Enhancing machine translation with human expertise: the power of post-editing

Ever popped over to Google Translate to find a few words to impress an overseas client, or to fake fluency in a foreign friend’s language?

Then you’ve experienced the magic of machine translation!

You may have experienced its limitations too, but more on that later.

So how does it work?

Put simply, machine translation (MT) is a language translation process that automatically translates text or speech from one language to another using computer algorithms and artificial intelligence. Instead of using human translators to produce multilingual content, machine translation systems analyse the input text and generate translations in the target language.

By analysing vast amounts of multilingual data and breaking down sentences into smaller components, it can find and replace content with new words and phrases in the chosen language.

Driven by breakthroughs in artificial intelligence and natural language processing tech, in the 70 years since a team from Georgetown USA demonstrated its first system in 1954, MT has come a long way.

And it’s just as well, as in today’s globalised world, with companies expanding into new regions with growing pace, a combination of automated translation systems and the linguistic expertise of experienced human translators, is helping businesses expedite the translation process and reach broader audiences quickly and efficiently. 

Interestingly, a report by Gartner indicates that by 2025, 80% of worldwide B2B sales interactions will occur in the digital realm and so the need for efficient, multilingual content generation is only going to grow.

It remains the case that the vast majority of content still needs a degree of human post-editing to deal with the translation challenges of accuracy, especially with complex or particularly niche languages, cultural nuances, and idiomatic expressions.

But as translation technology becomes more reliable, accurate, and easy to use, it’s becoming clear to businesses that in certain circumstances machine-based translation, when followed by a more in-depth human post-edit, can offer a faster, cheaper alternative than the purely personal touch.

Don’t worry, you don’t need a PhD in computer science to understand this bit, we’re just going back to basics to offer a little more insight into what goes on in the ‘mind’ of machine translation.

• MT analyses input text to understand structure, syntax, and semantics so it can break it all down and work out grammatical elements such as nouns, verbs, and adjectives.

• Various algorithms and models generate translations; these are either rule-based, statistical, or neural network based, and sometimes a model will use a combination of all three:

–      In statistical machine translation (SMT) algorithms learn patterns and associations between words and phrases from a collection of translated texts in the source and target languages to generate translations.

–  Neural machine translation (NMT) uses artificial neural networks to predict the likelihood of a sequence of words and can model complete sentences more effectively than traditional SMT approaches.

• Once the MT system of choice has analysed the input text and generated a translation, it outputs the translated text in the target language.

That’s the technical talk over; let’s look at the pros and the cons of it all…

In the pros corner

1.     Translation efficiency and speed – machine translation systems can translate large volumes of text quickly and efficiently, making them ideal for time-sensitive tasks or processing large datasets.
   
2.     Cost-effective – compared to human translation services, machine translation is often more cost-effective, especially for high-volume or repetitive translation.
   
3.     Translation consistency – machine translation ensures consistency as it generates the same output for identical input texts, reducing the risk of inconsistencies that may occur with human translators.

In the cons camp

1.     Translation accuracy issues – MT systems may produce inaccurate or incomplete translations, especially for complex or ambiguous texts, idiomatic expressions, or specialised fields such as finance or medicine.
   
2.     Lack of cultural understanding – machines can’t be relied upon to understand the tone, style, and cultural nuances of a text, leading to translations that miss, or even change, the intended meaning of the original piece.
   
3.     Translation quality – depending on which languages are being paired for translation, how specific material needs to be, and the calibre of data being drawn from, the MT output doesn’t always make the grade.

What’s evident from the pros and cons discussion is that machine translation delivers in terms of speed and cost – the technology accelerates the initial translation phase of taking content from one language into another – but it can’t think, and it can’t ask questions!  

However, exciting advancements in neural networks and AI-powered translation have helped to improve how MT deals with ambiguous words and phrases, allowing it to consider the context in which they appear so it gets closer, more often, to the original intent.

And of course, the more data that’s available, and the amount of ‘training’ the networks have, the more effective their learning from more sources and adapting better to different language pairs and specialisms.

But, and it’s quite a big but, no matter how much data is available, and how much training a machine has, the fact that they aren’t sentient beings is always going to be a problem when they’re being tasked to deal with human language.

Machines lack the linguistic and creative abilities of human translators, making them less suitable for tasks that require creative or subjective interpretation, such as literary translation or marketing copy, or material that needs high levels of technical accuracy such as medical, financial, and health and safety documents.

Expert and experienced translators are worth their weight in gold by providing:

Quality assurance: they meticulously review machine-generated translations to identify and correct errors, inconsistencies, and inaccuracies.

Linguistic expertise: bringing a deep understanding of language nuances, idiomatic expressions, and cultural references to the table, experienced translators check and refine machine-generated translations so that the finished piece can truly resonate with its target audience.

Contextual understanding: where machines can’t see context, humans can! They can also question ambiguities and make informed decisions to produce translations that accurately convey intended meaning.

