Why post-editing matters – the true cost of machine translation

Machine translation, AI and LLMs have transformed the way businesses approach multilingual content. With promises of speed and reduced costs, it’s no wonder so many turn to automated translation solutions. But machine translation isn’t quite the cost-cutting quick win it seems. The reality is more complex.

Beneath the surface, there are hidden costs. Especially if you publish raw machine translation output. Quality issues can damage your brand and cost you customers. It can even lead to legal or compliance risks. That’s where post-editing makes all the difference.

Contact Planet Languages today for expert post-editing support. We’ll work with you to achieve the perfect balance between cost, efficiency and quality.

It is possible to strike the right balance between cost-efficiency and quality. Knowing what machine translation really costs is the first step. It’s the foundation for building a sustainable multilingual content strategy. Let’s explore what these costs involve, and how post-editing can help you maximise ROI.

Understanding the complete cost picture of machine translation

AI translation might seem like a quick way to cut costs. After all, low monthly subscription fees or character-based pricing are hardly budget-busters. But the full cost picture includes much more than just the price of the software.

Direct costs include:

• Licensing and subscription fees for MT engines
• Token-based fees for LLMs
• Integration costs with your CMS or translation management system
• Training costs for your internal teams

Add to that the indirect costs:

• Inconsistent output
• Brand dilution
• Time spent correcting errors

Plus the costs involved in improving the MT output so that it’s tailored to your content:

• Maintaining glossaries
• Creating style guides and other reference material

The price of poor quality in machine translation

Raw machine translation output can contain surprising – and sometimes alarming – errors. The internet is full of examples of machine translation fails. These range from mistranslated restaurant menus to unintentionally offensive product descriptions.

If you’re a business that uses machine translation, what sort of errors can you expect? Typical errors include:

• Incorrect handling of proper nouns: Places, brands or people’s names might be wrongly translated.
• Gender and number mismatch: Common in languages that tend to omit the pronoun.
• Inconsistent terminology: Especially across longer documents.
• Over- or under-translation: A tendency to add or skip parts, especially in longer texts.
• Over-smoothing: The translation sounds fluent, but key details can be wrong, even the opposite of what was meant, like do versus don’t. Think about the impact of such a mistake in instructions for taking medication.
• Bias: The output may reflect training data biases.

How might these errors affect your business?

Let’s consider the impact machine translation errors might have on a global retailer. Automatically translated product descriptions could contain misleading information. This could lead to:

• a surge in returns
• reduced revenue
• a spike in negative reviews

The price you pay often comes in the form of lost trust and missed opportunities. And fixing these errors later is often far more costly than getting it right the first time.

How post-editing drives value

AI translation post-editing involves editing the raw MT output. The aim is to achieve a result that is comparable to human-quality translation. At Planet Languages, our machine translation post-editing services are fulfilled by specially trained linguists.

Post-editing can be faster than traditional translation workflows

Post-editing can be significantly faster than traditional translation workflows. How much faster? A report published by industry researcher Slator cites productivity gains of around 50%. But this depends on a lot of factors:

• the language pair
• the MT engine or LLM you use
• the post-editor’s skills
• the content type
• the expected quality level

Machine translation + post-editing = better quality

Machine translation output always benefits from post-editing by a specially trained linguist.

Your user manuals will be clearer. Confusing instructions lead to confused end users. Post-edited user manuals mean fewer support calls for your customer service team to handle.

Your product descriptions will be accurate. Misleading claims lead to unhappy customers. Post-edited product descriptions mean fewer product returns and improved sales KPIs.

Post-editing is often cheaper

Post-editing costs between 30% and 60% of the cost of traditional translation. The higher the quality expectations, the closer the cost gets to traditional translation rates. But the savings can be significant, especially when you consider the cost of errors.

• Inaccurate product descriptions: Retailers estimate that processing returns costs about 66% of the original item’s price.
• Mislabelled food products: The average cost of recalling a food product is $10 million.
• Unclear user instructions: A call to an IT service desk could cost between $20 and $100, depending on the complexity of the caller’s query.

Post-editing food product label translations of around 300 words in five languages might cost you €300. It compares very favourably to handling a recall.

Finding the right balance

Not every piece of content needs the same level of editing. Understanding the different types of post-editing can help you choose the right approach.

Light post-editing

Light post-editing focuses on correcting major errors. Style and consistency don’t come into play. This is a good option for user-generated content, where speed is more important than polish.

Full post-editing

Full post-editing, on the other hand, aims to bring the translation up to near-human quality. It’s best for materials requiring clarity and accuracy.

Ask yourself:

• Who is the target audience?
• How visible is the content?
• What are the potential consequences of errors?

Matching the post-editing level to your goals helps you stay efficient. It also ensures you’re spending your translation budget where it matters most.

Measuring success through quality

To justify any investment, you need metrics. Machine translation post-editing is no exception. There are various metrics out there, each with pros and cons.

Before post-editing

Machine translation quality estimation

Quality estimation uses algorithms to automatically predict the quality of machine-translated text. Within a CAT tool, it provides a quality score per segment. It can help you decide which machine-translated segments need post-editing. This technology is still in its infancy. Systems offered by commercial providers may be prohibitively expensive. And open-source systems require someone with programming skills to set it up.

After post-editing

Edit distance statistics

Edit distance statistics shows how much work a post-editor did overall. This is measured by a score of between 0 and 1. 0 means the post-edited version is no different from the MT output. 1 means that the post-edited version is completely different from the MT output.

At Planet Languages, we document the edit distance for every single MT post-editing project. This means we can draw on years of data relating to performance by language pair, MT engine and content type. This data guides our approach to pricing and ensures that we find the most appropriate solution for each of our clients.

Beyond metrics, you need robust quality assurance processes. Regular reviews and terminology checks help maintain standards. This is especially important when working with multiple languages or markets.

Working with experienced post-editors who understand your business is key. Their insights can help refine MT output and workflows over time.

Conclusion

Machine translation has its place in modern localisation programmes. It should always be paired with expert post-editing, though. When you invest in the right level of post-editing, you protect the quality of your content, your reputation, and your bottom line.

Understanding the true cost of machine translation and post-editing leads to informed decision-making. It helps you create multilingual content that delivers the results you need.

Key takeaways:

1. Raw machine translation often has hidden quality costs
2. Post-editing balances efficiency with accuracy
3. ROI depends on choosing the right MT strategy

Optimise your MT strategy with expert post-editing support

Contact Planet Languages today for expert post-editing support. We’ll work with you to achieve the perfect balance between cost, efficiency and quality.

About the author

Bethan Thomas has worked in the language services industry for 20 years. She helped Planet Languages achieve certification to ISO 18587 soon after the standard was first published. This standard demonstrates commitment to incorporating MT into workflows responsibly.