Machine Translation (MT) software is getting better all the time. They can help you speed up the translation process and save money on a number of different projects.
But with there being so many different types of Machine Translation out there, how do you pick the right MT engine for your business?
In this article, we’ll show you how.
What types of Machine Translation engines are there?
The first step in your decision-making process should be to check what’s available.
Let’s get a brief overview of the types of Machine Translation software out there:
- Rule-Based Machine Translation (RBMT) – these were the first kind of MT software to be created. RBMT engines use a series of rules based on the software’s analysis of the source and target languages. These rules are then developed by human translators to improve their accuracy.
- Statistical Machine Translation (SMT) – SMT engines use algorithms and statistical models to create translations after analysing substantial amounts of data. In order to train an SMT engine, you need a great deal of already translated source and target documents of proven quality for the machine to “study”. But, when it’s properly trained and the subject domain is narrow enough, SMT can result in translations of reliable quality.
- Neural Machine Translation (NMT) – Neural Machine Translation using deep learning is producing increasingly accurate translations. In particular, NMT engines are light-years ahead of other varieties of MT software in terms of fluency. That said, the encoder-decoder approach which NMT relies upon seems to struggle with unknown terminology or unusual words. Even the latest NMT software can’t handle complex sentences.
- Hybrid approaches – though it would be convenient if all MT fell into one of these categories, most systems actually use a hybrid approach. SMT systems might follow RBMT guidelines. Or an RBMT system might use SMT for post-processing or to improve output quality. Either of these hybrids might also use input from an NMT system.
When Asian Absolute is helping you choose the right MT engine for your project, we believe it’s vital to do so from an unbiased vantage point.
Because we’re not tied to any one MT platform, we’re able to be completely objective in our recommendation. That’s something you’ll want to watch out for no matter which Language Service Provider you use.
The factors we use to decide which specific MT engine is the best for you or any of our other clients will include a whole range of details…
How to pick the right Machine Translation engine for your business – points to consider
1) Your type of content and quality goals
If you’re translating literature or creative writing, you almost certainly won’t want to use MT. Even top of the line Neural Machine Translation using deep learning has trouble with this kind of content.
On the other hand, if you have a large volume of content and you need it translated in a short space of time, MT might be the choice for you. Particularly if you don’t need the end result to be pitch-perfect in terms of accuracy.
The ideal sort of content for MT (this is sometimes called Fully Automated Useful Translation or FAUT, as usually there are no humans involved) might include:
- Internal communications
- Product descriptions
- Customer reviews
If you use a human post-editor alongside your MT engine though, you suddenly make it possible to use this type of software on a much wider variety of content. Plus, you’ll know that as a human linguistic expert has assessed it, it’s going to be highly accurate.
There is no specific MT engine which will be the perfect match to your type of content. There are, however, custom engines which can be trained on certain types of data and certain subjects…
2) Your subject matter and target languages
SMT engines, in particular, require a great deal of training data before they can deliver solid results. This is fine if you have a language pairing like English and German, where there is a huge amount of bilingual data available and the quality is all well verified.
However, when it comes to less common pairings, this can become a difficulty. The same is true in certain fields and industries. There might be a ready body of equivalent data in both languages to use to train your engine…
Then again, there might not be.
Speaking to a Machine Translation specialist is usually your best first port of call. If they’re anything like the experts on the Asian Absolute team, they’ll know about the state of play in any given field or language pair. Or they’ll know where to look to find out.
3) Your budget and turnaround time
Using Machine Translation is a very useful way to meet short deadlines and keep your budget small for large-volume work.
Be wary though, that there is a vital human role if your content is going to be highly visible or if you’re going to be relying on its accuracy…
The importance of post-editing
A post-editor is a human linguist who, as you might expect, is responsible for checking the accuracy of your MT engine’s output. They review the accuracy and style of your software’s work, ensuring that all the phraseology, terminology and so on meets your quality criteria.
Without post-editing, it’s a bad idea to use your MT output anywhere it’s going to be very visible. Selecting the right linguists is as important here as it is in any translation project. When we’re choosing a post-editor for a particular project, we look for certain qualities:
- Subject matter knowledge, in the form of industry experience and/ or qualifications
- Native speakers of the target language
Picking your post-editors will be a key part of the process of selecting the right type of Machine Translation software for your company. Asian Absolute can help you with that, of course. As well as in setting up the next important tool…
Using Translation Management Systems – and why you need to
Machine Translation software works pretty well on its own. But if you use the right Translation Management System and properly integrate it with your MT software, you will massively improve the quality of the translations you generate.
A Translation Management System (TMS) is a piece of software which makes coordinating even immensely complicated translation projects relatively easy. Just picture how a team of translators works:
You’ll have multiple specialists, usually in several different countries. There will potentially be multiple languages involved, as well as different content and assets like LiveDocs for those languages.
Your TMS will provide all kinds of helpful functionality to help you overcome this:
- Translation Memory management – a Translation Memory (TM) is essentially a database of translated phrases and part-phrases specific to your projects. Using TMs will save you money, particularly on high-volume content where there’s going to be repetition. It will also ensure your use of phrases remains consistent across a project or across multiple projects. Having your Translation Management System handle your TMs centralises their use and makes them far more efficient.
- Termbase management – Termbases are much like TMs but for terminology. They’re particularly good for helping you adhere to your brand style guide no matter which region you’re targeting.
- Managing workflows – this is another place where your TMS helps your translation team work together no matter where in the world they’re based. You can automate certain processes – like flagging content your CMS is posting online to your translation team ready for it to be localised for other regions – while making communication and managing documents easy.
- Analysis and reporting – your TMS can be called upon to report on progress, costs and output. All the work is happening through it, after all. So you have the data you need close at hand.
When to use Machine Translation software
Machine Translation is a tool which you can get a lot of value from. You can have an MT engine translate thousands of words into multiple languages every minute. Yet your engine cannot understand context in the way which a human translator can.
By choosing the right kind of Machine Translation software for your project, by integrating it with the right Translation Management System and then selecting the right post-editor, you’ll be ready to wring all of the advantages out of these increasingly handy tools – while avoiding any potential downsides.
Asian Absolute can help you plan every part of your next Machine Translation project. Get in touch now or comment below to learn more.