Machine Translation Post-Editing (MTPE): Go or No-Go?

Flitto Localization looks into the rising translation methodology to see if it’s worth it.

Machine Translation Post-Editing, commonly known as MTPE, is a translation project methodology where AI machine translation intersects with human translators or editors. It is experiencing steady growth with the advancement of AI translation technology.

If you are considering adopting or utilizing the MTPE method for your localization projects, here are some key points to consider.

Pros and cons of MTPE

MTPE services offer advantages and disadvantages compared to fully human-translated services. Here are a few:

Pros of MTPE services

• Relatively cost-effective
• Scalability
• Possibility to yield decent results in certain domains

Cons of MTPE services

• Dependent on the variant of translation model used
• Dependent on the human editor’s editing capacity
• Not as “speedy” as some might expect

These strengths and weaknesses make MTPE services more suitable for some industries than others. If the material mostly contains terms that have an equivalent in another language, it can be ideal to utilize AI technologies, as long as it’s done right.

However, machine translation technology so far notoriously falls short when it comes to translating deeply contextual conversations.

For instance, applying MTPE to translating media content requires appropriate human editor resources to ensure high-quality, engaging results. Editing poor-quality machine-translated outputs can consume more time and resources.

What are some types of MTPE?

MTPE services can be found in various contexts, which we can categorize into two broad project types:

For business purposes

This is the domain that when comes to mind when we think of translation services. It includes common documents, apps and websites, books, etc. As we previously mentioned, certain industry domains definitely thrive more with MTPE than others.

Crucially, the translated outputs are intended for human usage.

For scaling (AI, research, etc.)

Artificial intelligence development and research require an extensive amount of a specific type of data called the parallel corpora. They refer to bilingual (or multilingual) datasets of one source text arranged side-by-side (hence parallel), so that the source text can correspond to its multilingual counterparts. These corpora are used to train translation engines and other natural language processing tools.

While the applicability of MTPE can depend on factors like the domain of the datasets, it is common to find cases where MTPE is utilized for AI dataset scaling purposes.

The translated outputs are often for utilization by AI systems.

What to keep in mind

If you are considering adopting the MTPE framework for your next translation project, there are several factors to ensure optimized results.

• First, consult whether the domain of your source file is suitable for machine translation.
• Next, ensure that the base AI model used by the language service provider can actually produce high-quality outputs. (The base AI model is more important than you may think; Some AI models are better than others in certain linguistic pairs. You can consult your LSP for more details.)
• Consider having your translation data (e.g. industry jargon, proper nouns, brand and product names) trained into the private AI engine of your LSP. For instance, Flitto Localization offers machine translation API options where partnered companies can train their own data into their version of our base model, instead of using publicly open AI models. This is especially useful in large-scale translation, or in confidential projects where security is paramount.

Flitto’s very own translation API services with uncompromised security

Main takeaway

The quality achievable through the machine translation post-editing framework boils down to how much we can rely on the skills of the human editor.

We know that machine translation has come a long way for it to be at the level of fluency as it is in right now. Nonetheless, it still cannot fully grasp contexts as humans can. It’s still going to take a long time until we can solely rely on the AI to translate error-free. Until then, the existence of skilled human editors, well-versed in their own domains, remains indispensable. If the role of human editors is irreplaceable, what the advanced machine translator can do in MTPE projects is save their time.

At Flitto Localization, we make sure we make the most out of the technology available to us to produce the best outputs in projects like MTPE. After all, in today’s rapidly advancing technological landscape, being able to fully utilize the potential of AI is a virtue in itself.