Picture this: You need to translate 10,000 words for a game in multiple languages in record time. This means that translation from scratch for such an amount is near to impossible, thus it needs to be machine-translated. However, the output comes across as “robotic”, disregarding more metaphoric expressions.
This is one of the main concerns with machine translation (MT). Despite how much MT has come along (think about NMT and GenAI for example), there is still room for improvement, especially when it comes to culturally dependent information, informal or metaphorical language. This is where post-editing (PE/MTPE) comes in, transforming those clunky translations into something that makes sense.
In general terms, MTPE involves working on MT outputs to enhance their accuracy and fluency. Although the newest developments aim to have automated PE, this is usually carried out by human translators. In this article, we’ll delve more into PE and give some pointers for recommendations for companies and post-editors.
Post-Editing Basics
From Past to Present
Early MT systems produced translations that were often laughable and labelled as “useless” to the language industry. In this way, during the early days of MT, post-editing was considered a mandatory final step in the process. Human editors needed to correct very flawed translations to make them up to the industry’s standards. As a result, post-editing became a tedious and heavy task often in charge of non-bilingual editors. Much has changed since then, as advancements in artificial intelligence and neural networks have significantly improved MT quality.
As MT technology advanced, PE became a more specialised process. Nowadays PE is often minimal or unnecessary due to the good quality of MT outputs. However, MT can still falter, especially with complex texts or idiomatic expressions. This is where post-editing shines, as it bridges the gap between raw machine output and polished translations.
Today’s post-editors have a wealth of tools at their disposal. Translation management systems (TMS) often include features specifically designed for post-editing. These tools can help manage terminology, track changes, and even provide quality estimates for MT outputs.
Source: Nunes Vieira, 2019.
Levels of PE
Post-editing isn’t a one-size-fits-all process. As with everything in translation and localization, it all depends on the translation purpose. Taking this into account, MTPE generally falls into two categories:
- Light Post-Editing: Involves making minimal changes to ensure the translation is understandable and meets basic quality standards. It’s quick and often used for content where precision isn’t critical, like internal documents or casual communication.
- Full Post-Editing: This is a more thorough approach, where the post-editor meticulously reviews the translation for accuracy, coherence, and style. It’s essential for high-stakes content, such as legal documents, marketing materials, or anything where nuance and usability matter.
To PE or Not PE?
One of the most pressing questions in the translation industry is whether to post-edit machine translations or translate from scratch. As mentioned above, this decision or the type of PE depends on the desired quality of the MT output and the specific project requirements. Consider the following:
- Time: Do you need good translation for multiple languages in a quick turnaround? If so, MTPE might be the right solution.
- Quality: What type of text are you looking to translate? In many cases, post-editing can save time and resources, but if the MT quality is poor, it might be more efficient to start from scratch.
- Content: Are you translating technical in-domain content or more literary/metaphorical content? From this, consider if the MT output is good enough for your purposes.
- Durability: Do you need to translate important documents that are to be constantly used or are they more of a one-off kind of use? If so, MT alone or a light PE can do a good enough job.
The Language Specialist Rise (Pointers for post-editors)
As MTPE is more and more used in translation pipelines, the role of linguists has shifted towards the post-editing end. That means that linguists need to deal with MT systems and develop a wide set of skills to accommodate for this role.
Being tech-savvy is a game-changer in the world of post-editing. Understanding how machine translation works can help post-editors identify potential pitfalls and errors in the output. Familiarity with the latest MT engines and tools can also enhance efficiency and effectiveness.
Essential Skills
As mentioned above, post-editing shares some similarities with traditional translation. Here are some essential skills for successful post-editors:
- Attention to detail: A keen eye for detail is crucial for spotting errors and inconsistencies. Think about it; the human brain “autocompletes” missing letters in words or sentences to catch the gist of a message. A good post editor must train themselves not to let these things slide.
- Language proficiency: Strong language skills in both the source and target languages are a must. In the early days of post-editing, editors only needed to be proficient in the target language. As a result, the effort was high and translations were still not good enough.
- Cultural awareness: Understanding cultural nuances can help post-editors make informed decisions about phrasing and tone. This is what makes the difference between an average translation and a good one.
Tech-Savy
For post-editors looking to maximize efficiency, here are a few quick wins:
- Use glossaries: Maintain a glossary of key terms to ensure consistency across projects. This saves time and reduces errors. Most CAT tools have integrated glossaries, find where they are and use them at your convenience.
- Set up shortcuts: Familiarize yourself with keyboard shortcuts in your post-editing tools to speed up the process.
- Make use of QA tools: Most PE interfaces have quality assessment features that allow you to check source and target segments to make sure you’re not missing anything.
Training opportunities
Changes in the industry are reflected also in the training institutions. In the European Master’s in Translation Competence Framework 2022, for instance, technology skills are one of the basic competencies for translators in training. Within those competencies, understanding MT and integrating it into translation workflows shines as an addition to the switch in the translator’s education landscape.
Most translation programs include introductory modules about CAT tools and project management software. As researchers’ and stakeholders’ advocacy for MTPE training within translation programs grows, training on translation technologies increases. New proposals for including MTPE courses in translation programs emerge to reflect the changes and advances in the translation industry.
However, not everybody is happy about the upcoming changes. There are also valid concerns regarding the future and status of post-editors as opposed to translators. The issues raised varied from compensation according to post-editors efforts, their status, to what extent must MT outputs be post-edited to be considered good quality. A discussion that deviates towards clear guidelines in MTPE as well.
Measuring Impact
How do you know if the post-editing efforts are paying off? Measuring the impact of post-editing can be done through various metrics, such as:
- Turnaround time.
- Translation Error rate (TER) and Post Editing Effort (PEE) scores.
- Client feedback.
In production, turnaround time is indicative of productivity; however, this must not be the only measure of success. There must be also an assessment of quality that can be achieved by using metrics such as TER and PEE or any other metric that suits your process.
Post-editing should be viewed as an iterative process. For both language service providers (LSPs) and language specialists, it is important to gather feedback to identify areas for improvement. This continuous learning loop will help refine the post-editing process and skills (for LS) over time.
What’s Next?
The future of post-editing is bright, with advancements in AI and machine learning set to revolutionize the field. As MT technology continues to evolve, post-editors will need to adapt to new tools and methodologies. Areas not usually using MT will most likely increase their use due to the improved accuracy and ease of integrating PE in the pipelines.
By refining machine-generated text, post-editors play a crucial role in delivering high-quality translations that resonate with audiences. Embracing post-editing practices in a well-thought-out and ethical way will be key to navigating the ever-evolving landscape of translation.