Post-editing efficiency: while human translation from an original source text can be time-consuming, post-editing by humans of machine-generated material is much quicker.

Adaptation to style and tone: translations that meet the style, tone, and voice of the target audience is a must in many instances and can make the difference between success and failure, particularly across global marketing campaigns. Articulating ideas with the right language to suit the market and keeping branding consistent are important aspects of the human post-edit stage.

Specialised knowledge: where specific terminology or subject expertise is needed, human translators bring valuable knowledge and insights that machines may lack. They can accurately translate technical jargon, industry terms, and complex concepts.

Subjective judgement: being able to think and apply subjective opinion is a very useful human trait! Human translators can assess translation fluency, do a sense check, and make informed decisions to enhance translation where necessary.

Continuous improvement: through post-editing, human translators provide valuable feedback to machine translation systems, helping identify areas for improvement and fine-tune algorithms.

How deep the post-editing phase needs to dive is very much project- dependent. 

Here’s a brief run down of why it’s needed, what it involves, and what the end result should be:

Full post-editing:

When it’s appropriate – where translated content will be published or distributed to external audiences, and a high level of quality is required. It’s often used for official documents, marketing materials, or client-facing content.

Its scope – thoroughly reviewing and revising machine-generated translation, a human editor will check for accuracy, fluency, and that the text sticks to original meaning. This includes correcting errors, improving language flow, and refining style and tone.

The result – meeting professional standards, at this level it should be hard to tell the difference between human and machine-translated material in terms of quality and readability.

Heavier revision post-editing:

When it’s appropriate – when machine-generated translation is of such poor quality, with a high degree of inaccuracy, it’s time to call in the big guns of extensive reworking and revising to get it to an acceptable standard. Hopefully the least-used level of post-editing!

Its scope – the translator may need to re-translate portions of the text or even start from scratch to ensure the accuracy and clarity of the final output. The focus is on achieving a high-quality translation that accurately reflects the source content.

The result – polished, coherent, and error-free translations; basically, it’s a start a-fresh approach that ends up being more human than machine.

Now for the post-editing benefits of mixing machines and the human touch

While machines are quick, cheap, and efficient, humans understand nuance, can ask questions, check for accuracy, and find work arounds to tricky translations.

Put them together and you’ve got the perfect partners to look after a whole world of translation tasks!

Along with some case study stats, here are some of the tangible benefits of combining machine translation with human post-editing:

• Making the most of machine efficiency: MT delivers quick, and cost-effective translations and when followed by human post-editing, there’s a significant reduction in the time and effort required for translation.

Research from professional services company SLD found that post-editing machine-translated content was up to 40% faster than translating from scratch and maintained comparable quality levels.

• Keeping consistency flowing: MT systems can produce consistent translations for repeated phrases or terms within a document, or across multiple documents. Human post-editors ensure that this consistency is maintained throughout the translated content, providing a cohesive and professional result.
• Enhanced accuracy: while machine translation algorithms have advanced significantly, they may still get things wrong, especially in complex or ambiguous language scenarios. Human post-editors can correct errors, provide clarification, and ensure accuracy through linguistic expertise and cultural understanding.

A study on eBay’s localisation efforts reported that post-editing machine-translated content allowed them to maintain quality and accuracy while reducing translation costs by 30-50% compared to traditional human translation methods.

Managing the translation workflow

We’ve seen that overall, depending on the level of accuracy and quality required, working with a combination of machine translation and human post-editing offers a powerful option for organisations looking to produce content for global audiences, and optimise time and resources.

But striking the right balance between automated systems and human involvement is crucial to making the most of the benefits this translation strategy offers.

To achieve maximum efficiency in terms of output, accuracy, and costs, businesses and their localisation partner, need to:

Put quality assurance mechanisms in place – regularly check standards and incorporate human checkpoints within the workflow; this maintains quality while keeping the efficiency gains of automation intact.

Track performance – collect feedback from translators, reviewers, and end users. Invest in ongoing training, skills development, and technology upgrades to enhance the capabilities of automated tools and human translators.

Conduct a cost-benefit analysis – evaluating the ROI of automation versus human involvement in translation workflows means adjustments can be made to resources or workflow strategies to maximise efficiency and reward.

Working with language service professionals is the best way to make the most of machine translation and expand global reach effectively and efficiently.

As genuine language experts, at Comtec, we’ve finely tuned our machine translation models to give our clients the best quality output.

Every piece of our machine-translated content is always reviewed by a real human translator, who we select for suitability for each project. Depending on your budget, the review process can be as involved or light-touch as time and expenditure allows.

We continually invest in the most up-to-date tech-enabled software; from handling translations at scale to generating content in multiple languages at the touch of a button, we will help you hit your business goals.

If you’d like to find out more about translation trends, or our translation and localisation services, we’d love to hear from you!

GET IN TOUCH