With the rapid advancement of artificial intelligence (AI), both the quality and efficiency of machine translation (MT) have been significantly improved. In particular, the application of neural machine translation (NMT) has led to breakthroughs in both the accuracy of text conversion and processing speed (Cui, 2021: 26). The swift development of AI has opened up new possibilities for translation and continues to reshape translation practices and the industry ecosystem (Han and Chai, 2024: 3). The field of translation is undergoing profound transformations (Li and Li, 2025: 25).

Regarding translation, the technological revolution presents a dual effect. On the one hand, the application of machine translation has markedly enhanced the efficiency of translation practice, particularly in preliminary text processing and basic terminology conversion (Korol, 2022: 577). On the other hand, current machine translation systems still exhibit shortcomings in cultural adaptability and handling complex contexts, especially in scenarios involving philosophical, cultural metaphors, rhetorical devices, and domain-specific terminology. These limitations often result in semantic deviations and inappropriate expressions.

Consequently, the era of AI has presented new requirements for the cultivation of English majors, with a growing demand for interdisciplinary talents. Translation learners are now expected not only to consolidate their foundational linguistic knowledge but also to develop digital literacy, cross-disciplinary knowledge, and the ability to utilize intelligent tools effectively. Additionally, they must cultivate critical revision awareness and multimodal content processing skills (Deng et al., 2024: 95).

This study selected English-Chinese translations produced by undergraduate English majors at a Chinese university and those generated by the Chinese large language model (LLM) DeepSeek as the research subjects. Using a mixed-methods approach that combined quantitative analysis and qualitative research, it systematically evaluated the differences in text quality between human and machine translations. Moreover, it delved into the types and causes of errors in students’ translation practices. Against the backdrop of the deep integration of AI and translation learning, how students can efficiently leverage machine translation to enhance their practical competence and translation proficiency constitutes a key research focus of the paper.

In recent years, machine translation has attracted extensive and in-depth attention in both academic and applied domains worldwide. Within the field of translation studies, machine translation has achieved numerous groundbreaking results in theoretical research and practical application. Its scope of application has expanded from traditional general text translation to highly specialized areas such as medicine, law, finance, and technology, gradually transforming translation practices and enhancing translation efficiency. The study focused on the theoretical developments and academic trends in artificial intelligence translation technology by systematically reviewing research directions in both Chinese and English literature. The source databases included CSSCI and Peking University Core Journals from the CNKI database, as well as SSCI-indexed journals from the Web of Science platform. The literature retrieval was confined to the period from 2020 to 2025, using a combination of keywords such as “computer-aided translation”, “neural machine translation”, “machine translation”, along with their corresponding English terms. As of the data collection date on September 8, 2025, a total of 1,139 valid publications were identified after cross-database deduplication and type filtering, laying a solid data foundation for subsequent in-depth analysis.

Statistical analysis revealed that research on machine translation by scholars both domestically and internationally demonstrates a trend of cross-disciplinary integration spanning multiple fields. The literature primarily focuses on the following key dimensions: At the technical level of machine translation models, numerous scholars had conducted in-depth analyses of the working mechanisms and challenges of LLM. Studies include, for example, investigating the principles and mechanisms of language generation in large language models (Khurana et al., 2023); categorizing and summarizing error types in machine translation (Yuan, 2016; Li, 2022); examining the risks and governance of generative AI from perspectives such as ethics, data privacy, and controllability of translation quality (Ye, 2021; Gao and Ren, 2025; Liu, 2025); and comparing the linguistic features of human and machine translations to identify translation traces in neural machine translation (Yang, 2020; Shen and Huang, 2024).

At the application level, research exhibited characteristics of diversification and specialization. Examples include experimentally comparing different post-editing strategies and methods (Feng and Li, 2016; Yao, 2024); investigating the adaptability and optimization pathways of machine translation for specific text types based on their distinctive translational features (He & Li, 2019; Popel et al., 2020; Geng and Hu, 2023; Yu, 2024)); enhancing translation performance by innovating model architectures, optimizing parameters, and improving training algorithms to develop new machine translation models (Liu et al., 2020; Du et al., 2025; Li and Zhang, 2025); and designing novel models for specialized domains such as healthcare and sign language translation through the construction of domain-specific corpora and the integration of professional knowledge graphs (Armanious, 2020; Camgoz et al., 2020).

