March 25, 2019


Extracts from Robert J. Moore and Raphael Arar’s book – Chapter 1 : “Conversational UX Design : An Introduction”

All conversations are interactions but not all interactions are conversations”.

Today’s chatbot platforms enable the designer to create a variety of different styles of interaction. Moore and Arar have observed four distinct interaction styles in today’s chatbots and virtual agents: system-centric, content-centric, visual-centric and conversation-centric styles. They are not mutually exclusive, but each emphasizes a different aspect of the interface.

System-centric interaction styles are like interactions with web search engines, but also with voice control systems. In this type of interface, the agent only recognizes one class of user actions: queries or commands. They cannot recognize other kinds of actions for managing the conversation itself, such as appreciations, “thank you,” or clarification requests, “what do you mean?”. The interaction is entirely user-driven, where the agent merely responds to commands from the user but does not initiate courses of action on its own. In addition, system-centric interfaces can handle only two-turn sequences, as in web search or voice control. Each next user action is like starting a new conversation: the agent forgets what it was talking about only a moment before.

Content-centric interaction styles are like interactions with frequently asked questions (FAQ) web pages. Interactionally, content-centric styles may be the same as system-centric ones. They are entirely user-driven and limited to a single class of user actions and to two-turn sequences. Where they differ is in their turn design: the content-centric style is characterized by long, document-like responses to user questions. This length is due to a primary concern for accuracy and thoroughness. As a result, every potentially related detail, example and contingency is built into an extended monologue.

Visual-centric interaction styles borrow heavily from web and mobile interfaces, all of which rely on the principle of direct manipulation (Shneiderman 1982) instead of text commands or natural language. With visual-centric styles, graphical elements, such as buttons or lists, are mixed into the interaction alongside natural language utterances, and dialog techniques from web and mobile interfaces, such as “Was this answer helpful?” or thumbs up/down buttons are incorporated. Visual-centric styles, rely heavily on buttons and visual elements, and are not designed to work well with the text inputs and outputs alone. Unlike system- and content-centric styles, visual-centric styles tend to be more agentdriven, although they allow some mixed initiative. The agent tends to lead the user down a somewhat linear path.

Finally, conversation-centric interaction styles are more like a natural human conversation than the other styles. First and foremost, conversation-centric styles can handle some level of conversation management, that is, they can recognize actions that function to manage the conversation itself. For example, “what did you say?,” “none of your business,” “okay,” “never mind” are all operations on other utterances in the interaction. This is a major part of what makes an interaction a conversation. In addition, the responses in a conversation-centric interaction are relatively short, or bite-sized. This enables efficiency and speed as conversations happen in real time, either through voice or text. And conversation-centric styles are fully mixed-initiative and non-linear. Either the user or the agent can lead, and the user can always redirect the conversation. Consequently, they also rely more on the user to decide what to do next.

If you’d like to know more, you can read Robert J. Moore and Raphael Arar’s book on “Conversational UX Design”: