Machines are becoming more accurate at translations, but they still aren't on par with trained human linguists. However, using human linguists for all aspects of a translation project can be costly and time-consuming, particularly those with high-volume content types with low intrinsic value. To achieve efficient, cost-effective and accurate translations for such content types, you can combine machine translation with human editing (MTPE).
Before Editing Begins
Much of the work occurs before the actual translation (or editing) begins. The better the machine output is, the less post-editing that will be required.
Carefully Created Source Text
Start with a well-written source text that is free of spelling and grammatical errors and uses the standard word order. If possible, reuse content that has already been translated rather than creating from scratch.
Write short sentences and use active voice when possible. Avoid idioms and humor; machines cannot pick up nuances as humans do. Keep terminology and formatting consistent and avoid numeric formats for dates. For example, use January 5, 2022, rather than 1/05/2022 or 5/01/2022.
The Right Translation Engine
Consider the best translation engine for your text. Some engines are better for certain language pairs or specific industries. Choosing the right one can improve the quality of your raw output. Use your experience to determine the right engine, or try a sample of your text in various generic engines to see which works best.
Another option is to train an engine yourself, using your own data. A third option is to use a translation management system that automatically chooses the best translation engine for you.
Post-editing
Once the machine has produced its translation, determine how much post-editing you want. Weigh quality against time and cost constraints.
Strategies
One strategy is to do a light post-editing. In this strategy, the editor looks only for major errors that render the text illegible, wrong, or offensive. For example, one translation engine erroneously translated the name of the Chinese leader to a vulgar word. A light post-editing would catch this error.
A full post-editing reviews the machine translation for any errors, major or minor. Errors might be grammatical, tonal, or cultural. After a full post-editing, a native speaker of the target language should believe the document was originally written in that target language. Obviously, this strategy is slower and more costly than light post-editing, but it results in more accurate output.
In some cases, your experience tells you that a machine translation will provide good output. If 100 percent accuracy is less important than having a quick translation, you might opt to forego any post-editing in this case.
Of course, you can choose an editing strategy for each project. Those that require accuracy might have fuller editing than those where a general translation is acceptable.
Tools
Almost all CAT platforms offer some tools to support post-editing. These include:
- Terminology management systems that aid in consistency, such as reference manuals, translation memory, and terminology databases
- Quality assurance tools that can identify overlooked issues or new errors introduced in the process
- Machine Translation Quality Estimation (MTQE) which provides scores for each machine translation and helps you know what to prioritize.
Post-editing Training
Although post-editors possess skills similar to translators, they aren't really translators. Many undergo formal training. A good post-editor has some of these skills and attributes:
- Cultural knowledge
- Linguistic and textual competence in both the source language and the target language
- Translation competence
- Research and information processing skills
- Technical competence in MT engine models (such as neural, statistical, and rule-based), including the most common mistakes of each model
- Knowledge of CAT tools
- Knowledge of domain content
- Speed and accuracy
- Flexibility and willingness to accept and adapt to machine translation technology
Evaluation
To continue to improve, you must evaluate the results of your post-editing machine translation or PEMT so that you can improve with the next iteration. Several tools exist to help with this evaluation. For example, the translation management system Memsource provides information on how much post-editing work each passage required. If a passage required a disproportionate amount of time, you might consider adjusting the source text, for example, or using a different machine translation engine model.
Besides using tools, you might also want to survey the humans involved in the process, including content editors, post-editors, project managers, and clients, to receive their feedback on what worked well and what could improve.
Andovar can provide turnkey custom solutions for your translation needs, including machine translation, post-editing, and human translation, where appropriate. We localize documents, websites, apps, gaming, software, voiceovers, e-learning, subtitling, e-commerce automation, and media.
We have six global offices so that we can offer support around the clock. We translate more than 200 language pairs and can help you grow your business globally. Talk to an expert today.
For more tips and content on global growth, please visit our blog.