At the level of translation pedagogy, with the widespread adoption of machine translation technology in the translation industry, the reform of translation teaching models and the cultivation of translation talent had become key academic focuses. Literature both domestically and internationally primarily emphasized adapting to the evolving competency requirements of the translation industry in the new era. This included reforming traditional translation teaching paradigms, integrating machine translation technology into pedagogical systems, and developing students’ ability to perform human-machine collaborative translation (Lee, 2020; Hu and Tian, 2020; Sui et al., 2025). Moreover, recommendations were proposed across multiple dimensions—such as curriculum design, practical teaching, and faculty development—aimed at cultivating versatile translation professionals who possess not only solid language skills but also proficiency in machine translation technology and translation project management, thereby meeting the diverse demands of the translation industry in the digital and intelligent era (Wang, 2017; Anthony and Yu, 2024; Wang and Wang, 2024).

With the continuous improvement and innovation of machine translation models—evolving from early rule-based systems to statistical machine translation and now to the dominant neural machine translation—academic research in this field has deepened significantly. The perspectives of research have also diversified, expanding from a narrow focus on translation accuracy to multidimensional investigations encompassing the translation process, quality assessment frameworks, ethical considerations, and the integration of translation technology into pedagogy. These developments have provided a theoretical foundation and practical guidance for assessing machine translation quality and cultivating translation talent, substantially promoting the synergistic growth of machine translation in both academic and applied contexts.

However, against the backdrop of an accelerating digital and intelligent transformation, empirical research remains relatively scarce on how English majors adapt to and implement human-machine collaborative translation modes. Enabling English majors to swiftly adapt to and effectively utilize human-machine collaborative translation during their academic training is a crucial step in cultivating high-quality, interdisciplinary translation professionals who meet industry demands. The study aims to provide new empirical evidence for human-machine collaborative translation mode and pedagogical reform through a comparative evaluation of English-Chinese translation quality produced by humans and machines. It also seeks to offer theoretical and practical insights for nurturing versatile translation talents capable of fulfilling the requirements of the digital and intelligent era.

The study selected both machine-generated and human-produced translations as research objects. For the machine-generated component, the Chinese LLM DeepSeek was chosen. As a leading representative of the new generation of artificial intelligence, DeepSeek has rapidly gained prominence across various regions and industries (Liu and Wen, 2025). In terms of English-Chinese language conversion quality, it demonstrates significant advantages over other translation platforms commonly used in the domestic market.

For the human participants, the author employed a sampling method by selecting translations produced by English majors from the School of Foreign Languages at a Chinese comprehensive university as part of their final examination for the course “English-Chinese/Chinese-English Translation (I)”. This course is a compulsory component for third-year English majors, spanning one semester with a total of 32 instructional hours. A cohort of 80 students participated in the course.

The source texts used in this study were selected from the final examination of the aforementioned course. The two English passages of different genres were chosen by the course instructors from online sources. The first text was an excerpt from the essay “Women and Men”, comprising 282 words. The second text was drawn from the second chapter of the 2013 essay collection The International Imperative in Higher Education, titled “Globalization and Forces for Change in Higher Education”, containing 303 words. The professors, who have extensive experience in translation pedagogy, are well-acquainted with the overall proficiency level of third-year English majors. Therefore, the selected texts were deemed appropriately challenging for evaluating the participants’ E-C translation abilities.

The procedure was conducted as follows: Prior to the research, both source texts were input into the DeepSeek interface with the instruction, “Translate the following passage into Chinese”, to obtain the machine-generated translations. The human participants were required to complete the translation of both texts within a two-hour time frame. The use of electronic devices such as mobile phones was prohibited; only a paper-based dictionary was permitted, ensuring an accurate reflection of the participants’ authentic translation competence.

Upon collecting the examination responses, it was observed that six students had not completed the translation of the second passage. These incomplete submissions were subsequently excluded from the analysis. The final sample thus consisted of 80 valid translations for the first passage (denoted as “Human Participants I”) and 74 valid translations for the second passage (denoted as “Human Participants II”). After compiling the samples, errors in both human and machine translations were systematically annotated. Finally, a comparative analysis of translation errors across human and machine outputs was conducted to examine the respective strengths and weaknesses of human and machine translation quality.

