Research on Post-Editing Skills in the Age of AI

Introduction

In the era of artificial intelligence, machine translation technology is continuously updated, with large language models represented by ChatGPT disrupting traditional translation methods. While improving translation efficiency, it also raises ongoing questions regarding the roles and relationships between humans and machines. Currently, the “machine translation + post-editing” model has become a widely popular work and service model, and the importance of post-editing skills is increasingly recognized by both academia and industry. Recent books and articles discuss the definitions, model explorations, and skill development methods related to post-editing capabilities. This issue focuses on a book and two papers that concentrate on the study of post-editing skills in the AI era. Wang Huashu and Liu Shijie (2022) provide insights on building and developing a post-editing capability model based on real industry needs. Zhao Qiurong (2023) offers specific suggestions for cultivating post-editing skills based on research and model exploration. Wang Lu and Wang Xiangling (2023) empirically discover that the interactive post-editing capability cultivation model of “ChatGPT + MTPE” helps students form a more flexible and comprehensive knowledge system of post-editing strategies, enhances translation quality assessment abilities, and improves professional knowledge levels. The study of post-editing skills in the AI era still holds significant exploration potential for the future, and we hope this issue can inspire translation education, research, and industry practices, continuously seeking methods and pathways to enhance human-machine collaborative translation efficiency and quality.

Wang Huashu, PhD in Translation Studies, Associate Professor, Master’s Supervisor at the School of Advanced Translation, Beijing Foreign Studies University. His main research areas include translation and localization technology, foreign language education technology, and language service management.

Liu Shijie, PhD candidate in the School of Foreign Languages, Shanghai Maritime University. His main research areas include maritime terminology extraction and text mining, and translation technology.

Exploration of Post-Editing Skills in Machine Translation Era

Introduction

Currently, translation services based on translation technology scenarios have become a new service model in the language service industry, transforming towards collaboration, crowdsourcing, platformization, and integration. Language service companies have raised higher demands for translators’ post-editing skills, requiring not only traditional translation abilities but also proficiency in using machine translation systems, post-editing platforms, efficiently identifying and correcting machine translation errors, and possessing certain risk assessment capabilities, data ethics awareness, and service consciousness. Against this backdrop, what skills post-editors should possess and how to cultivate these skills have become urgent topics for discussion.

1 Foundation of Post-Editing Skills in Machine Translation

Post-editing skills in machine translation refer to “the knowledge system and cognitive literacy required to process and modify the original output of machine translation according to certain purposes and requirements”. Through systematic review, we find that no widely accepted post-editing capability model has been summarized in academia and industry, and there are many details that require further refinement.

2 Construction of Post-Editing Skills Model

This paper analyzes previous research results on post-editing skills, translation skills, and proofreading skills, as well as the descriptions of post-editors’ skills and qualifications in ISO 18587. We selected ten mainstream job recruitment websites to collect job postings or project recruitment notices that require post-editing skills from translators, limited to information published between January 1, 2020, and July 31, 2022, resulting in 150 job postings after manual screening. Content analysis research methods were used for processing and analyzing these postings. Based on the analysis and synthesis results, the new post-editing skills model roughly covers seven categories of skills (see Figure 1).

2.1 Service Ability

Service ability refers to the capacity of post-editors to provide services to others throughout the entire process from finding post-editing tasks to delivering tasks, playing a role in coordinating the overall model and connecting various skills. Post-editors should fully demonstrate their professional qualities and client awareness during communication with clients, be familiar with the translation market, industry demands, and pricing trends, strictly adhere to ethical standards and guidelines, and ensure that post-editing proceeds in an orderly manner.

2.2 Language Ability

Language ability refers to the knowledge and skill level that post-editors possess in editing, revising, and quality assessing during bilingual conversion, enabling smooth cross-cultural communication activities between source and target languages.

2.3 Tool Ability

Tool ability refers to the technical literacy of post-editors in utilizing tools/technology resources to solve translation challenges, ensuring high-quality completion of post-editing tasks, covering pre-editing skills and terminology management, machine translation/post-editing tool operation, complex document parsing, corpus application, information retrieval and discernment skills, understanding of machine translation principles, and content management tool application abilities. Additionally, post-editors should possess basic programming skills.

