Should AI Chatbots Employ Dialect?

More and more companies are embracing chatbots as part of customer service. They are proven to answer routine queries quickly. Moreover, machine learning equips chatbots to handle problems with greater complexity.

So why are some companies slower in rolling out chatbots than others? It is arguable that chatbots may never match the human mind’s ability to problem solve, or have the necessary empathy to deal with certain issues. But there’s evidence that chatbots can save businesses time and money. However, many people are still of the attitude that they would rather be served by a human being than a chatbot.

Should people be more ready to embrace chatbots? The solution could be to program chatbots to communicate in a way that’s more ‘human’.

The art of conversation

A global survey of 3,500 people worldwide found that 65% of customers using an online chat function would prefer to be speaking to a human rather than a bot. 14% of customers then went on to say that their main complaint about chatbots was that they had ‘robot-like engagement with few human qualities.’

Dialect and accents are two of the main ways we differentiate how human beings speak to one another. Therefore, incorporating them into chatbots could make them seem more ‘human.’

This doesn’t mean you’re suddenly going to encounter a slew of Cockney chatbots. Instead, chatbots will be programmed to understand the nuanced differences in how humans communicate with one another.

Just act natural

Since the beginning of the 20th century, scientists and programmers have set out to create machines that can communicate exactly like humans. The Turing Test, for example, has been a benchmark for success for over 65 years. The test involves a series of keyboard conversations between a machine and a panel of judges. If an AI can pass for a human in 30% of these conversations, they have passed the Turing Test.

Communication gaps between chatbots and humans can be bridged by developments in Natural Language Processing (NLP). Broadly speaking, NLP is the process by which chatbots decipher and analyse human language. This is the technology behind services such as Google Translate, as well as voice-activated assistants such as Siri and Amazon Alexa.

NLP has already become considerably advanced. Without it, there’d be no way for chatbots to understand humans at all. The next step is for communication between chatbots and humans to become indistinguishable from conversations between two people.

The imitation game

Researchers at the Moscow Institute of Physics and Technology recently created AI that can interpret accents at a far greater success rate than preceding models. The researchers took resources from the George Mason University’s Speech Accent Archive, an extensive collection of speech samples across a range of languages and dialects. These resources were used to create a scalable phonetic dataset.

The first issue the researchers encountered was coming across terms and phrases not present in the archive. They rectified this by creating a predictive model, which then allowed the AI to process many more potential variants of words. The AI listened to speech samples in an attempt to recognise accented words. The AI accurately recognised 59% of 800,000 words.

Considering how many computers fall short of the 30% benchmark set by the Turing Test, a 59% accuracy rate for detecting accents shows how far the technology has come, not to mention its future potential.

At We Build Bots, we are excited about how bots can transform companies, leading them to not only save time and money but to provide better service. To help realise the potential of artificial intelligence for your company, visit our bank of AI-related resources. Alternatively, you can contact us today to book a demo.