Sales is about conversation. Chatbots bring this age-old element of commerce to the online world. Multilingual chatbots make on-demand information accessible to a company’s global customers, creating value in marketing as well as customer support.
Gartner forecasts that 25% of customer support and services will be integrated with virtual assistants by 2020. While no figures are available on how multilingual the bot population is, some locations are more inclined to use chatbots than others. European consumers are more receptive to chatbots than those in the US: 50% of French consumers hold a positive opinion on bots as opposed to only 32% of Americans.
Currently, chatbots are mostly used in the leading languages of global ecommerce: English, FIGS (French, Italian, German, and Spanish), Portuguese, Arabic, Russian, Chinese, and Dutch. They have also made steady inroads into long-tail languages from India and Africa to support the regions’ emerging e-commerce industry.
Not all chatbots are born equal. While some are purely menu- or button-based and pose the least challenge, localization-wise, chatbots that are keyword- and context-based aim to provide a more fulfilling user experience. Hence, the latter need to have better language skills.
Menu/button-based chatbots are pre-loaded with canned responses. They are good enough in short conversations, but have limited capability to respond to questions with more variables or long-tail questions.
Keyword-based chatbots essentially dip into a knowledge base to respond to queries from the user after they “listen” to specific keywords that users type in. They make use of previously translated data from the knowledge base. However, when queries become more nuanced or similar keywords are used, keyword-based virtual assistants can get confused.
Smarter Does Not Mean Easier
Chatbots that remember, try to understand meaning, and then respond appropriately need to have better language skills than their humbler peers. People believe that anything they talk to must necessarily have human-like capabilities in language, thus, expectations can be high.
There are also the challenges brought on by people using multiple languages in the same chat or transliterating into another language. For example, in India, it would be common to have users typing out Hindi queries in English. In Africa, people may mix English and Swahili in Kenya or French and Arabic in Algeria.
Plugging into a machine translation API can go some way by virtue of keyword detection, but may not be ready to work with live customers at scale. Even when the likes of Facebook have bots that could easily detect language given all the information in the user’s profile, it is still not enough for a fluent user experience.
Conversational user interfaces (or CUIs) — which is essentially what a chatbot is — need to be supported by systems powered by natural language processing and machine learning. Language designers (a new job description) are required to aid this learning, manage terminology, and adapt the script to new locales.
Designing virtual assistants to conduct a human-like conversation is a tall order and requires a different set of skills than those supplied by developers; it needs the creativity of poets. And when the chatbot goes international, it is required to be similarly steered through various languages by transcreators, who combine the expertise of copywriters and translators. Additionally, domain-specific fine-tuning would require collaboration with subject matter experts.
The technology is still relatively new and chatbots may currently be best suited to filling in the gap before the handoff to human agents. Even at this stage, chatbots are expected to bring in cost savings of USD 11bn in the retail, banking, and healthcare sectors.
However, with advances in artificial intelligence, human-machine interactions are expected to become more commonplace, not just more efficient. And longer conversations may develop. Humans and machines are currently working on that.