Can we design voice assistants for SA?

South African languages pose quite a challenge to voice assistants. The country recognises twelve official languages, but these subdivide into many regional, cultural and colloquial accents. Most South Africans speak more than one language, further colouring their speech.

Many dialect possibilities in the South African lexicon make it a significant challenge when designing voice assistants for this market, both in accents and different languages.

“It’s said that South Africans communicate in such a way that even their accents have accents,” says Brandon Meszaros, CEO of CXG, a member of Digital Solutions Group (DSG). “Even native South Africans aren’t always able to understand or interpret many of these accents, so expecting a voice assistant to easily achieve this is one of the major issues facing bot designers and developers working on conversational AI solutions.”

Appropriate conversational bots make an enormous difference for customer service and engagement, as well as employee support and efficiency. Hence the challenge: South African organisations benefit significantly from locally relevant voice assistants and chatbots but must navigate a complicated language landscape to get there.

The Digital Solutions Group (DSG) has been part of the chatbot and virtual assistant movement since the late Nineties. Through that experience, it isolated and tested several best practices. If your organisation wants to adopt voice assistants directly or through a partner, consider the following insights from Meszaros:

Understand the ‘why’ before the ‘how’: It is vital to understand the “why” because this makes the ‘how’ more apparent. Work closely with stakeholders to understand the rationale for automation, evolve the designs and bring the experience to life. In essence, develop a strong use case motivated by real business needs.

Define the customer’s language: A voice assistant must specialise in certain types of language, and the organisation to determine what that style of language should be. There are three key elements: utterances (what the caller said), intent (the customer’s objective for the call, and entity (information or objects referred to by the caller). The variability of common phrases and accents influences these elements. Work with front-line teams to gain a strong understanding of the types of questions and issues that customers would call about.

Keep training the AI: We continually learn new ways to speak; voice assistants also need to keep training. Train the voice assistant on customer- and use case-specific audio data—as the voice assistant handles more calls and iteratively retrains, it learns to handle more complex conversations.

Know its place in the organisation: The voice assistant will be useless if it operates in isolation. Where does it fit in? For example, it might integrate with a front-end telephony system for swift and accurate call answering. It might query databases or systems to confirm contact details and customer verification. To find such answers, cooperate with all the stakeholders, including reviewing and categorising operations. If possible, quantify each interaction type’s cost, complexity, and volume to determine which are most suitable for automation. Then map the necessary underlying workflows, systems, data, and processes.

“The variety of dialects and accents that are spawned from these means that the linguistic landscape is as wide ranging as the Karoo and as challenging to navigate as the Orange River,” says Meszaros. “Don’t let that overwhelm you. Start with a purpose, define a use case, and then work on the right organisational fit and training. Build a proof of concept or pilot to achieve those goals, and in the process, you’ll establish the processes and knowledge to keep evolving the voice assistant.”

But don’t do it all by yourself. Companies like DSG and AI trailblazers like Elerian have extensive experience and tested delivery to create regionally adept voice assistants and natural language bots. They help collapse the final barrier between humans and machines: fluent and natural communication.