based on transformational generative theory; Bell (1991) proposed a translation process model based on cognitive psychology and information processing theory, which uses semantic representation as a bilingual conversion intermediary; and Gutt (2004) combined relevance theory to propose the “relevance translation model.” Compared to translation models centered on strategies, these models focus on the translation process and the cognitive thinking of translators, getting closer to the essence of translation activities and helping to reveal the workings of the translator’s brain.

2.3 Research on Translation Models Based on Organizational Forms

Translation organizational forms refer to the cooperation methods or organizational structures among participants in translation projects or tasks involving multiple people. A typical representative of this translation model is collaborative translation, where participants in translation activities complete a translation task through cooperation, such as the “oral translation” model in Buddhist scripture translation (Fu Liangyu, 2005) and the “dialogue cooperation” model in academic translations (Wang Shanshan et al., 2020). Additionally, with the development of modern communication and information technology, the “crowdsourced translation” model, where internet users claim translation tasks from the client, has increasingly emerged and become a hot topic in translation research (Hu Anjiang, 2017). Such model studies focus on the core of translation organizational forms, emphasizing the cooperation forms among translators, reflecting the diverse organizational practices in translation activities.

2.4 Research on Translation Models Based on Technological Means

With the rapid development of artificial intelligence, translation technology has made significant progress and has been widely applied in translation practice. Against this backdrop, research centered on translation technology has increasingly gained attention. Early research on translation models mainly focused on Computer-Aided Translation (CAT), gradually expanding to Machine Translation Post-Editing (MTPE) and Interactive Machine Translation (IMT). Research topics mainly involve descriptions and introductions of various models (Bowker, 2002; Cui Qiliang, 2014; Kong Lingran, Cui Qiliang, 2018; Huang Guoping, 2017), as well as their specific applications in practice and education (Zhang Zheng, Zhang Shaozhe, 2012; Knowles et al., 2019), and comparisons of efficiency and quality among different models (Underwood et al., 2014; Wang Xiangling, Wang Tingting, 2019). Currently, research on technology-driven translation models has become one of the hot topics in translation studies, showing a vigorous development trend.

In summary, scholars at home and abroad have conducted a series of studies on translation models from different perspectives, which have extremely important reference value for the theoretical construction of translation models. Among them, research on technology-driven translation models has achieved fruitful results but also has certain shortcomings. Most early studies started from a micro perspective, limiting their content to the introduction and application of specific translation models or comparisons of translation efficiency or quality among different models. So far, few researchers have examined the morphological evolution of translation models driven by technology and their underlying driving forces from a macro perspective, and a clear outline of the development of this emerging model has yet to be drawn.

Given this, this article will first introduce the morphological evolution of technology-driven translation models, then further analyze their driving forces, and explore the implications of translation model changes for the translation industry, translation education, and translator training.

In the era of artificial intelligence, innovations in translation technology have triggered disruptive changes in translation practice models. Translation technology refers to the collection of software, tools, facilities, environments, techniques, etc., applied in translation practice, research, and teaching, mainly including three categories: “Computer-Aided Translation Technology” (such as translation memory, terminology management, quality assurance), “Machine Translation Technology” (such as statistical machine translation, neural machine translation), and “Translation Management Technology” (such as workflow management, language asset management). Overall, the technology-driven translation model has mainly undergone the following evolution process:

