Utterances are messages that can be interpreted by your natural language processing engine and can be used as training data. To design a great conversational experience, Conversation Designers can help the team to generate utterances that can improve the AI models – but this is not a one-person job and many other roles can help on this task.
Getting familiar with utterances and the common terms
If you’re not familiar with these concepts, these are some of the most common terms that you may encounter in a conversational AI project:
- Utterances: examples of messages that can be sent by a user
- Intents: it’s a collection of utterances and it represents what a user is trying to accomplish or expressing in a conversation. Ex: greetings, asking for something, or specifying a location
- Entities: these are keywords that can be extracted from a user message. Ex: a telephone number, person’s name, product’s name, etc.
- NLP or NLU Engine: NLP stands for Natural Language Processing and NLU for Natural Language Understanding; both are similar concepts and most engines in the industry combine these to map intents, extract entities and use utterances as training data for an AI model.
Think that for every answer that your chatbot or voice bot can give, you need to generate utterances or write examples to help your NLP engine to find the best answer to a particular question.
For example, let’s say you’re designing a chatbot for a coffee shop that helps a group of users orders an espresso. To design the conversation while improving the AI model, you will need to think about different ways that this user can express the intent of ordering an espresso:
- I would like espresso
- I want an espresso
- Can you give me an espresso?
- Please, an espresso!
In short, with this approach, you will be able to cover most things a real user would say in real life, but also help your engineers to improve the AI model at the same time. And this can be done in partnership with AI Trainers, Copywriters, UX Writers, AI Engineers, and more! In this article, I will focus more on Conversation Designers and their take on generating better utterances for AI models.
So, how do Conversation Designers design for AI and generate better utterances?
A generic approach for any conversation designer is to understand the problem, think about solutions or ways to design the conversations, write conversations, validate with real users, improve the conversations, hand over to enginers, then repeat.
When it comes to designing conversations for an AI assistant, Conversation Designers can face a new conversation design approach. They need to be more intentional when it comes to design the conversations and think about ways to generate better utterances that can improve the AI model as a whole.
To improve the AI model while designing conversations, Conversation Designers can:
- Conduct usability testing with real users – while validating the content and the flow of the conversations, they can collect organic examples or utterances during these sessions
- Write the conversations and scenarios with multiple examples – for some companies, it’s a good practice to generate many utterances while designing use cases, so this can be rapidly incorporated into the ai model
- Use AI to train the AI model – this is a Botsociety feature that enables you to test your design with an AI, helping you to automatic generate utterances and examples; sometimes, it even helps you to think about new paths
On Botsociety, Conversation Designers can design conversations and evolve them as they improve they conduct usability tests, write conversations and use the automatic test their designs. If you’re curious about how to do this and generate many utterances for your conversation designs, check this Botsociety Tip to learn how to do that.
In the end, the best way to generate better utterances as a Conversation Designer is to design conversations and test with real users that can help you to think about different ways of expressing things. Botsociety helps Conversation Designers and teams to rapidly prototype and iterate their design while generating utterances. With few clicks, you can design your conversations and export them to top-edge NLU engines.