.png?width=740&name=machine%20translation%20experiments%20(1).png)
Would you like to translate 1,500,000 words within one working day into a couple of languages at once?
Every business in the world would. Thinking about global expansion, what you want are solutions that can be implemented fast. Maybe not 1,500,000 words in a day fast, but fast enough to reach your global audience in time to win them over.
Machine Translation (MT) may seem like a shiny, red Corvette among the translation services. But a Corvette driven by an inexperienced driver can be dangerous to the driver and other road users.
Likewise, experimenting with Machine Translation can have unpleasant consequences for your localization budget, but also to your company's global expansion.
Related content: 7 Tips to Avoid Wasting Your Translation Budget
Machine Translation is a fast and good solution as long as it is correctly implemented into your localization process, and if its quality is properly measured and adjusted when needed.
Working with multiple international clients, we’ve seen the repercussions of their experiments with Machine Translation. If it is implemented without a proper plan, preparation, and a few stages of quality check, it may result in incorrect translation of your content.
This can lead to a reduced interest in your offer and loss of trust in your services. In more serious cases, the offer or product specifications may be incorrectly translated, which may inadvertently mislead the buyer.
This article explains why some experiments with MT cause problems and how you can make sure your global expansion won’t follow any of the bad scenarios.
A good way to start would be explaining some terms used in this article.
Machine Translation (MT) is fully automated software that translates content from one language into another. This technology should not be confused with translation software called CAT tools (Computer Assisted Translation) and Translation Memory (TM).
Machine Translation translates without any human help and produces new translation using previously analyzed translations. The output is based on algorithms, statistics, rules or a mixture of them all.
Translation Memory contains previously translated sentences and prompts them whenever the same or a similar sentence needs translation. A translator can use the suggestion or not, but the TM does not translate by itself.
Related content: A Guide to Understanding Translation and Localization Terminology
MT technology bloomed in the 1950s. Today, MT is not a futuristic concept anymore. It is one of the solutions that can be applied to your localization project and speed up your time-to-market.
There are three main types of MT:
• A rule-based MT,
• A statistic MT, and
• Neural, which is based on Artificial Intelligence (AI).
There are also solutions that use more than one approach, and these solutions are called hybrid MT.
Different approaches are used to train an MT engine - the brain of Machine Translation. The main job of the MT is to improve the speed of translation.
There are more and more words written every day. The demand for translation is growing, but there are not enough linguists to keep up with the ever-increasing amount of content waiting for translation.
There are billions of posts and comments produced on social media. The users come from various places in the world and all of them want to understand their friend's posts and comments.
One of the reasons for the fast development of MT technology is safety. Huge amounts of money are invested by various governments and organizations into the development of machines and artificial intelligence in the service of breaking language barriers.
When it comes to MT solutions available on the market, you can choose between big players like Omniscien Technologies, the paid version of Google Translate, or Microsoft Translator. There are also other alternatives like: Slate Desktop, Kantan or Systran. You may also want to check free MT solutions like DeepL.
But how do you know that the content translated by a machine is correct? How can you make sure, that your content won't end up in the compilation of the funniest translation errors like "eating carpet strictly prohibited"?
What you need is a solution called: Post-editing.
Machine Translation Post-editing is based on revising the translations provided by the machine and either approving them or modifying them slightly so that they reach human translation quality.
If you experiment with MT and skip the post-editing step, the quality of published or printed translation may not necessarily make you and your clients happy.
Post-editing is a service provided by professional post-editors. They are linguists that understand the original text and are proficient in the language of translation.
They also need to be subject matter experts when it comes to what the text is about. They understand the rules of localization which makes your content not only correct in terms of meaning and grammar, but it also follows local buyers' style preferences, trends, legal requirements, etc.
Post-editing is much faster (by 20% - 40%) than translation from scratch. Also, the better the quality of the MT, the fewer corrections are needed and the sooner the content can be distributed among your new potential clients.
There are some cases when post-editing can be skipped.
One of the scenarios is when the translation provided usually by a free MT is meant for internal use only. You just need to grasp the general meaning of the text in a different language.
It may be an interesting article or a review of your product in another language that you want to translate. After that, you may decide if you need a professional translation, or not.
Another example of MT that can be used without post-editing is to translate content created by the users.
.png?width=480&name=machine%20translation%20for%20social%20media%20(1).png)
For example, chats or comments on social media. It can be Tweets or Facebook posts. It wouldn't be possible to post-edit content fast enough for the information to be available in multiple languages at the same time.
Actually, it would be quite funny.
Can you imagine a group of linguists waiting for let's say @badgalriri to post something and let Instagram automatically translate it into Italian, German, Japanese, Dutch, Spanish, French, Polish, Arabic, Chinese and Greek and ping the linguists to immediately post-edit it? And then doing it again with all the comments?
In order to fully use contemporary Machine Translation solutions, it’s good to start with deciding which languages you want to translate into.
The development of translating machines is based on statistics, language rules and Artificial Intelligence but it also results from processing huge amounts of previously translated texts. We may say that machines learn from reading translations. They need the original documents and their translations in order to be able to start learning.