3.1 From Manual Translation to Computer-Aided Translation

For thousands of years, manual translation has been the mainstream model of translation. Until the second half of the 20th century, as international communication deepened and expanded, the demand for translation surged, and traditional manual translation could no longer meet the growing industry demands. Additionally, at that time, the demand for software localization translation continued to grow; as software adopted a continuous iterative upgrade development model, the translated texts after each upgrade had a high repetition rate compared to the previous version, leading to a continuous increase in the demand for translation reuse. With advancements in computer technology, people began to attempt using various software and tools to assist in translation. Against this backdrop, Computer-Aided Translation (CAT) emerged. Unlike machine translation, CAT does not rely on automatic translation by computers; instead, it involves translators conducting translations with the assistance of translation software or tools. The history of computer-aided translation can be traced back to the 1990s, when the Swiss STAR company released the first commercial CAT tool, Transit, in 1991. Since then, various CAT software and tools have emerged rapidly and have been widely applied in the industry. Currently, mainstream CAT software or platforms include Trados, memoQ, Déjàvu, Wodfast, and YiCAT, with core technologies being Translation Memory (TM) and Term Base (TB). Translation memory can provide references for translators to complete current translation tasks, achieving parallel corpus resource sharing and translation reuse. Compared to traditional manual translation, CAT greatly reduces the time and effort required by translators, significantly improving translation efficiency while ensuring consistency in translation style and terminology usage. Although CAT liberates translators from heavy manual translation, its level of automation and intelligence is not high, urgently requiring a new model to enhance its production efficiency.

3.2 From Computer-Aided Translation to Machine Translation Post-Editing

Following CAT, machine translation technology made significant progress, especially since Google launched its neural machine translation (NMT) engine in 2016, the high usability of machine translation outputs has gradually been accepted and recognized by the industry, becoming increasingly favored. However, the output of machine translation cannot meet the high-quality practical demands and must undergo human editing to enhance translation quality, giving rise to a new translation model—Machine Translation Post-Editing (MTPE).

Machine Translation Post-Editing refers to the process of enhancing the output of machine translation through human and partially automated means to meet specific quality objectives (DePalma, 2014). In other words, post-editing is the process of editing, modifying, and processing the initial output of machine translation. Depending on the translation purpose, client needs, and the quality of machine translation output, translators can adopt light post-editing or full post-editing strategies. Compared to CAT, MTPE represents a significant advancement in production efficiency. Thanks to high-quality machine translation output, translators can complete translation tasks or projects in a shorter time, making post-editing an ideal choice for large-volume, low-cost translation projects, which has “already become a new trend in translation” (Li Mei, 2021:93). However, in industry practice, the post-editing model still faces many practical challenges. Currently, the quality of machine translation output still far fails to meet the needs of end clients. If the quality of machine translation output is too poor, correcting the numerous errors in the translation will consume even more time and effort from the translator (O’Brien, 2014). Furthermore, as post-editing heavily relies on machine translation, translators can only edit based on the output of machine translation, inevitably being influenced by the initial translation, leading to varying degrees of “machine translation tone.”

3.3 From Machine Translation Post-Editing to Interactive Machine Translation

In recent years, with the convergence of big data and artificial intelligence technologies, especially advancements in deep learning technology, machine translation technology has made new progress. The emergence of Interactive Machine Translation (IMT) has attracted widespread attention and has seen preliminary applications in the industry.

The concept of Interactive Machine Translation was first proposed by Church & Hovy (1993), whose core idea is to achieve an organic unity of the accuracy of human translation and the efficiency of machine translation through interaction between translators and machine translation engines. Currently, there are not many interactive machine translation systems available, mainly including CASMACAT and Lilt developed abroad, as well as Tencent’s self-developed TranSmart platform and the LanguageX translation platform developed by Oracle. The main advantage of Interactive Machine Translation lies in interactivity, with the core being online adaptive technology. During the translation process, the system automatically predicts the upcoming content based on the parts already translated by the translator and dynamically generates subsequent translations for reference. Translators can accept the translations provided by the system or modify them according to their thoughts. The system takes each input as feedback to “learn” and adjusts in real-time, and this interactive process continues until the translation task is completed. This not only improves translation quality but also avoids the frustration of translators repeatedly correcting the same errors during the post-editing process. Although interactive machine translation systems are still in the prototype stage, with advancements in machine translation technology, especially the development of adaptive technology based on deep learning, “interactive machine translation is expected to become one of the options for human translation” (Huang Guoping, 2017:21).

In summary, the technology-driven translation model has evolved from computer-aided translation to machine translation post-editing, and then to interactive machine translation. Each translation model has its own characteristics and has developed and innovated based on previous models, presenting some typical features in terms of automation level, translator subjectivity, and human-machine interaction capabilities, as shown in Table 1.

