The challenges of building human-like conversational experiences

October 7, 2020

In my earlier articles on building conversational experiences, I explored how they have become highly attractive to businesses, whether they are text-based (a chatbot) or voice based (a voice assistant). I also provided some ideas as to how organizations can use them to better connect with their customers. In this concluding article, I want to focus on the challenges you might encounter in building your conversational interface.

Providing customer feedback

All graphical interfaces have the capability to provide feedback to the user after a specific action. But with voice assistants or chatbots we are not able to use animations, colors or raw data as you would in a website or app.

If you have tried a conversational interface, you might have noticed that interactions provide feedback by using part of the original message in every response. For example, if I ask Alexa to “set a timer for 15 minutes”, the assistant will reply “15 minutes, starting now”.

While we try to imitate human interactions as much as possible, when using conversational interfaces there is a greater need for some kind of “confirmation”. This is why, while repeating part of the message as a confirmation might seem odd to people, the alternative of following an interaction without providing any confirmation or giving some sort of feedback about whether it has been correctly understood, can leave people with the sensation of an incomplete transaction.

As conversation user interfaces (CUIs) improve and people gain confidence in using them, we will probably see less of this approach but, for now at least, it is a necessary constraint.

The issue of context

While lots of our human interactions are simplified or shortened by omitting some information, CUIs are not that advanced. We need to provide them with context.  When you are asking about the weather for example, you will need to provide at least the context of the location you are referring to.

The workaround for these types of interactions is to set up a profile that gives the initial context of the user, and the information we might need to know, in advance to correctly solve their requests.

Building context may not necessarily mean creating a profile. Sometimes you will need to prompt with follow up questions to complete the intention.

Complexity of simple conversations

Designing conversational interactions does not mean passing the Turing Test. But, to successfully create a functional product, you do need to provide human-like responses, not just confirmations and errors. 

Yes, this may mean going out of scope of your actual application: you need to invest in providing “smalltalk” feedback, adding a certain personality to responses, and using past interactions to predict future behaviour.

Complexity then means eliminating the robotic or automated ´feel´ when converting visual feedback into more robust textual responses for our users.

Mixing languages

Different regions with different languages and dialects might use words that are not necessarily the same, and this can lead our assistants to fail with their translation. This is mainly due to the fact that the first action that an assistant performs is speech to text translation. When a pronounced word is outside of the language dictionary it might lead to it being mapped to something else. After that, when the text reaches the point of natural language understanding, there can be trouble matching the pattern which results in the assistant not understanding a command.

Those issues might not surface in a chatbot for example, since that speech to text translation never happens and that particular word can be added as a training example to our models.

Next steps for conversational experiences

While CUIs are far from the chatbots of 10 years ago, which could only reply to a specific set of words, there’s still a way to go to achieving fully linguistic comprehensive AI.

With the spread of many voice enabled devices, and their increasing popularity, the amount of interactions and available apps will increase, bringing together major improvements over the different platforms.
Hopefully by now, if you have followed this series, you will have a better understanding of conversational interfaces, how they work and how they can be applied to your business solution. Please don’t hesitate to get in touch with us to find out more about the work we are doing and how we are helping organizations.

Subscribe to our newsletter

Receive the latests news, curated posts and highlights from us. We’ll never spam, we promise.

More From

The Data & AI Studio harnesses the power of big data and artificial intelligence to create new and better experiences and services, going above and beyond extracting value out of data and automation. Our aim is to empower clients with a competitive advantage by unlocking the true value of data and AI to create meaningful, actionable, and timely business decisions.