What Is Machine Translation? A Guide for Busy Content Professionals

September 6, 2021
Post feature
Is machine translation the right solution for your content process?
Machine translation holds the promise of faster, cheaper, and easier-to-use translations. But, are those promises really true?
There are several pitfalls and challenges that you can encounter if you approach machine translation in the wrong way. As a busy content professional, it’s all too easy to overlook some key considerations and end up with an expensive mistake.
This simple guide provides an overview of how you can get the most from machine translation and overcome common challenges.

What is machine translation?

Machine translation is a software-driven approach to translating content. Linguistic algorithms are used to automatically convert text from one language into another. Approaches to machine translation include classical rule-based algorithms, statistical translation, and neural machine translation.
As a content professional, the basic benefit of using machine translation is that it can provide almost instantaneous translation of your content.
Unlike conventional translation — which exclusively uses human translators — the automated software holds the promise of significantly reducing your translation timescales.
Machine translation services are often sold to users as being easy-to-use, quick, and cheap.
But, this is not exactly true…
In practice, using machine translation requires much more skill and strategy than many content professionals first realize. However, it is possible to put machine translation to good use in your content creation workflow as long as you know what you are doing.

Can machine translation replace human translation?

A common question from companies interested in machine translation is whether automated systems can be a replacement for human translation.
The answer to this question is a rather unhelpful: Yes and No.
More specifically…
  • Yes — Machine translation can replace human translation in some specific use cases. The algorithms can be applied to certain steps in your content creation workflow to help you achieve particular strategic objectives.
  • No — Machine translation shouldn’t completely remove human translation from your workflow. Although some companies do try to use it in this way (for example with auto-translated versions of their websites) the end result is a poor quality translation that reflects badly on the company and doesn’t meet the needs of the consumer.
Like any other type of automation or technology, machine translation is just a tool.
How you use that tool will determine how successful it will be for you and your company.
Inline image

How does machine translation work?

There are 2 levels that we can use to look at how machine translation works. The basic level, for content professionals, and the complex level, for software engineers.

The basic level: The practical level

At the most fundamental level, all machine translation algorithms follow this 3-step process:
  1. Your source content is fed into the algorithm as text.
  2. The algorithm converts this into the target language.
  3. The software outputs the new content translated into the target language.
This is basically all you need to know to use most machine translation systems.

The complex: Machine translation algorithms

Let’s look a bit deeper at the functional part of machine translation — the algorithm itself.
The most common algorithms fit into the following groups:
  • Rule-based machine translation — This is a rather outdated form of machine translation that was first used in the 1970s. It splits the source text into words or word-chunks then directly translates those words into the target language. As it ignores context, it is usually filled with errors but can be useful for specific applications such as weather reporting where you need to translate exact terms in the same way every time.
  • Example-based machine translation — This is similar to the previous approach but takes entire phrases of text as the input and translates them using pre-translated phrases. As a result, it’s similar to what is done by terminology management systems.
  • Statistical machine translation — This has been the most common approach to machine translation for years. Machine learning algorithms are used to automatically detect statistical similarities between thousands of texts in the source and target languages. The algorithm then uses the learned knowledge to generate translations automatically.
  • Neural machine translation — The state of the art in machine translation. Instead of using statistical methods based on phrases, a set of neural networks automatically extract the “information” from the text in the source language. Then, a second set of neural networks converts that information into the new language.
Which approach you need will depend on your specific situation.
In some use cases, the less “intelligent” approaches are better as they provide higher accuracy for specific phrases. In other cases, you may need to use a combination of statistical and neural algorithms.

Problems with machine translation

Whichever approach to machine translation you use, there are always going to be some problems. Machines are just not as adaptable as human translators and are thus prone to making mistakes.
For example, with neural machine translation, 3 core sources of problems are:
  1. Systems perform badly when they are used in a different domain from the one the system was designed for. Even if a system works well for general translation, it might be inadequate for your unique industry.
  2. Systems don’t work well for uncommon words or phrases, which are often found in the content created by global businesses working in specialist industries.
  3. The longer the sentences, the worse the quality of translation. This means that systems tend to be less accurate if you’re translating into, say, legal Portuguese where sentences are often very long.
You can overcome all of the potential problems you might encounter by applying the technology correctly. However, this requires specialist knowledge in the languages, domains, and technologies. A good translation provider will be able to help you navigate these challenges.
Inline image

Machine translation vs human translation

Probably the most helpful way to think about machine translation and human translation is not to compare them at all.
They are both useful approaches that have their place in a well-designed translation process. For example, you might combine machine and human translation to help you achieve continuous translation, as we do with some of our clients.
Even so, here are some of the key differences between machine and human translation:
Machine translation Human translation
Accuracy Accurate for some situations. Highly accurate and domain knowledgeable.
Speed Within a few minutes. Flexible to your workflow but takes days.
Cost Cheap in itself but quality assurance steps can be significant. Depends highly on the project but includes quality assurance at all stages.
Ease of use Variable ease of use but complex to set up optimally. Easy with the right provider. People are involved at all stages.
Knowledge required Requires specialist knowledge to get the most from it. Very little with the right provider. People will help you throughout.

How to decide if machine translation is right for you

Are you unsure if machine translation is suitable for your translation needs?
The best way to find out what approach could work for you is to talk over your needs with a content strategist.
You can book a free Global Content Strategy call with one of our strategists and find out more about our approach and how it applies to businesses like yours by contacting us directly.