Should we use human translation or machine translation for multilingual customer service? As a company expands, the need to build and maintain strong customer relationships becomes more important. Businesses with global ambitions recognise the need to communicate with customers in their native language and may require multilingual customer service.
Communicating with customers in their preferred language facilitates a sense of being heard. As a result of personalized messaging, customer satisfaction increases. But providing multilingual customer service is harder than it sounds. So where do you start?
Economies of scale and multilingual customer service
Many customer service teams employ people who are proficient in languages for which they receive a high volume of customer interactions or problems. However, there are many languages in which companies receive few inquiries. To scale multilingual orders, businesses often rely on automation or machine translation at this stage.
While Artificial Intelligence (AI) has helped humanity with many remarkable achievements, it is lagging behind regarding languages. A few years ago, for example, an earthquake in Indonesia killed 98 people and injured many more. However, several Facebook users discovered that their screens filled with balloons and confetti when they used certain phrases, such as Selamat, which means “safe” or “unharmed” in Indonesian. It turned out that in other contexts it also meant “congratulations”. Facebook’s technology was able to translate the phrase, but was unable to pick up the context, resulting in celebratory photos appearing at the time of the losses.
The most difficult path for artificial intelligence, machine translation (MT), has been in the spotlight for some time, and for good reason. However, the quality of MT is questionable, especially in terms of accuracy.
To understand the difficulties associated with Machine Translation, it is essential to understand how it works. There are four main approaches currently in use. The most basic approach is Statistical Machine Translation (SMT), which uses a huge amount of multilingual text to find matches between source and target words. At the other end of the spectrum is the more advanced approach of Neural Machine Translation (NMT), in which the system tries to mimic the neural networks of the human brain. The NMT database consists of previously translated texts that the machine analyses and learns over time.
Context, culture, human emotions
Although translation technology has come a long way, it still does not include the three most important variables that determine quality and accuracy: context, culture and human emotions. Let’s take a closer look:
Although translation technology has come a long way, it still does not include the three most important variables that determine quality and accuracy: context, culture and human emotions. Let’s take a closer look:
Context is essential to grasping the meaning of a translation that makes sense in a given situation. French, for example, distinguishes between formal and informal forms of address based on the nature of the relationship. Tu in French is a loose way of saying you, while vous is the formal version. Usually, tu is used to address a close friend, a close cousin or a younger person. These relationships cannot be analysed by machines.
Language and culture are closely intertwined. Unlike language, however, culture cannot be taught or programmed. Culture must be experienced and acquired. What is the hardest part? Each region’s culture is as vibrant and specific as the next.
Artificial intelligence and machine learning cannot understand or express human emotions. Machines can rarely distinguish between tone of voice and purpose. There is a fine line between, for example, being nice and sounding overly intimate. This line should not be crossed in a professional atmosphere, especially with clients!
Making it simple, machine translation can be compared to non-playable characters (NPCs) in video games. Do you remember Nigel Billingsley from the sequels to Jumanji? Every time he was addressed by another character, the character followed a predetermined script. It did not matter where he was, or who he was interacting with; his voice inflection was always the same! However, he only said what he was instructed to say.
Support tickets translated by humans
If multilingual customer support is not practical, i.e. you do not need to employ a large number of in-house linguists, and machine translation is not supported, how can you provide multilingual support? Intelligently, of course! The recently released Translate By Humans software for Freshdesk facilitates multilingual customer support by translating ticket responses by humans. The software allows Freshdesk agents to speak to customers in sixty or more languages, regardless of the agent’s mother tongue!
Freshdesk agents receive automatically translated enquiries in the default language of their choice, to which they respond in the same language.The response is then sent to Translate By Humans via the app. A professional project manager works with experienced translators from around the world to ensure that the response to the ticket is accurately translated and sent to the agent in record time!
Leveraging technology
Thanks to the app, you no longer need to employ in-house staff, pay additional project management costs or rely on faulty machine translations. By using translation technology, you can easily monitor all agents and their activities, manage orders and sort tickets from a single dashboard, while saving a significant amount of money!
The app provides the option to use a customised translation memory, which contains previously human translated text, tailored to your brand. If you need to translate the same or similar passages repeatedly, the database will make suggestions to the translator. If the suggestion matches the context, tone and cultural appropriateness, the translator will use it and you will pay less. As the database grows, translation prices decrease, but quality remains constant!