2.4 Strategy Ability

Strategy ability refers to the sum of decision-making, forecasting, and assessment abilities exercised by post-editors to comprehensively enhance editing efficiency and translation quality, covering tool assessment abilities, machine translation quality assessment capabilities, the ability to identify, classify, and correct neural machine translation errors, the ability to accurately interpret and execute post-editing guidelines, and the ability to make accurate decisions on the degree of post-editing based on specific text types, subject areas, and client needs.

2.5 Subject Area Ability

Subject area ability refers to the capacity to understand and master professional content generated in the source language and present it in the target language.

2.6 Psychophysiological Ability

Psychophysiological ability refers to the comprehensive ability exhibited by post-editors at the psychological (cognitive, attitude) and physiological levels.

2.7 Management Ability

Management ability refers to the capacity of post-editors to effectively coordinate relationships with clients, project managers, and other stakeholders in collaborative post-editing tasks to achieve high-quality task completion.

3 Development Path of Post-Editing Skills

The cultivation of post-editing skills has become an important topic in training translation professionals, requiring all parties to work together to cultivate truly needed professionals in the language service industry. The following discusses four aspects of the development path for professional post-editors’ skills.

3.1 Innovating Traditional Translation Cognition

Translation education authorities and training institutions should innovate their understanding of traditional translation, focus on the new requirements for translation professionals arising from industry development needs and national language strategies; actively expand the integration of intelligent technology into translation teaching practices, guiding students to correctly use translation technology tools.

3.2 Emphasizing Post-Editing Curriculum Development

Higher education institutions should clarify the gap between students’ personal qualities and actual industry needs, collaborate with enterprises to establish an integrated post-editing talent training and research mechanism, and jointly create high-quality post-editing courses.

3.3 Strengthening Digital Literacy of Teachers and Students

As the main body of translation teaching and practice, translation teachers and students should adapt to the digital living environment and be familiar with the application of intelligent technology in translation practice.

3.4 Focusing on the Construction of Technical Ethics Norms

At the national level, efforts should be made to accelerate the construction of a technical ethics system, providing macro policy guidance; industry associations should assist in regulating the technical ethics norms of diverse behavioral entities, monitoring and assessing data and technical risks; enterprises should establish internal ethical rules to enhance employees’ awareness of technical ethics; and universities should actively guide teachers and students to correctly understand the pros and cons brought by machine translation, pay attention to issues of responsibility attribution and intellectual property protection, and enhance risk assessment and response capabilities.

4 Conclusion

In the era of artificial intelligence, the machine translation post-editing model has become a new topic of interest in academia and industry, and the language service industry has raised higher requirements for translators’ post-editing skills. This paper attempts to construct a new post-editing skills model, clearly defining the connotations of various skills and elaborating on the development path of post-editing skills from the perspectives of innovating translation cognition, emphasizing curriculum development, strengthening digital literacy, and focusing on ethical construction. This model has received some support and validation based on industry talent needs, and the current research results can provide important references for the training of domestic post-editing talents and the formulation of post-editing capability standards.

This article was published in Chinese Science and Technology Translation, Volume 35, Issue 4, pages 21-24. Thanks to the authors and the editorial department for granting permission for reprint. Due to space limitations, the content has been abridged, and notes and references have been omitted.

Zhao Qiurong, Professor and PhD supervisor at Beijing University of Science and Technology. His main research areas include Translation Studies and Corpus Linguistics.

