A Brief Intro to Conversation Design Terminology and Software
--
I’ve talked about the basics of designing a VUI and using Voiceflow to produce conversational prototypes. To make your life easier, I will keep this article short and efficient when introducing you to the concepts of VUI that any aspiring conversation designer should be familiar with.
If you have further questions and feedback, feel free to add a comment below.
The Basics
As these have been covered plenty of times, I will only give a quick recap of their meaning. However, they are key in understanding conversation design.
- Linguistic Objects: Sentences, words, abbreviations, …
- Intent: The user’s objective when interacting with an AI-based assistant.
- Utterance: The exact phrase used by the user during the interaction.
- Prompt: Users tend to jump from topic to topic (as used by Voiceflow). Prompt makes non-linear conversations possible by listening to intents when the user and the assistant are already having a conversation.
- Slots: Slots are used in utterances, for example when mentioning a time or destination, whereas variables are broader in terms of storing information.
- Variables: Variables are broader in terms of storing information, like a global variable that is being used and remembered throughout the entire experience.
- Skills / Actions / Conversational Apps: Every system uses different terminology when referring to conversational experiences. For the sake of simplicity, we will be referring to conversational apps in this article
Understanding and Synthesising Language
Often (incorrectly) used interchangeably: Natural Language Processing, Natural Language Understanding, and Natural Language Generation
Natural Language Understanding (NLU)
In order for a machine to process linguistic objects, it needs to understand them first. It is an integral part of NLP. NLU picks up the meaning of an intent, sentences, and words (symantics). It is also meant to understand the context (pragmatic) and grammar of text (syntax). Additionally, NLU even analyses sentiment.
Natural Language Processing (NLP)
NLP, as the name suggests, is processing natural language data, often in large quantities. For a complex voice assistant, their NLP can be considered as an engine. It makes sense of unstructured language data into a format that a machine can understand.
Natural Language Generation (NLG)
Simply put, NLG produces sentences that make sense to the human. It can be considered as the front-end piece of the conversation design system that interacts directly with the user.
Text and Speech
Text-to-Speech (TTS)
TTS renders text into speech, a particularly popular technology for reading out websites and ebooks. TTS is quite underrated in fact, however, many users want a solution where they can just listen to the news, social media content, translation purposes, and so forth.
Speech-to-Text (STT)
On the other hand, we have STT. This is where conversation design becomes a lot more technical. This is the science of converting human speech into text.
A great example is the subtitle feature that Youtube rolled out. It analyses what is being said during the video and converts it into text, in the form of subtitles.
User Testing
There are many ways to test a conversation-based experience. These could be simple application that you build out and prepare for testing (we will discuss software in the next section).
However, the Wizard of Oz technique is used a lot in these scenarios. For those not familiar with this type of testing, it’s basically faking the experience. The user has the impression that she or he is interacting with computer system, while, in the background, everything happens manually by a human being. The advantage of this technique is the flexibility, unlike a scripted prototype for example.
Software
I’ve briefly summarised the conversation design capabilities of Voiceflow, Adobe XD, and Protopie. Depending on how far you want to take your conversational journey, we can split these tools into conversation-first applications (Voiceflow) and all-arround UX applications (XD and Protopie).
Voiceflow
If you want to build a full-on conversational design experience, then you should starting building in Voiceflow. This piece of software is becoming, and in my opinion already is, the industry standard for prototyping and building chat and VUI experiences.
Futhermore, their latest release allows designers to create multimodal designs by adding screens and making the experience go beyond voice and chat.
If you’re interested in conversation design, then it’s almost a must to pick up and become familiar with Voiceflow.
Adobe XD
Many of us designer have the full Adobe subscription, which includes Adobe XD. Although limited, the application allows speech recognition and playback. Ideal if you want to get started or test a VUI experience.
Protopie
Perhaps less famous, but a fairly powerful tool. Protopie, similar to Adobe XD, has features that allow designers to create speech related experiences.
Conclusion
Conversation design isn’t new, but it’s gaining popularity, especially in cars and households. Furthermore, big brands, with the help of their AI technologies, are heavily investing in conversation-based experiences.
It seems very different from traditional UX Design, however, the process and many of the techniques are the same. The biggest difference is the focus on chat and voice, rather than visuals. However, many new experiences require both the visual as well as the conversational part.
It is becoming a fast growing domain in the UX industry. Even if conversation design is not what you’re looking for, as a designer I still recommend upskilling yourself in terms of building VUIs and how they could potentially impact modern user experiences, because they will.
We offer multimodal product design and strategy to deliver experiences that your users will love.