Firstly, the intelligence level of translation is increasing. As shown in Table 1, early computer-aided translation technologies had relatively low complexity, mainly relying on translation memory technology (i.e., bilingual parallel corpora). In contrast, under the post-editing model, the machine translation engine technology used by translators is much more complex, requiring strong artificial intelligence and deep learning technologies as support. Interactive machine translation introduces even more advanced adaptive technologies on the basis of the former, further enhancing the system’s intelligence and interactivity. It is evident that as translation models continue to evolve, the automation and intelligence levels of translation are increasingly high, and the contribution of technology to translation efficiency and quality is also growing.

Secondly, the subjectivity of translators has weakened. In the computer-aided translation model, although various tools and software play an important role in the translation process, they serve merely as auxiliary functions, with most of the translation work still completed by translators, who are the primary agents of the translation activity. However, in machine translation post-editing and interactive machine translation, most work is completed by machine translation, and translators only modify and process the output after the system produces the translation, resulting in a lack of full expression of initiative and creativity. Although interactive machine translation consciously increases human participation in the post-editing process, the subjectivity of translators still cannot be fully manifested.

Finally, the degree of human-machine interaction is continuously enhanced. In both the computer-aided translation and post-editing models, the level of interaction between humans and machines is relatively weak. Translators can only refine translations based on translation memories or process and modify static translations produced by machine translation, without receiving timely feedback and adjustments. In the interactive machine translation model, however, translators can receive immediate feedback from the machine translation system and make dynamic adjustments, with modifications being stored in the translation memory for later translations. This human-machine interaction continues until the translation task is completed and is considered “the mainstream translation model for current and future professional translators” (Wang Huashu, 2020:85). In the interactive machine translation model, both the efficiency of machines and the initiative of translators can be enhanced, complementing and benefiting each other.

The evolution of technology-driven translation models is an inevitable trend in the development of the translation industry in the era of artificial intelligence. Exploring the underlying driving forces not only deepens the understanding and comprehension of translation models but also holds significant reference value for industry development, technological innovation, and translator training. The following sections analyze the driving forces from three aspects: language service demand, technological tool innovation, and socio-cultural factors.

4.1 Language Service Demand as the Fundamental Factor in the Evolution of Translation Models

Changes in language service demand are the fundamental reasons driving the evolution of translation models. On one hand, with the acceleration of global economic integration, the demand for language services has surged dramatically, placing unprecedented emphasis on the speed and efficiency of translation. According to a survey by the international authoritative research organization CSA Research, the global language service output value was only 14 billion USD in 2008, reaching 52.01 billion USD in 2022. The rapidly growing demand for translation poses challenges to the traditional translation industry, as it is difficult to complete such a massive volume of translation relying solely on manpower and simple organization. Introducing translation technology to enhance production efficiency can serve as an ideal solution to this problem. On the other hand, changes in translation objects have also driven the transformation of translation models. With the advent of the digital age, “the objects of translation have not only included traditional paper texts but have also emerged in various forms of symbols covering text, images, sounds, videos, etc., such as INDD, XML, YML, JSON files” (Xie Tianzhen, 2015: 14). Without the support of translation technology tools, translators cannot access the files that need to be translated, and thus cannot begin these text projects. This has largely compelled the translation industry to continuously upgrade technological tools and innovate translation models to meet the ever-changing industry and market demands. Furthermore, as language service demand evolves toward multilingual, large-scale, and high-quality directions, translation models have gradually shifted from individual translation to team collaborative translation, inevitably necessitating enhanced project management communication and coordination, division of labor, and sharing of project information. This has greatly promoted the emergence of various cloud translation platforms (such as Phrase, YiCAT, and Yima Network).

4.2 Technological Innovation as the Direct Driver of Translation Model Evolution

Innovations in translation technology provide a continuous driving force for the transformation of translation models. Fundamentally, translation technology is a product of the continuous development of artificial intelligence technology, based on software technology, internet technology, and big data technology. It evolves alongside advancements in artificial intelligence technology, continuously iterating and upgrading, providing direct impetus for the evolution of translation models.