Cultivation of Post-Editing Skills

In today’s information age, pure human translation can no longer meet the demands of the language service market, and possessing post-editing skills has become one of the core competencies of translators. The characteristics of translation activities determine that the translation process is a continuous decision-making process for the translator. Although translation skills and post-editing skills overlap and intersect, scholars (such as Nitzke & Hansen-Schirra 2021; O’Brien 2002) point out that there are still many differences between translation skills and post-editing skills, with post-editing skills emphasizing the editing of translations generated by machine translation, i.e., identifying machine translation errors and making appropriate modifications, which is also a continuous decision-making process for the translator. Many language learners have a certain resistance to technology; Pym (2013) emphasizes that translation teaching should guide learners to understand when to use technology. In the process of acquiring tools, translation students should ideally form study groups to promote and improve each other. O’Brien (2002) emphasizes that machine translation and post-editing should be part of translation teaching, including the history of machine translation, basic programming, terminology knowledge, and controlled language knowledge. Given the importance of post-editing skills, Kenny & Doherty (2014) suggest incorporating post-editing into translation teaching and translator training courses. Koponen (2015: 5-6) believes that post-editing courses should include both theoretical and practical introductions to machine translation and post-editing, as well as practical exercises based on machine translation and post-editing, such as post-editing controlled language and machine translation without the original text, as well as studying post-editing processes, quality standards, assessments, and skill cultivation. Arenas & Moorkens (2019: 223-224) outline eight components included in post-editing modules, specifically basic definitions, quality, types, rules, common errors, costs and outputs, pricing, and practical exercises.

Domestic scholars also attach great importance to the cultivation of post-editing skills, proposing to strengthen post-editing training (Cui Qiliang 2014; Feng Quanguang, Cui Qiliang 2016), and to construct a three-dimensional model of post-editing skills (Feng Quanguang, Liu Ming 2018). Drawing on foreign post-editing courses, Feng Quanguang and Zhang Huiyu (2015) suggest that post-editing course settings should include 12 modules, namely machine translation, pre-editing, an overview of post-editing, current industry applications, common machine translation error analysis, introduction and application of post-editing tools, overview of text (discourse) knowledge, post-editing practice I, post-editing practice II, machine translation and computer-assisted translation, basic programming introduction, and others (including post-editing pricing strategies). Zhong Wenming and Shu Chao (2020) propose course suggestions for teaching, involving three modules: pre-editing text processing ability, post-editing text processing ability, and basic general ability. The core of post-editing courses is to help students become familiar with machine translation technology, including recognizing common machine translation errors, understanding the potential and limitations of machine translation, and maintaining a positive attitude towards machine translation. (Konttinen et al. 2021: 199; O’Brien 2002).

Post-editing courses in translation teaching and translator training are still in their infancy. When or to what extent post-editing should be introduced into translation teaching remains a topic of debate (Plaza-Lara 2020). However, many scholars believe that post-editing skills should be included in the training module for translation skills.

We believe that the cultivation of post-editing skills still needs to focus on the following points.

First, embrace technology but avoid technological supremacy. Translators are at the center of translation or post-editing, and technology merely serves to assist translators in improving speed and quality. Teachers and trainers need to guide students not to blindly trust translation technology tools (Bowker 2005) but to form a correct understanding of translation technology through their own learning, becoming excellent talents that integrate language, technology, and expertise. Cui Qiliang (2014) points out that learners need to maintain a positive attitude, understand the thinking and methods behind machine translation outputs, understand the basic principles and common error types of machine translation, familiarize themselves with the project domain, terminology, and translation environment, quickly read and identify major errors produced by machine translation, and modify translations that are inconsistent with the original author’s intent. These are all skills needed to achieve high-quality post-editing.

Second, avoid excessive modification and insufficient modification. Post-editing should not be influenced by the translator’s personal style and habits, only necessary errors should be corrected. If the existing translation is grammatically correct and fluent, post-editing is unnecessary (Aikawa et al. 2012: 7).

The overall goal of post-editing course teaching is to enable translation students to master the basic principles of machine translation and post-editing, cultivating their ability to carry out post-editing practices using technology tools. The post-editing skills course is not meant to teach the operation of related software, but to teach students to be output-oriented, effectively combining machine translation with specific practices to complete post-editing projects as efficiently as possible. The ongoing discussion about whether machine translation will replace human translation still exists, but until fully automated high-quality machine translation emerges, machine translation will not replace human translation. Therefore, the current focus should be on how to make full use of machine translation. Translation capabilities are continuously changing, and the development of artificial intelligence technology and machine translation has made post-editing skills an essential skill for translators. The language service industry has begun to pay attention to post-editing training services. Due to limitations in teaching resources, there are relatively few universities offering post-editing courses, and teaching practices based on post-editing still need further expansion.

