Multilingual Chatbots: Teaching Your Bot New Languages
Chatbots that can communicate in multiple languages can help you reach more users. Before you get started, read up on these linguistic and cultural challenges.
December 11, 2018
Only about 7.5 percent of the world’s people call English their native language. That being the case, it’s no surprise that many chatbot developers are adding multiple languages to their bots. There’s a lot to gain from making your bot multilingual. There are also several hurdles to clear, if you want to do it right.
Creating a multilingual chatbot rarely means running your English dialogue through a language translator and considering it done. Language, after all, is more than just words. Effective communication requires cultural awareness and attentiveness to nuance—and effort. There’s actual work involved in building a bot that communicates effectively in multiple languages!
Here are some key issues to consider as you turn your chatbot into a polyglot.
Expanding your user base with multilingual chatbots
In case you’re still on the fence about making your chatbot multilingual, remember that the more efficiently your bot can solve a user’s problem, the better. And practically everyone communicates most efficiently in their native tongue.
This simple reality can have major business consequences. Even when your audience in Germany speaks fluent English, you can still make the experience better and easier by teaching your bot German. A multilingual chatbot can give you a strong competitive advantage.
Consider Ada Health, a company with a health guide application that includes a chatbot which engages users in conversations about their health and medical status. The company is based in Berlin, but the bot is quickly learning the languages of its continental neighbors. In addition to German, the bot speaks English, Spanish, and Portuguese. And it’s about to launch in French.
Given its geographical footprint, wouldn’t many of Ada Health’s users speak English as a common (second) language? Probably. However, giving users the option to communicate in their native language makes the experience far more optimal.
And a more optimal user experience results in positive ROI. It’s just good business.
Understanding other cultures and dialogue
Consider the other languages that your bot needs to speak. How do people typically greet one another? Compared to English speakers, do users who speak the target language expect different responses to basic questions? What new linguistic cues does your bot need to understand?
As anyone who has immersed themselves in a foreign language can attest, dialog and culture have a profound impact on effective communication. One language’s motivational platitude is another’s awkward compliment. In real life, ignoring linguistic nuance can lead to uncomfortable situations. And, with a bot, it can lead to user frustration.
This is why you avoid relying on an instant translator or a web application when teaching your bot another language. You miss the distinctions and subtleties of the language you want your bot to learn—and users notice.
Big companies understand how important this is. It’s why Amazon spends years preparing bots for new language markets. Alexa has been telling Americans what kind of weather to expect since 2014, but she didn’t start conversing in French until this year—and it was quite a project.
Working with language translators
So how do you build a multilingual chatbot that can engage users across cultures? By working directly with human translators, whenever possible. And, if they’re native speakers or otherwise understand the cultural nuances of your target language, even better. Compared to machine translation, human translators help you with:
For an experience that resonates with your users, they need communication on their terms. By making conversations feel more relevant—or, in some cases, more authentic—you can more effectively meet user demands. And you easily satisfy the basics for how a chatbot is supposed to work.
This is an area where machine translators have trouble. Context, culture, and linguistic norms impact not only words but also the intent behind those words. Humans can automatically intuit intent.
Let’s face it. Some users hit you with commentary that your script isn’t equipped to handle. Human translators can help make sense of this feedback and can determine whether any of it could improve the performance of your chatbot.
If your budget simply doesn’t have room for a human translator, machine translation might be better than no translation at all. The Google Translate API won’t always provide dialogue that’s grammatically sound and contextually appropriate, but it could suffice as a stopgap measure for localizing your bot or as a solution for highly predictable use cases.
Rethinking the bot development process
Building a multilingual chatbot typically requires changes to your development workflow. There are multiple ways to integrate multilingualism, and you need to determine which is the most feasible for your organization.
An ideal situation might involve multiple human translators working in tandem with script writers and developers to create multilingual dialogue as efficiently as possible. But if you don’t have the resources for an arrangement like that (and most organizations—especially startups—don’t), you might have to start with English and go from there. The same is true if your chatbot already speaks English and you’re planning to add languages in a piecemeal fashion.
IBM encourages enterprises to first train the chatbot in a single language. After that, you can translate the dialogue using the company’s Watson Language translator and pursue a manual editorial process to smooth out abnormalities and misstatements. It’s a combination of machine translation and human refinement.
In theory, this process could take place at any point after your English-language chatbot is ready to launch. You could also run your dialogue through a different language bot or translation tool.
Anticipating bot development challenges
No matter how you rework your development workflow, structure your translation team, or incorporate technology, you’re certain to experience challenges, including some of the following:
Your bot will encounter slang, misspellings, and emojis. It needs to be able to respond appropriately. Working with native speakers is incredibly helpful here. The more adept your team is with the target language, the fewer problems you’ll experience.
Attitude and context
A great chatbot knows when users are satisfied, frustrated, or confused. It can assess users’ attitudes based on what they type or say. When users converse with your bot, they assume a bot that speaks their language also understands the attitudes underpinning what they say. In other words, they assume a common cultural context. Once again, a capable translation team can help you develop a bot that’s culturally literate.
Is your chatbot prepared to switch to English if, for example, a Spanish-speaking user suddenly starts typing English words? Dealing with this situation might not be an issue if your bot dynamically switches among languages like a true polyglot. However, switching to English could indicate frustration with a chatbot’s inability to converse effectively in the user’s native language. Be prepared for this scenario, and understand what it might entail.
An American who has lived in France for 20 years might be totally fluent in French, until you open the hood of his Peugeot. If your chatbot deals in specialized terminology, consider bringing in subject matter experts who speak the target language.
All challenges aside, multilingual bots are absolutely possible—even if you’re on a budget and can live with “acceptable” language proficiency while exploring a new market. Perfection, after all, is the enemy of—well, you know.
If you think you might have a business reason to teach your bot a new language, you probably do. In the end, reaching more users—around the globe—is an endeavor that pays for itself.