Even machine translation software like Google still has languages that defy translation.
In April 2006 Google launched a service that quickly became every internet user’s go-to for quick language translations. Google now boasts that it can translate 108 languages covering 99% of the internet population.
Yet in a world with over 7000 languages why aren’t more being included?
Many languages defy machine translation software, even though they are spoken by millions of people such as Bhojpuri (52 million), Fula or Fulani (65 million), Quechua (8 million), or African languages such as Luganda, Twi, and Ewe. So why is it that languages like Czech or Swedish, who have relatively smaller or similar numbers to these other languages, get translation support while the others are barely recognised?
Machine translation, like Google Translate, rely heavily on algorithms that are learned from human translations that require millions of words of translated text called parallel corpus. For translation machines to be effective they require a staggering number of parallel corpora for each language. An ideal parallel corpus will have content from a variety of contexts such as novels, news reports, and other pieces of writing that make up a language.
For languages like Czech or Swedish, as they are part of the European Union, a large part of their parallel corpus comes from official parliament documents. These countries are also important for big tech companies in terms of eCommerce marketing, language translation services and more, meaning that they have a larger parallel corpus to work with. With other languages, a large basis of their parallel corpus has come from the bible, which resulted in some entertaining doomsday prophecies from Google Translate prior to 2016.
In 2016, Google started using a new technique called neural machine translation which claims to have reduced translation errors by 60%. Neural machine translation is a type of artificial intelligence that can mimic some forms of human thinking. Sounds like something out of a science fiction novel, right? The neural machine translator can associate meaning to certain words and phrases. It can look at a sentence as a whole rather than translating each word. However, the database required to make this neural machine translation effective is still substantial.
Neural machine translation has been effective for select languages but what about the thousands of others?
When West Africa was hit with Ebola or when Haiti was hit with an earthquake in 2010 difficulties occurred when those that were there to help could not communicate with the locals to get them the resources they needed. With little translation support for the languages spoken in these areas, it shed light on the need for diversifying machine language translations.
With COVID-19, health information has been needed in many languages which machine translation has been incapable of helping due to poor translation quality.
Further, for countries that have low literacy rates or no written language, locals may not even be literate in their mother tongue, using voice messages to communicate which increases the need for audio translation.
So, while expanding on neural machine translation is revolutionary in terms of very basic internet communication and translations it lags in terms of international need and diversity, especially in times of crisis. So what about for business?
Google Translate is a convenient tool so it would be a stretch to say never to use it as it usually gets the basic understanding of a text, however, it is far from ideal in getting a quality translation that would be needed for business. Especially if you are aiming to enter the global market, need website localization, eLearning translation services, or are bolstering your eCommerce platform.
Here is why Google translate is not effective enough to be used in business.
One of the appeals of Google Translate is the speed in which it produces translations, however, this comes at a cost as it doesn’t equate with a quality translation. When you use certified translation services you are guaranteed a properly formatted and grammatically correct quality translation. Further, Into23’s translation services offer 24/7 availability with a quick translation turnaround making them nearly as fast as Google.
Google does not have to be accountable for any inaccuracies in its translations as it is a free service. Any user can also manually input their translations and at times malicious and incorrect translations are allowed through. What’s more unnerving is that Google isn’t even accountable to your security or privacy as it collects data on whatever content you place into its text box to translate. When you work with translation professionals, your confidentiality and privacy are ensured.
The translations from Google Translate will not be catered to your specific business needs and you run the risk of having nonsensical or inaccurate translations. In today’s global market it is important to speak to clients and customers in their own language, such as with website localization, and if the first impression of your content is incorrect it sends the message to any prospective customers that you are not the right business for them.
Misinterpretations in formal and legal documents have the potential for serious safety or financial concerns which can lead to legal disputes. In a study performed on the terms and conditions of airlines, it detailed the risks of machine translations for legal documentation and its possible negative outcomes.
No matter what type of business you run if you need to translate in any language, using multilingual translation services is crucial for business success. From transcreation for marketing to eLearning or eCommerce translation services, Into23 offers high-quality translation services in any language. Into23 works on your time and your schedule with 24/7 accessibility and fast turnaround. Get a free translation services quote today by filling in the form below or uploading your files to our quick quote portal.