Guided by the Multidimensional Quality Metrics (MQM) framework, this study established a structured classification system for translation errors, organized into four core dimensions: Accuracy (encompassing mistranslation, over-translation, and omission), Linguistic Conventions (unintelligible), Style (unidiomatic style), and Locale Conventions (number format, time format). This resulted in a total of seven sub-category indicators. A quantitative scoring system was implemented to evaluate the samples in a structured manner. Drawing on empirical insights from senior evaluators of the China Accreditation Test for Translators and Interpreters (CATTI), errors were categorized and penalized according to severity: critical errors (3 points per instance), major errors (2 points per instance), and minor errors (1 point per instance). Specifically, mistranslation, omission, and errors in number and time formats were classified as critical errors and assigned a deduction of 3 points. Unintelligible and unidiomatic style were categorized as major errors, incurring a 2-point deduction, while over-translation was treated as a minor error, resulting in a 1-point deduction.

In accordance with the MQM scorecard approach, the quality assessment results of the samples in this study were quantified based on error type and severity. Deduction values were assigned to both human and machine translations (with results rounded to one decimal place) to enhance the clarity and objectivity of the evaluation (Lommel et al., 2014:165).

As indicated by the data in Fig. 1, human participants produced fewer errors and received lower penalty scores than the machine system in the first text, suggesting that human translation outperforms machine translation in handling literary texts. In contrast, for the second text — an academic essay — the machine translation demonstrated superior performance, with fewer errors and a lower penalty score compared to the human translators.

With the advancement of artificial intelligence technology, the field has entered an era where “translation is inseparable from technology” (Ren, 2019:47). The “machine translation + human translation” model has become a mainstream workflow (Feng and Cui, 2016:72), and human-machine collaborative translation is gradually becoming the norm (Yang et al., 2025:39). Traditional human translation is time-consuming and costly, making it difficult to meet the demands of rapid large-scale information exchange in the contemporary world. While machine translation may already achieve basic comprehensibility, producing high-quality, polished translations remains a considerable challenge (Fan and Yang, 2024:29). Therefore, human intervention is essential for quality optimization, establishing an integrated quality assurance system encompassing “pre-translation preparation, in-process monitoring, and post-translation review”. Within this framework, the machine translation system rapidly produces a draft, which is then post-edited by human translators. This approach not only ensures translation quality but also significantly shortens the delivery cycle (Wang and Zhu, 2023:39). The objective of human-machine collaboration is to augment human intelligence; by combining the efficiency of machine translation with the creativity and judgment of human translators, it is possible to achieve simultaneous improvements in both translation quality and efficiency (Chen et al., 2021:36).

Against the backdrop of rapid AI development, English majors must not only deepen their theoretical linguistic knowledge but also acquire operational skills in intelligent tools, while simultaneously cultivating critical thinking and ethical judgment regarding technology. Accordingly, students should learn to appropriately utilize machine translation and implement practical training strategies based on human-machine collaboration. For instance, during translation exercises, they may adopt a workflow consisting of “self-translation — machine-assisted revision — comparative analysis”. More specifically, students first complete a translation task independently, then use machine translation to assist in revising their work. Following revision, a comparative analysis is conducted to identify aspects worthy of learning from the machine output, such as lexical and phrasal usage, handling of complex syntactic structures, and discourse organization. Students with higher proficiency may further engage in error diagnosis and critique of the machine-generated text, annotating deficiencies, proposing alternative translations, and providing translator’s notes to justify modifications. This training model not only helps students improve translation quality and efficiency but also fosters their ability to critically employ technological tools and develop professional competence in translation quality assessment. Furthermore, it supports the development of a personalized translation style within a technology-enhanced environment, ultimately promoting the integrated development of theoretical knowledge, practical skills, and technological application, and achieving a synergy between technological empowerment and humanistic literacy.

Currently, there is an urgent need for the transformation and upgrading of talent cultivation, making the reform of translation education imperative. The integration of artificial intelligence in education is driving a structural shift in translation teaching models, characterized fundamentally by a transition from unidirectional knowledge transmission to interactive, human-machine collaborative pedagogy. This technology-enabled restructuring has facilitated the adoption of new instructional approaches, such as crowdsourced translation project-based learning, intelligent interpreting training integrated with speech recognition and real-time feedback, and dynamic corpus mining pedagogy powered by deep learning. These emerging teaching paradigms leverage technological intervention to achieve both spatial and temporal expansion of traditional translation classrooms, as well as a significant enhancement of instructional effectiveness (Kiraly et al., 2019).