The documents are mostly taken from Internet resources or other contributors. There is a lot of content available for the most common translation pairs like for example: English - Spanish, English - Italian, English - French.
Related content: 15 Languages for Translation in 2020 That You Should Know About
Experimenting with engines that were fuelled with an insufficient amount of translated content leads to low-quality translation. As a result, machine translation does not qualify for post-editing but a translation from scratch.
One of the other aspects to take into account is the domain and the type of content that you want to translate.
Naturally, previously translated texts that teach the engine of the machine should preferably describe your industry, service or product.
For example, if the engine was taught based on texts about animals and you want to use it for sports, the word "bat" may end up being translated incorrectly (as a flying mammal instead of a piece of baseball player's equipment).
Talking about the content type. MT will perform better when translating manuals, documentation, product specifications or instructions than a marketing text or literature.
Let's say you are all in favor of using the MT in your global expansion.
You have the English - Spanish MT engine trained, checked and ready to start translating. You had to invest money in the technology, but the amount of translated content within one week puts you ahead of your competition in the Spanish speaking regions of the world.
The cost is justified in this case.
But you also have some of the text to be translated into Italian. The number of documents is small, and you are not going to translate into Italian anymore.
In this case, investing in a costly English - Italian engine may not be logically justified.
Related content: 7 Tips to Avoid Wasting Your Translation Budget
Machine Translation, like any other technology, can be used in favor of your business or it can turn into a costly troublemaker. Fortunately, there are some tips to use in order to avoid some issues with MT.
There are a few things that you have to remember while experimenting with MT. One of them is using free, online solutions for translating content with sensitive data.
Related content: Translation Agencies vs. Google Translate Plugin: An Honest Comparison
The safety of the information you put into the free MT solutions is an inconvenient subject for technology providers. These solutions are open for everybody, their engines are placed on servers somewhere out there, and nobody knows what happens with the data you send.
One of the rules for using free MT solutions is to remove all personal data from the documents like people's names, phone numbers, email addresses.
Properly trained and set up MT translates with very high quality. Especially the neural engines, based on AI are performing so well that it is very easy to perceive such a translation as performed by a human translator.
The style and the flow of the sentences are natural, but sometimes, for reasons we don't yet understand, the engine completely misses the correct meaning.
Even for professional linguists, some of the sentences produced by MT sound like translated by a human. When they compare the translation with the original text, it turns out that both sentences (original and translated) mean something different, even though both are correct and sound natural.
For this reason, it is really a must to include post-editing on your machine-translated content. Even if your Italian speaking colleague takes a look at the translation and says "Wow, what a quality! This can't be translated by machines!?", don't skip the post-editing stage.
.png?width=513&name=benefits%20of%20machine%20translation%20(1).png)
The benefits of properly implemented Machine Translation start with a lot of time saved.
A regular freelancer can translate about 2,500 words of documentation in 8 hours. Machine translation can do 1,500,000. Taking into consideration the fact that the content translated by the machine needs post-editing, the time savings vary from 20% to 40%.
The final savings depend on the translation quality, language pair and the expected quality resulting from post-editing. We may talk about light post-editing, which should make the MT human-like or heavy post-editing, which aims for publishable quality.
Although machines do not produce a perfect translation and the texts need post-editing, there are some problems that the machines reduce. One of them is the problem of availability. Machines are never tired, sleepy, they don’t take sick leave or have bank holidays.
Machines also do not produce human errors. It is impossible for the machine to produce a typo or a spelling error.
.png?width=523&name=find%20localization%20company%20(1).png)
Pressing the "translate" button is very easy.
But, like with any automation, it is very difficult to undo the performed actions due to the complexity of the systems. So, when you start your MT, make sure you do it with a reliable translation services provider who knows how to handle it.
In order to check if your localization services provider can meet your MT post-editing needs, check if they are certified, post-editing providers.
ISO 18587:2017 provides requirements for the process of full, human post-editing of machine translation output and post-editors' competences.
If a company meets ISO 18587:2017 requirements, it is a good sign. It means that the workflow, assigned linguists and quality assurance models will meet your individual needs.
Another thing is the qualifications of the post-editors. Linguists can become certified post-editors by gaining a lot of experience as well as by training for certification.
A pile of certificates cannot substitute hands-on experience. Check if the company you want to work with has a portfolio or any success stories to share with regards to MT projects.
Last but not least, check their online resources. Look for blog posts on the topic and how they approach it. Do they sound like they are selling the one and only solution, or does the company look flexible enough to onboard different technologies in order to meet your globalization needs?
Related content: 50+ Translation Integrations That Speed Up Your Content Translation
Using new technologies to boost companies’ global expansion is exciting and we, at ATL, love to explore new ways of enabling our clients' success in new markets.
If you're looking to start your localization project based on MT and post-editing, contact us. We will assist you and help you find the best solutions for your localization needs.
Recommended articles:
How Much Do Translation Services Really Cost? [Rates & Factors]
How to Reduce Turnaround Time for Content Translation
A Lack of Translation Integration With Your System: Trouble Alert
