The way humans talk to each other has always been dependent on established codes that allow us to understand each other. Languages are a major part of our identity, because they are the gateway from our inner self to the world. They are expressed differently through slang, intonations and body language.
"Our focus is to understand all humans at their most natural and offer conversational automated interfaces, bots, that can interact with them. "
It is clear that if artificial conversational interfaces are going to become an important part of our society, they’d better be able to adapt to each of us smoothly. And language, in its entirety, is a key factor in this success.
Why is language such an important part of bots?
Well, bots strive to be the most natural way of interacting with our digital world. By using natural language, we allow the user to talk just as if they were talking to a friend or a relative, on the same channels (Messenger, Slack, Kik, etc). But we can’t do it only in a few languages! To allow a smooth integration on a global scale, supporting as many languages as possible is key.
And to do that, we’ve talked to a few chatbot actors to discover how they managed languages in their bot building. Citron, the chatbot that makes place recommandation as easy as asking a friend in both English and French, was designed to be multilingual.
"We three co-founders, because of our background and education, didn’t even question language: our bot had to be multilingual and international as soon as possible."
Louisa explains that they first launched Citron in French, and shortly followed with English in May 2017.
Matthieu Bietry, CTO of The Chatbot Factory, followed the same pattern.
"When we work on bots, we design our architecture to facilitate the implementation of another language, because, well, it doesn’t affect the conception much, but saves lots of time in the long run."
Why is language tricky to manage for bots?
Languages are complex. Some work by the same rules, but some are constructed very differently. How do you manage the different word structure? Words that mean different things in different languages? How do you add another language to a bot that’s already settled in one without derailing everything?
Louisa explained that when you have a well established bot in one language, broadening its understanding to other languages is quite easy, because you don’t have to recreate anything! Translating its training can be time consuming, especially the small talk, but the process is quite smooth nevertheless. However, she recognizes that Citron was lucky: the team is fully bilingual, making training way easier.
Matthieu explains that with languages that are close to our own, both in their structure and their cultural heritage, the integration into a bot isn’t too hard because it’s just a matter of translating. However, supporting widely different languages means adapting to widely different cultures, which will impact your bot structure and flow.
He also explains that some words get tricky. French, English or German speakers casually use the word “ciao”. It’s tricky to teach a bot that “ciao” is not only used in Italian when it has been programmed to recognize the language of a sentence and respond accordingly. The Chatbot Factory team had to develop ingenious ways to allow every non-Italian to say pizza, and every non-French to say baguette.
What’s your advice to developers who want to create a bot in many different languages?
Louisa’s advice is rooted in the core of Citron: community. As we know, training your bot is essential to its success. Having a large international community is a god send when it comes to training a new language! When you start adding more than 3 languages, it becomes hard to find native speakers within your team or social circle, so a strong community is key.
Matthieu’s advice is to put effort into the bot’s translation, from its training to its answers, but not to stop there: are the APIs you’re connected to multilingual? If not, how are you going to translate their data? Is your bot dependent on time zones? Then do you want to link a particular language to a particular time zone? What about the cultural acceptance of bots and AIs?
You have to remember that a bot isn’t just code and sentences, but a tool that people will only use if the experience is smooth and effective. And when we work with different languages, we’re working with different expectations.
So if you’re thinking about building a bot in Spanish, Japanese, Mandarin, Finnish, Dutch, Portuguese, or anything else, you’re making the first step in the direction of bots adaptable to everybody, anywhere. At Recast.AI, we did it: we made language a no-brainer by supporting any language there is. Now, you have every tool you need. Good luck!
Also published on Medium.