Machine Translation (MT) software translates source text into a target language automatically. Yet not all MT software is created equal…

Everyone is familiar with the biggest name in machine translation:

Google Translate is used by over 200 million people every month. But Google Translate is in no way reflective of the quality which MT engines can produce these days.

Instead, Google Translate is an example of what’s called a generic machine translation engine. You might think of this kind of software being jacks-of-all-trades, masters of none.

The other part of the MT field is made up of custom machine translation engines. These are the masters of certain topics.

Custom machine translation is the way companies in industries around the world are starting to increase their productivity while reducing the costs associated with some kinds of translation…

What is a generic Machine Translation engine?

Google Translate is the best-known example of a generic machine translation engine. Other big-name platforms like Microsoft and Naver also provide similar ones.

These generic MT engines are used to translate millions of words every single day. There’s no denying that they are an incredibly useful tool.

However, the quality of translation they offer tends to be relatively low. You wouldn’t want to use them for your business.

So, why do they exist?

When are generic MT engines a good choice?

If you have a snippet of text in a language you don’t understand and you want to get the gist of what it says, using something like Microsoft Translator or Google Translate can be a good idea.

Unfortunately, there is a high chance that there will be syntax and grammar problems and other errors in the translation this kind of software produces.

This makes non-custom MT engines unsuitable for anything more than getting a general idea of what a certain bit of text says.

Can generic MT engines be used for business?

If you have a snippet of text in a language you don’t understand and you want to get the gist of what it says, using something like Microsoft Translator or Google Translate can be a good idea.

Unfortunately, there is a high chance that there will be syntax and grammar problems and other errors in the translation this kind of software produces.

This makes non-custom MT engines unsuitable for anything more than getting a general idea of what a certain bit of text says.

What is a custom Machine Translation engine?

A custom MT engine is a machine translation system which has been trained to be specialised in a set field or domain, such as medicine or manufacturing.

Custom machine translation engines provide a far higher quality of translation than a generic engine ever could.

To do this though, they need to be allowed to read large volumes of source and already-translated target data relevant to the domain in question. This “training” needs to be handled by experts using the right kind of data.

In the short term, it’s a complex task. In the long term, it leads to all kinds of benefits…

When should you use a custom MT engine?

Custom MT engines can be trained for any given industry, topic or individual projects.

They are particularly useful for situations where you have very large volumes of text you need to translate. We’re talking millions of words – something no human translator could handle at the necessary speed.

For example:

  • Many companies in the tourism and hospitality industry use custom MT engines trained on the kind of language people use when writing reviews.
  • E-commerce companies like eBay have customised engines trained to understand the specific terminology people use in online transactions.
  • Businesses in every vertical use custom MT engines trained on their own brand and industry terminology to translate their internal communications.

How to train a custom Machine Translation engine

Properly training a custom machine translation engine is critical if you want to get real value when you use it. Remember:

The goal is to create a machine which automatically translates text in a specific domain to a high enough standard to need as little human attention as possible.

To train an engine, you will need high-quality bilingual data which is:

  • Related to your field – the data you have needs to be relevant to the job you want your engine to do.
  • High in volume – the more data you have, the better. A figure of 90 000 words of high-quality bilingual text is sometimes quoted as a minimum.
  • “Clean” – errors in the training text will be “learned” by your engine. All the data you use needs to be processed by a human MT specialist in order to eliminate “noise” like this.

When choosing an MT specialist for this job, it’s always a good idea to make sure they are a native speaker of your target language. Not only does this reduce the time and cost of the engine training process, but it also produces bilingual training data which is much more accurate.

It’s the reason why Language Service Providers who know what they’re doing will only use native linguists to “clean” MT training data.

Training MT engines on “dirty data”

MT engines can be trained on “dirty data” – i.e. data that has not been cleaned by a professional. However, it’s a little like teaching a child to read using books you haven’t checked for spelling or grammar…

You don’t know the “rules” they’re going to pick up by learning from it until after they start making mistakes.

Using dirty data to train your MT engine may still result in a higher quality of translation than a generic MT engine. Yet it’s unlikely to be good enough to use for any professional purpose.

The good news is that if you do have dirty data which you’ve used to train your MT engine, you can have that data cleaned rather than starting again from square one.

The benefits of custom machine translation” Return On Investment continues to increase

The Return On Investment which custom machine translation software provides continues to grow as MT technology advances and the quality of output continues to improve.

Once again, the reason for this is that, after the initial investment, ongoing machine translation costs are tied to the amount of human post-editing time the output requires.

By properly training your machine on large amounts of clean, high-quality, in-domain data, you dramatically decrease the amount of human post-editing required – and thus your costs.

Many custom MT engine users recoup their initial machine translation investment by the end of the first project they use it for.

Productivity increases

A properly trained custom MT engine can as much as triple productivity on a project.

For some very high volume projects, a properly trained custom engine is the only practical solution.

Confidentiality

It’s a comparatively little-known fact that text translated by generic MT engines like Google Translate is stored by the owners – sometimes to use for later translations.

If your data includes potentially sensitive information, this is not a good thing. It’s another advantage of custom MT engines – and of many of the Language Service Providers who train them.

Asian Absolute, for instance, practices the kind of data security typically used by international banks.Getting the most from your machine translation software” Custom MT engine training is a critical step in making sure you reap the full rewards of the huge increase in productivity and excellent Return on Investment which custom machine translation can give you.

For those who have got used to the relatively poor quality results a generic MT engine like Google Translate can produce, the benefits of a custom machine translation engine have to be seen to be believed.

Do you need to know more about MT? Or do you already have a custom MT engine which you need to start getting better results from?

Asian Absolute helps our clients in every industry increase the productivity and profit they get from their custom machine translation software.

Comment below if you have a quick question or contact us directly for a free, friendly chat.