Throughout history, every significant social transformation has closely correlated with scientific discoveries and technological inventions. Over the past 30 years, disruptive technologies that have changed translation production methods have emerged, significantly impacting the traditional manual translation model. In the 1990s, increasingly mature corpus technology spurred the emergence of computer-aided translation software and platforms based on translation memory, with the computer-aided translation model gradually becoming mainstream in the translation industry. The rapidly changing machine translation technology, especially the emergence of neural machine translation technology in 2016, broke the dominance of computer-aided translation, bringing machine translation post-editing to the forefront. In recent years, adaptive technologies based on deep learning have injected new vitality into the transformation of the translation industry, and the concept of human-machine interaction has gradually gained recognition and acceptance in the industry. The translation model of “human-machine collaborative translation” has also transitioned from an ideal to a reality. The release of large language models represented by ChatGPT in 2022 has accelerated the application of interactive machine translation models. It is evident that every transformation of translation models is inextricably linked to innovations and developments in translation technology.

4.3 Socio-Cultural Factors as Regulators in the Evolution of Translation Models

Socio-cultural factors also play an important regulatory role in the transformation of translation models. With the widespread application of technological tools in the translation process, translation efficiency has significantly improved, exhibiting enormous productivity. While marveling at technological advancements, people often overemphasize speed and efficiency, becoming overly reliant on technology applications while neglecting the humanistic and social aspects of translation activities. This “technocentric” mindset has driven the application of new models such as computer-aided translation and machine translation post-editing, but has also led to an over-reliance on machines and tools. However, translation is fundamentally a communicative activity that encompasses both humanistic and social dimensions, where humans play a dominant role. Only by fully utilizing human agency and creativity can various tools and technologies be reasonably employed and harnessed to achieve the goal of knowledge and value transmission, rather than being enslaved by tools or technology. With the return of value rationality, academia has begun to critically reflect on translation technology and its derived translation models, focusing on exploring the socio-cultural factors within translation activities, gradually aligning with a “human-centered” philosophy. Under the guidance of value rationality, an increasing number of researchers are attempting to reconstruct the subjectivity of translators, leading to a “trend of integrating instrumental rationality and value rationality” (Li Yan, Xiao Weiqing, 2018:1), with the interactive machine translation model gradually emerging and developing under this trend.

In the era of artificial intelligence, the transformation of technology-driven translation models not only profoundly affects the development of the language service industry but also presents opportunities and challenges for technological innovation, translation education, and training. Based on the aforementioned analysis, this article proposes that construction should be carried out in the following areas to promote the sustainable and healthy development of the translation industry.

5.1 Establish a Harmonious Human-Machine Symbiotic Ecosystem

In the face of the rapidly changing development of artificial intelligence technology, the language service industry should adapt to the trend of the times, lead the development of translation technology and production models, and establish a translation ecosystem where humans and machines coexist harmoniously. In this ecosystem, translators and machines do not compete or replace each other but rather support each other and progress systematically, ultimately achieving close coupling between humans and machines. To achieve this goal, it is essential to strengthen human-machine interaction and guide the human-machine relationship toward a symbiotic direction. Through human-machine interaction, machines can better adapt to human control, perception, and cognitive abilities, providing flexible and personalized services, while translators can fully leverage their initiative and creativity through intelligent translation technologies or machine engines, thereby enhancing translation efficiency and quality. Davenport & Kirby (2016) have foresightedly pointed out that the combination of technological thinking and human wisdom is an inevitable trend for future development, and whether humans and machines can coexist harmoniously depends on our choices.