This article is excerpted from Research on Translation Competence (Foreign Language Teaching and Research Press, 2023, click here for book details) Chapter 5. Due to space limitations, notes and references have been omitted.

Wang Lu, PhD in English Language and Literature, Lecturer at the School of Foreign Languages, Hunan University. His main research areas include post-editing research in machine translation, cognitive processes in human-computer interactive translation, and AI-assisted translation.

Wang Xiangling, Professor and PhD supervisor at the School of Foreign Languages, Hunan University, Director of the Research Institute of Translation and Interpretation Cognition. Her main research areas include translation cognition, human-computer interaction in machine translation, and theories and empirical studies of translation education.

Research on Post-Editing Skills Cultivation Model in the ChatGPT Era

1 Introduction

In the era of artificial intelligence, translation technologies are emerging and iterating rapidly, leading to an increased degree of technological integration in the translation process, making post-editing skills an important component of translators’ professional qualities and an indispensable part of translation talent cultivation. How to integrate AI technologies like ChatGPT into post-editing teaching practices in higher education to promote the development of students’ post-editing skills remains a topic for ongoing exploration. This research aims to construct an interactive post-editing skills cultivation model based on social constructivism and project-based teaching methods, comparing the mastery of post-editing knowledge and translation quality between the experimental and control groups through teaching experiments to explore the effectiveness of this model.

2 Literature Review

2.1 Post-Editing and Post-Editing Teaching in the ChatGPT Era

Scholars at home and abroad have analyzed the components of post-editing skills from the perspective of multi-element models of translation competence or translation revision competence, pointing out that post-editing skills consist of bilingual ability, language proficiency, search ability, problem identification/classification/correction ability, and post-editing soft skills (Rico & Torrejon, 2012; Robert et al., 2022). Feng Quanguang and Liu Ming (2018) explain the connotation of post-editing skills from the perspective of the skill acquisition process, covering knowledge, skills, and cognition across three dimensions.

The existing results of post-editing teaching research are mostly summaries of teaching experiences (Zhong Wenming, Shu Chao, 2020), designs of teaching modules (O’Brien, 2002; Feng Quanguang, Zhang Huiyu, 2015), construction of teaching models (Guerberof & Moorkens, 2019), and theoretical and empirical studies on post-editing skills cultivation models incorporating ChatGPT are relatively scarce, with almost no research focusing on the knowledge acquisition process of student translators in post-editing teaching.

2.2 Project-Based Teaching Method Based on Constructivism

Since the early 20th century, scholars at home and abroad have begun to shift towards the paradigm of social constructivism, conducting exploratory research on translation teaching models from both theoretical and empirical perspectives to break the limitations of traditional translation teaching models (Kiraly, 2016; Zhang Zheng, Wang Yun, 2020). Kiraly (2000) shifts from cognitive psychology research in translation teaching to social constructivism, suggesting the application of real translation projects in translation teaching practice. Mahboubeh & Masood (2019) use group learning approaches in teaching translation technology. Additionally, multiple empirical studies indicate that project-based translation teaching helps cultivate students’ tool competence, information competence, and professional qualities (Zhang Wenhe, Wen Jun, 2020; Li, 2022).

3 Construction of Post-Editing Skills Cultivation Model in the ChatGPT Era

This research attempts to introduce ChatGPT into the post-editing classroom, theoretically constructing the “ChatGPT + MTPE” interactive post-editing skills cultivation model (see Figure 1). As shown in Figure 1, the “ChatGPT + MTPE” interactive post-editing skills cultivation model can be divided into three stages: construction of post-editing knowledge system, “ChatGPT + post-editing” working mode, and “ChatGPT + post-editing” project practice.

4 Teaching Experiment

To verify the feasibility and effectiveness of the “ChatGPT + MTPE” interactive post-editing skills cultivation model, this research conducted preliminary exploratory studies in a post-editing class of the MTI translation program at a comprehensive key university in central China.

4.1 Research Subjects

This paper randomly sampled two translation master’s degree (MTI) classes from local universities, dividing students into experimental and control groups (20 students each). Before the experiment began, the declarative knowledge of post-editing was measured for both groups based on a post-editing knowledge questionnaire. Independent sample t tests showed no significant difference in post-editing knowledge between the two groups (p = .834 > .05), indicating that the differences in post-editing capabilities between the experimental and control groups were within acceptable ranges.