Technology-enabled reform in foreign language curricula urgently requires breakthrough research, and institutions and instructors must undertake systematic adjustments to existing program structures (Xu, 2023:12). At the institutional level, there is a need to re-examine the educational objectives and curriculum design of translation studies, and to construct a comprehensive teaching framework integrating “humanistic literacy + digital technology + interdisciplinary collaboration”. Blended teaching models incorporating AI can effectively enhance the engagement and practical effectiveness of English majors. Departments should reconfigure the interactive relationships among key elements of translation pedagogy: the role of teachers should evolve into that of intelligent instructional designers, students should develop into critical users of technology, and teaching resources should advance into knowledge systems reflecting cutting-edge developments.

At the level of English language instructors, from a micro perspective, teachers should focus on students’ vocabulary acquisition, textual deconstruction skills, Chinese language proficiency, and mastery of conventions in English and Chinese symbolic formats. From a macro perspective, instructors should be trained in Technological Pedagogical Content Knowledge (TPACK). In the classroom, they must transition from traditional knowledge transmitters to designers, organizers, and technical mentors of translation learning. This requires not only deepening expertise in linguistic theory and intercultural communication competence, but also mastering the operation of intelligent tools, while fostering critical thinking and ethical judgment regarding technology. First, within the curriculum, in addition to developing students’ bilingual abilities, teachers should incorporate modules aimed at cultivating AI technological literacy. This involves guiding students to thoroughly understand the strengths and limitations of machine translation, thereby alleviating potential anxiety toward technology. Second, in daily instruction, alongside improving students’ English proficiency, teachers should instruct students on the appropriate use of machine translation and implement collaborative human-machine practical training schemes. For example, when assigning translation tasks, instructors may require students to submit an initial human-translated draft, a revised draft assisted by machine translation, and a polished final draft edited manually, supplemented with error diagnosis exercises. This approach not only helps students progressively improve their translation quality and efficiency, but also guides them in the rational use of machine translation technology. Furthermore, it cultivates their ability to evaluate the accuracy and precision of machine-generated translations, thereby enhancing critical thinking. Finally, in the era of AI, where translation pedagogy research is flourishing, teachers should transform their teaching models by introducing representative methodologies such as case-driven teaching and flipped classrooms. These approaches help students develop their own translation style, highlight individual strengths, and achieve a balance between theory and practice, as well as between technological empowerment and humanistic cultivation. Such approaches can effectively address the core issue of misalignment between conventional translation teaching methods and contemporary technological developments (Li et al., 2024:43), enabling innovation in translation teaching and practice through AI empowerment. This transformation entails not only innovations in teaching methodologies but also a systematic rethinking of the paradigm for educating future translation professionals.

The study adopted a mixed-methods approach combining quantitative and qualitative analysis to evaluate the quality of English-Chinese translations produced by both human and machine translators. By comparing and analyzing the performance differences across multiple textual dimensions, this research provided an in-depth examination of the challenges encountered by English majors in translation learning, identifies underlying causes, and proposes potential strategies for improvement. The findings indicate that in the current era of widespread machine translation, translation learning must adapt to technological trends, and talent development models require iterative upgrading. Firstly, English majors need to enhance their bilingual competence and proficiency in applying translation technologies. They should master the use of machine translation tools to leverage their efficiency while compensating for technological limitations through their deep understanding of both English and Chinese. Secondly, instructors should focus on cultivating students’ critical thinking and leadership capabilities, guiding them to become a new generation of translation professionals who are “proficient in languages, familiar with technology, skilled in coordination, and capable of innovation”—versatile talents who meet the demands of the translation industry in the digital and intelligent age.

Future research could expand the scope of this study by incorporating a wider variety of text types and translation scenarios to further validate the generalizability of the conclusions. Additionally, efforts should be made to refine both the theoretical framework and practical implementation pathways for human-machine collaborative translation. These research directions will contribute to enhancing the competitiveness and multidisciplinary development of English majors, while also helping to reshape and advance the future of the translation industry.