5.2 Promote Human-Centered Technological Innovation

As artificial intelligence technology is widely applied in the translation field, some problems and shortcomings have gradually emerged, with the most significant issue being that many designs do not adequately consider the “human” factor. Currently, the adaptive technology of interactive machine translation effectively utilizes the information brought by human-machine interaction to enhance the translation model’s effectiveness, serving as a good leading and demonstrative example. However, the current human-machine interaction technology is still immature and needs improvement. For instance, interactive machine translation systems still adopt traditional left-right interaction schemes, which somewhat affects translation efficiency (Huang Guoping, 2017). The enhancement and progress of translation technology cannot rely solely on advanced computational models; it also requires fully tapping into the information obtained through human-machine interaction to enhance translation capabilities, “providing humanistic care and ethical guidance to its developers and users to promote the in-depth development and interdisciplinary advancement of machine translation” (Wang Yun, Zhang Zheng, 2022: 112). This necessitates industry users to provide support and feedback to developers of translation technology software and systems, guiding the development of translation technology rather than being led by technology. In the future, human-machine interaction and mutual learning will become critical to the advancement of machine translation technology and even artificial intelligence technology.

5.3 Emphasize Translation Data Security and Professional Ethics in Technology

As technology-driven translation production models continue to expand and extend, issues of data security and professional ethics in translation activities are becoming increasingly prominent. On one hand, the use of internet-based translation technologies and cloud platforms has significantly improved translation output efficiency, but it also poses data security risks. For example, terminology databases and translation memory stored on cloud servers may be vulnerable to hacking and virus attacks, leading to data breaches; using online machine translation also carries the risk of personal privacy leakage, as machine translation service providers may utilize users’ actual translation data for machine learning and training (Wang Huashu, Liu Shijie, 2022). Therefore, ensuring the security of translation data is an urgent task. On the other hand, translation technology is a double-edged sword; its misuse can have negative impacts on the language service industry. For instance, in machine translation post-editing, if the machine translation output is not rigorously and professionally reviewed after editing, it may lead to a decline in translation quality, harming the professional image of translators and potentially causing reading difficulties for readers. Thus, it is crucial to emphasize the training of translators in professional ethics and enhance the ethical constraints on translators’ technology. The “Code of Ethics and Conduct for Translators” issued by the China Translation Association in 2019 can play a good guiding role in this regard.

5.4 Focus on the Dual Enhancement of Translators’ Technical Skills and Humanistic Literacy

The transformation of technology-driven translation models poses significant challenges to translation education and translator training, as graduates trained under traditional teaching models find it difficult to adapt to the rapidly changing translation industry. To meet the current needs of translation technology and industry development, it is essential to reform the existing teaching system, adding courses on translation technology to enhance students’ technical skills. However, translation is not merely a language conversion task; it is a cross-cultural communicative activity that encompasses social, cultural, symbolic, creative, and historical dimensions (Xu Jun, 2003). This means that a qualified translator needs not only solid bilingual skills but also extensive cultural knowledge, rich emotions, and imagination. If the focus is solely on developing technical skills while neglecting humanistic spirit cultivation, translators may “face the risk of being ‘mechanized,’ gradually becoming workers on the language assembly line” (Wang Huashu, Liu Shijie, 2021:89). In future translation education, it is essential to cultivate students’ technical abilities while also significantly enhancing their humanistic literacy, focusing on developing aesthetic, investigative, and logical thinking skills, and stimulating their initiative and creativity.

06

In the era of artificial intelligence, translation technology is entering a phase of rapid development. The transformation of technology-driven translation industries and models aligns with the national strategy of developing language intelligence and is also an inevitable trend of historical development. This research on the evolution of technology-driven translation models reveals an increasingly evident trend toward intelligence and automation, alongside a growing reliance of translators on technology, leading to a weakening and decline of subjectivity and creativity. However, this does not imply that the two are entirely opposed and irreconcilable; rather, they are interconnected and mutually dependent contradictions. In the future, the key to achieving harmonious development between the two lies in promoting the development and application of human-machine interaction technology: on one hand, the development of translation technology should focus on being human-centered, enhancing human-machine interaction, and stimulating the initiative and creativity of translators; on the other hand, translators should maintain an objective and rational understanding, actively accept and guide the development of translation technology, and work towards realizing the ideal of mutual dependence and symbiosis between humans and machines.