4.2 Research Design

To minimize the interference of unrelated variables such as teachers, teaching materials, and class hours, the teaching tasks for both the experimental and control groups were completed by the same teacher, and the teaching content, progress, and class hours (16 weeks, 2 hours per week) were kept consistent. The only difference was that the experimental group was taught according to the new model, while the control group was taught according to traditional teaching methods. During the teaching process, both groups created dedicated QQ groups for the course to share learning and research materials and communicate. The difference was that the course group of the experimental group included subject experts, professional translators, and target language readers. After the teaching experiment, ten students from each group were selected for online interviews, and all students were required to submit translations and online interaction records, completing the post-editing knowledge questionnaire to form individual learning portfolios.

4.3 Research Methods

This research designed a post-editing knowledge measurement questionnaire based on Albir’s (2017) translation knowledge questionnaire, categorizing post-editing knowledge into five key elements — translation quality assessment, post-editing issues, machine translation errors, post-editing strategies, and professional knowledge.

The translation quality assessment questionnaire was based on the evaluation methods of Melby et al. (2014) and Galán-Mañas & Albir (2015), setting three variables: accuracy, fluency, and textual functionality.

Semi-structured interviews were conducted via Tencent Meeting, with each student spending about ten minutes. Semi-structured questions included learning experiences and gains, assessments of ChatGPT’s post-editing performance, evaluations of the teaching model, self-assessments of translations, and suggestions for the teaching model.

5 Results and Discussion

5.1 Comparison of Post-Editing Knowledge

This paper conducted independent sample t tests on the five elements of post-editing knowledge between the experimental and control groups using SPSS 26.0 to explore the differences in post-editing knowledge mastery between students in project-based interactive classrooms and traditional post-editing classrooms, ultimately verifying the teaching effectiveness of the “ChatGPT + MTPE” interactive post-editing skills cultivation model. The results show that the experimental group scored significantly higher than the control group in terms of post-editing strategies, translation quality assessment, and professional knowledge levels, indicating that the “ChatGPT + MTPE” interactive post-editing skills cultivation model helps students form a more flexible knowledge system of post-editing strategies, promotes the improvement of translation quality assessment abilities, and enhances professional knowledge levels. Detailed results can be found in Table 1.

5.2 Comparison of Post-Editing Quality

This study conducted independent sample t tests on the quality of post-editing outputs from the experimental and control groups. The results show that the post-editing quality of the experimental group scored significantly higher than the control group in terms of fluency and textual functionality, indicating that the project-driven post-editing skills cultivation model empowered by ChatGPT is conducive to producing more fluent and communicatively functional translations. Details are provided in Table 2.

6 Conclusion

This article adopts a quasi-experimental research design based on experimental and control groups to explore the teaching effectiveness of the “ChatGPT + MTPE” interactive post-editing skills cultivation model. The research results show that the experimental group scored significantly higher than the control group in post-editing strategies, translation quality assessment, and professional knowledge, indicating that this model helps students form a more flexible and comprehensive knowledge system of post-editing strategies, promotes the improvement of translation quality assessment abilities, and enhances professional knowledge levels. Additionally, students from both groups expressed a desire to learn more about translation quality assessment, reflecting the importance of quality assessment abilities in post-editing teaching. Furthermore, the post-editing outputs from the experimental group scored significantly higher than the control group in terms of “fluency” and “textual functionality,” indicating that the project-driven post-editing skills cultivation model has a certain promoting effect on improving translation quality.

This article was published in Foreign Language Electronic Teaching, Issue 4, 2023, pages 16-23. Thanks to the authors and the editorial department for granting permission for reprint. Due to space limitations, the content has been abridged, and notes and references have been omitted.

Related Reading

New Book Release | Research on Translation Competence (Frontier Research Library on Core Topics in Foreign Language Disciplines)

Focus on Hot Topics | Research on Self-Revision in Translation

What Does Translation Competence Research Do, and How to Do It?

How to Assess Translation Competence?

Leave a Comment Cancel reply