Introduction to Post-Editing in Machine Translation

In recent years, although machine translation technology has made significant progress, overall, the quality of machine translation still cannot be compared with human translation. In this context, the model combining machine translation with post-editing has rapidly developed, leveraging the speed (efficiency) of machine translation while also harnessing the accuracy (quality) of human translation. This achieves a balance between translation quality and efficiency, utilizing the advantages of human-machine interaction.Post-editing (PE) is defined as “the process of reviewing and correcting the output of machine translation,” which combines machine translation with high-level professional human post-editors to produce high-quality translations that meet publication standards.

Introduction to Post-Editing

Post-editing can be distinguished between directly editing translations obtained through machine translation and editing translations obtained in an integrated translation environment consisting of translation memory, machine translation, and translation management systems. Using an integrated translation environment ensures translation quality and efficiency, with professional editors conducting manual reviews or partial automated revisions of the initial output based on specific quality objectives.

1. Application of Post-Editing

With the development of economic globalization, almost every industry has translation needs. According to data provided by the Localization Industry Standards Association (LISA), the translation demand for (product) user documentation, information technology, and marketing materials accounts for the largest proportion, while sales support, legal regulations, medicine, and life sciences also occupy a significant share. Moreover, the subjects that require translation in today’s market are becoming increasingly diverse, and the purposes and clients (users) for translation exhibit variety, leading to different requirements for translation quality, cost, and efficiency. The combination of these factors not only determines whether post-editing is needed but also the extent of post-editing required.

At the same time, post-editing of machine translation has already been commercially applied in the language service industry (especially in software localization). For example, some products or projects localized by companies such as Google, Microsoft, Symantec, Autodesk, Lab126, SDL, and WPP have utilized machine translation post-editing, accumulating practical experience in post-editing. The ISO 18587 draft standard (ISO, 2014) for machine translation post-editing divides the post-editing process into preparation, post-editing, and post-processing stages. It describes the knowledge and skills required of post-editors. The outputs of post-editing are categorized into two levels:

Light Post-Editing and Full Post-Editing. Light post-editing aims to utilize the original output of machine translation as much as possible, focusing on correcting mistranslations, cultural discrepancies, and reorganizing sentence structures without altering the translation style; whereas the goal of full post-editing is grammatical and punctuation correctness, with the translation being accurate, easy to understand, and stylistically consistent, emphasizing grammar, punctuation, spelling accuracy, translation and terminology accuracy, modification of culturally inappropriate translations, and consistency with client requirements.

As a new growth point for language service companies, post-editing is an important manifestation of human-machine interaction in translation, and it can provide feedback for the improvement of machine translation, representing the future direction of translation services. Many language service companies have implemented translation processes that combine machine translation and post-editing, partly to meet client demands and partly to maintain translation profitability and efficiency.

The expectations, roles, and backgrounds of clients (users) regarding information acquisition determine the purpose of translation and the quality requirements for the translated text. If the goal is just to understand the basic content, and the user is a professional technician, then the output from machine translation can meet the requirements. However, if the user is a non-professional and the accuracy of the translation affects the user’s understanding and use, or even endangers health and safety, high-quality translations are required, and at this point, relying solely on machine translation cannot guarantee translation quality.

Combining the user’s reading expectations for translated texts with the quality requirements for different subjects is more beneficial for analyzing which fields are suitable for post-editing after machine translation. Depending on the subject matter, the requirements for translation quality vary, and the translation methods can include Machine Translation (MT), Machine Translation with Post-Editing (MTM+PE), and Human Translation (HT). Subjects suitable for post-editing include chat logs, web browsing, and information retrieval for basic information; literary works, poetry, advertisements, books, and contracts demand high translation quality and are suitable for human translation; professional materials such as product user manuals, patents, telecommunications, and automotive design, along with software and websites, have moderate quality requirements and high translation efficiency needs, making them suitable for post-editing of machine translation.

The translations produced using machine translation outputs are the work objects for post-editing. The Translation Automation User Society (TAUS) analyzed four application scenarios for the use of machine translation in professional translation projects: (1) Time-sensitive content. (2) Content where translation quality requirements are not high (e.g., product information displayed in online stores). (3) Initial drafts that require human post-editing. (4) As a means of identifying translation issues that require further revision by reviewers.

In summary, the specific translation method should be determined based on different translation subjects, quality requirements, delivery timelines, translation budgets, and the translators’ proficiency with translation technology. Post-editing is suitable for subjects with specific requirements for translation quality, efficiency, and cost.

2. Establishing Post-Editing Based on Translation Quality Requirements

Just as a good translation requires a translation style guide, effective post-editing also necessitates a post-editing style guide. The post-editing style guide defines the work objectives, common types of translation errors, general and specific rules for post-editing, etc. The goal of the post-editing style guide is to ensure translation quality while improving the standardization and efficiency of post-editing, achieving better return on investment.

The work objectives of post-editing are influenced by the type of material being edited, the quality of machine translation output, client quality requirements, deadlines, costs, and users’ reading purposes. Generally, the work objectives of post-editing can be divided into two types: first, high-quality post-editing, aimed at producing high-quality translations comparable to those of professional translators; second, light post-editing, aimed at producing understandable and readable translations without considering details such as translation style. To achieve high-quality post-editing, it is necessary to obtain semantically correct translations with no omissions or additions, modify offensive or culturally inappropriate content, and pay attention to issues related to grammar, syntax, terminology, punctuation, and formatting.