Machine Translation Post-editing (MTPE) is becoming a standard approach for a broad scope of industries in handling their translation needs. This phenomenon is shifting from the traditional translation method because of its practicality which offers better speed and lower cost in handling a very large translation batch. This translation procedure is even better and many have proved the reliability in achieving publishable quality that is almost equal to the conventional translation services which frequently cost higher.
Although MTPE is a standard practice in the global industry, there are still few studies that support its application so fewer methods are available compared to the conventional translation method. The lack of cohesive standards has made it difficult to provide uniform translation results which both accurate and publishable since different editors would give different measures and analyses.
The common over-edit or under-edit mistakes by editors needs to be minimized by providing a clear and applicable strategy of MTPE. We have tried to summarize the most practical and powerful technique that is also applied as our standard practice in handling MTPE projects.
Machine Translation Post Editing Guideline
The current global industries require a real-time translation process that has the capability to function on different platforms or media. Machine translation post-editing (MTPE) is emerging as a new solution in the translation industry regardless of several issues of machine translation quality of several language pairs.
The role of editors is even more important than the translators in MTPE, but their translation experiences will help editors perform better in analyzing the text. Therefore, in qualifying the editors for MTPE, it is better to use in-house translators or freelance translators with good records from the translators’ pool. Initial qualification and simple training will be necessary to make them ready to handle the jobs.
The role of editors in MTPE could be different according to the type of text and the quality of the machine translation. All in all, the practice of MTPE is categorized into two standard practices which are a full-editing practice and a regular editing practice.
The full-editing practice is only for machine translation results of certain language pairs that have notable errors and poor quality. This is also often required by clients who want a high-quality translation. The scope of editing is described as follows.
- Text format and punctuation. The writing format of each language can be different while several machine translations only translate text. This case often occurs when translating two language pairs of different writing systems or alphabets such as English to Japan and vice versa. The complicated writing systems with lots of loan words make Japanese tricky for machines. The editor should be aware of these issues and give some extra efforts which often requires retranslation instead of editing.
- Proper nouns. The proper nouns refer to the name of a person, place, or company that does not need translations. There are also often some ‘do not translate’ orders from the clients for several terms that they want to keep. The common machine translation cannot recognize this and give confusing or ambiguous results. However, the current translation environment that integrates machine translation into CAT helps the editor to put a list of ‘do not translate’ elements. It can also exclude tags and other types of exception words.
- Outliers. The length difference between the source and target text is a crucial issue in MTPE. Text expansion, contraction, omission, and addition are some instances that occur in the machine translation process, though those can also happen in human translation. When dealing with a project that will be used in websites, apps, or other printable texts, text expansion or contraction can create a problem since space matters a lot. Editing such things could take much longer time but it is really important so the results can meet acceptable and publishable quality.
- Morphology and Sentence structure. Typo and grammatical errors often happen in the MT process, especially in languages with time, gender, and subject reference. Machine translation cannot yet recognize different varieties of lexicons or lexical complexity of every language and translate them properly. This kind of error in using and constructing lexicons can be easily spotted in the word levels but sometimes need the editors’ extra awareness in the phrase or sentence levels.
- Terminology and exact match. In a translation project, terminology and some exact match elements should be translated consistently. The use of CAT tools can be very helpful to spot such inconsistencies, yet in the manual editing process, an editor may need to make a personal note with a list of terminologies and sentences that appear multiple times.
In the case of reliable machine translation results of two identical languages’ characters or structures, regular editing can be carried out. The parameters should only focus on the use of terminology, general meaning, text expansion, and morphological errors. The regular editing process is way faster and does not require highly skilled editors with lots of experience and knowledge in translations.
The editing procedure in MTPE seems more structural and has more predictable errors compared to editing the result of human translation. The human translation carried out by professionals may have fewer mistakes that are hard to spot, yet those done by regular translators in crowdsource platforms could suggest a more random error that happens repeatedly. Therefore, the practice of MTPE would be more beneficial and also practical that costing less time and money.
