From Manchester to Texas: How can I make my bot understand?

There are around 4500 languages in the world –excluding the ones that have less than 1000 speakers— but while doing research, we realized that about 80% of the most well-known bots are built to understand and answer in English.

And that made us think, what happens with the rest of the customers? There is still a large part of the market to be covered and almost no one seems to be taking advantage of this opportunity.

According to VentureBeat, most of the bots launched are specifically designed for Asia and North America, in particular, Japan, Korea, China and the United States. So, unless you speak one of the languages of those countries it is unlikely that you have had a great user experience with a chat bot. And this is one of the issues in bot adoption around the world for most businesses.

Let’s take as an example BMW, the German car manufacturer. If they decide to launch a bot, what will they do? Go for an English one? We don’t think Germans will be very happy with the decision.

In the process to take this decision there are different scenarios that should be considered:

- Two countries with two languages:

International companies face the problem of catering to clients from different countries who speak different languages. For this problem, there are two solutions, having several bots each trained to understand a different language; or training a multilingual bot that responds to the input the user gives. The main difficulties in getting to these solutions are:

  • Having the linguistic knowledge to train a bot to understand and elaborate meaningful responses to the input it receives.
  • Including pragmatic knowledge to achieve successful interactions with the user. Forgetting about this can result in very awkward interactions where the bot makes an unfunny joke, or even worse, an offensive comment (that might not be offensive in a different culture/language).
  • Engaging the user and adapting to their speech as much as possible. The bot is doing a customer-facing work, it represents the business to the public and intends to sell or inform while engaging the customer and selling the brand. The more human-like we can make it, the more successful the UX will be.

- Two countries with the “same” language:

The other scenario is having a bot that can adapt to different varieties of the same language. For example, if a company has a branch in Manchester and another one in Texas, we cannot use the exact same bot for both countries. Why?

  • Because we need a specific lexicon for each country or at least some specific entries for each one. Even though both varieties are mutually intelligible for a human, a chat bot without the proper training could struggle to understand the user.
  • Also, our bot needs some pragmatic knowledge. Something that could be perfectly acceptable in one of the two countries could be deeply offensive to the other. Or perhaps a neutral form of treatment in one country is not acceptable to address an older person in the other country.

For example, the expression “Don’t knock me up, please.” In British English means, “Don't knock on my door and wake me up, please”, however in American English it means something completely different: “Don't get me pregnant, please.”

Another example: in some Spanish speaking countries like Colombia, people will say “Me regalas una Coca Cola” meaning “May I have a Coke?”  while in Spain that sentence will have a completely different meaning “Can I have this Coke for free?” and the obvious answer in any restaurant will be no.

If we include this knowledge in the training data of our bot, we will improve the UX greatly and the chatbots will succeed in providing the user exactly what she needs, regardless of the language or the variety of language she speaks.



Subscribe Here!