Advances in machine translation (MT) mean enterprises now have a sophisticated translation solution in their toolkit that can translate quickly and at scale. Long gone are the days of weird menu translations and Yoda-like results. But given the recent rapid advancements in artificial intelligence and machine learning, companies must navigate how to optimally deploy this productivity-boosting approach alongside human translation. Knowing where and when to use machine translation will ensure translations are cost-effective and fit for purpose. Embracing translation technology and innovation in the right areas is the way to increase engagement and efficiency. Read on to find out the criteria you need to consider when deploying the latest machine translation solutions.
Though you may think language translation technology is a relatively modern phenomenon – after all, computers have only been around since The Babbage Difference Engine back in 1822 – its roots stretch back, all the way to the Arabian peninsula in the 9th century, where one al-Kindi translated ancient Greek mathematics, science and philosophy texts that had been lost to European civilisation, helping spark the Renaissance in the process. He developed various systems based on frequency analysis and statistics, key concepts in MT. Now, AI translation software enables clients to customise according to subject area, such as meteorological reports. This has massively widened MT’s applicability and usefulness.
Machine translation was initially developed in the 1950s, and has since been transformed through continuous advances, diverging into four categories: SMT, NMT, RBMT, and Hybrid Machine Translation. SMT, or statistical machine translation, automatically maps sentences in one language into another, whereas NMT, or Neural Machine Translation, encompasses a neural network that relies on algorithms working together to process highly complex data inputs. RBMT, or Recurrent Batch Machine Translation, replaces the input texts with translations of a set of translations of the same text, and Hybrid Machine Translation combines elements of both NMT and RBMT.
The advantages of using machine translation mean it is a very effective and efficient solution in a company’s toolkit. Firstly, machine translation is incredibly fast. It can process huge volumes of text in a near instant. Therefore, it improves efficiency and productivity. Companies that deploy machine translation typically see an improved profit margin, all else being equal. Secondly MT is scalable. If you need to translate a short document or an entire library’s worth of text, MT can handle it. Lastly, and partly as a result of the first two points, machine translation is much more cost-effective than human translation. Before you rush out to onboard a range of machine translation solutions, there are many circumstances where human translation is preferable, and numerous others where a hybrid approach of machine translations processed by human editors is best.
Getting the best out of machine translation requires optimally deploying it. Several factors will determine the ideal approach. A Nimdzi survey of 33 localisation buyers found 22.6% report extensively using neural machine translation. The survey notes that sectors like media, video gaming and marketing are laggards in MT adoption, mainly because they require high levels of cultural sensitivity and creativity that MT as of today can’t match. That’s not to say MT isn’t making inroads into these areas. For example, world top-10 gaming company Electronic Arts (EA) adopted MT tech quite early in its development. Notably, in areas where content is intended to prompt emotional engagement, EA uses human-edited MT translations. The survey also found that in circumstances that directly impact business revenue, human translation is preferred.
Applying those findings to business activity, MT is useful for quickly transmitting a message to large numbers of people in various locations. Even in this instance, it’s always wise to have a human check the copy. The content lifecycle is also a consideration. For short-lived content, such as product specs on a short run of merchandise, then the return on investment is not there for human translation, which costs more than MT. Here, the requirements for quality and timeliness are key determinants.
Another great use case for MT is where recipients are aware that the content they are consuming is translated by a machine. This enables them to read with caution.
In a sign of just how far MT could go in the future, a group of scientists recently launched a project to decode sperm whale ‘speech’ with a view to enabling whale-human communication. That would be an interesting one to add to the digital translation services already available. The Cetacean Translation Initiative is using AI to understand whales’ clicking sounds, known as codas. The scientists are deploying natural-language processing, which processes spoken and written communication, to that end.
MT is improving all the time, and as it does, it becomes applicable to an ever-expanding set of scenarios. However, we aren’t anywhere near the point where MT is good enough for businesses to abandon Machine Translation Post-Editing (MTPE). Finding the right balance is key, and a professional translation agency will help you navigate the optimal configuration of MT, human translation and MTPE. Get in touch today to speak to an expert and explore your options.