How to Make AI Chatbot Conversations More Natural?
-
This article translates Anna Prist's Medium post 'How to make your Chatbot Sound Natural,' summarizing key points in designing chatbot dialogues. It also analyzes three essential elements for making chatbot conversations more natural, combined with the translator's insights.
I’ve always struggled with direct translation, so I prefer paraphrasing after understanding the content. Recently, I came across some excellent articles about chatbots and decided to translate them while adding my own thoughts.
Working daily with chatbot platforms, multi-turn scenarios, and various parsers, I find it refreshing to explore different perspectives—it often sparks new ideas.
Today’s article is based on Anna Prist’s Medium post, 'How to make your Chatbot Sound Natural.' Here’s a summary of the six key points Anna highlights for designing chatbot conversations:
We’re accustomed to rapid technological advancements, to the point where it’s hard to imagine a future without them. As we progress, interaction devices and designs continue to evolve. Thanks to novels and movies, we’ve learned how to interact with machines—using voice commands, gestures, and virtual screens, much like Tom Cruise did in 'Minority Report.'
(You can search for 'Minority Report’s gesture-based user interface' on YouTube to see it in action.)
In daily interactions, we effortlessly use touch, voice, and gestures—no learning required.
The term 'natural' is used because interaction is fundamental to human behavior. From the moment we’re born, we naturally engage with our surroundings, grabbing or moving objects, speaking, and communicating. These interactions should feel just as natural in human-machine communication.
Bill Buxton, a principal researcher at Microsoft, once said that voice user interfaces might be the most natural, especially while driving. Clearly, when your hands are on the wheel and your eyes are on the road, voice becomes the most efficient way to convey information. Technological advancements now allow us to converse and interact with machines.
Speech is a common human skill, so it’s safe to assume your users already possess it. For VUI developers, the next challenge lies in creating dialogues/skills/behaviors and training chatbots/virtual assistants to communicate effectively and naturally.
This challenge is daunting because machines must understand our intentions while grasping contextual dialogue. To sound natural, they also need personality. Below are some tips for designing chatbots and virtual assistants.
As humans, we naturally use context without even thinking about it. We instinctively know how to adjust our tone and style when talking to children, parents, friends, or colleagues. At home, we might speak loudly and casually, but in public, we’re more mindful of our words and tone.
Chatbots and virtual assistants lack this contextual awareness. That’s why 'context' is crucial. Basic data—such as user queries/responses, authorized user information, or expressed preferences—should be leveraged. Avoid asking for information the bot already knows or bombarding experienced users with beginner guides.
Personality makes chatbots and virtual assistants sound natural. For example, Alexa is playful and has opinions that vary by country—ask it about beer preferences in the U.S. versus Germany, and you’ll get different answers. Amazon developers initially aimed to make Alexa sound less robotic, but its personality unexpectedly won users’ affection and built trust. While this principle may limit automated responses, it’s vital for user experience.
Concise wording reduces cognitive load, saves time, and sounds natural. Trim text to highlight essential information, omitting redundant details. If your bot has a screen, consider displaying summarized or hidden information there.
Anticipate users changing their input mid-conversation or answering questions in varied ways.
Chatbot speech should avoid repetition and overly formal language. Use implicit confirmations and active listening techniques to acknowledge key information. Don’t forget polite phrases like 'thank you,' 'please,' or 'goodbye.'
To prevent users from feeling lost, design dialogues with clear next steps—use prompts, guided markers, or buttons to steer the conversation.
While dialogue design is constrained by technology, these tips can help create smoother, more natural conversations. This field is still young, and we’re learning through trial and error—so don’t fear mistakes.
Remember: Good conversations are natural conversations.
That concludes the translation. Now, let’s delve deeper into a few points:
Interacting with chatbots typically involves four dialogue types: open-domain chat, task-driven conversations, Q&A (FAQ), and recommendations.
Often, these functionalities come from different teams, leading to disjointed user experiences. Imagine chatting with a friend whose tone suddenly shifts to their partner’s—you’d notice immediately.
Human language is incredibly flexible and nuanced. The same word can carry different meanings based on context or tone, posing a significant challenge for chatbots.
Even the same functionality can have varying expectations across scenarios. For example, my recent interactions with Tmall Genie:
One late night around 2 a.m., I said, 'Tmall Genie, set an alarm for 8 a.m.' It responded at full volume—startling me, especially if I’d been listening to music loudly earlier. After lowering the volume, the next morning at 7:59, it announced, 'Your alarm is about to ring…' at normal volume, jolting me awake again.
This shows how even a simple alarm-setting feature should adapt to different contexts—smarter, more considerate, and more natural.
That’s all for today!