7

Reading some of the more recent articles on Medium about conversational user interfaces seems to suggest that this is already an emergent trend that will set the scene for future interface design. However, other than extending existing UI design patterns to a slight degree or applying the existing voice-recognition APIs, what are some examples of user interface design that specifically supports conversational user interfaces?

And what are some of the main challenges that need to be overcome to ensure that it will bring a more fluent user experience compared to existing standards?

  • This is an interesting question. There's a fair body of existing academic research about the social 'rules' of conversations....: en.wikipedia.org/wiki/Conversation_analysis – PhillipW Jun 11 '17 at 10:18
  • @PhillipW how much do you think applies to a human-to-AI conversation? And how much applies to conversations between multiple AIs and humans? – Michael Lai Jun 11 '17 at 12:45
  • Quick response: I think it's partly down to the expectations of the human party: If you think you are talking to a 'dumb machine' then humans will tend to grade their language - a bit like they use BabyTalk with babies (en.wikipedia.org/wiki/Baby_talk)- and expect BabyTalk back. If they think the machine is smart, then they will use a different kind of - more complex - speech. I think this is an interesting question because it takes us back to the basic principles of interaction - but without the 'graphical' element. (ie what's the audible equivalent of a 'loading' indicator ?) – PhillipW Jun 11 '17 at 13:25
5

These are three nice pieces from the past couple weeks:

On Chat as Interface by Alistair Croll

All Talk and No Buttons: The Conversational UI by Matty Mariansky

When it Comes to Bots – The Beer Test Might be More Important than the Turing Test by Amir Shevat

4
  • Understand the intent of user
  • Extract relevant information

  • Good Analysis

  • Converse with user in a natural way

Understanding the intent of user is the hard part in natural language and there are certainty rate involved for this(in many cases). In addition to this; each language has its own specific features. Questions and commands are generic forms that can be extract from user text/sound/image. For instance, English grammar has

  • Yes-No Questions
  • Wh-Questions
  • Tag questions
  • Choice Questions
  • Hypothetical Questions
  • Embedded questions
  • Leading questions

User can define his/her intention by using different methods. Machine will extract information from such sentences. The extract information shall be analyzed and a relevant response shall be decided before sharing it to user. The information shall be shared with user in a conversation form. By this simple framework, I would like to explain some crucial points to overcome to design a better conversational interface...

Crucial points here to overcome;

  • If user intent is not going to be understood, the results are not going to satisfy the user. As in example; user may need a camera for 10 days and if machine does not understand the rental options and just bring buying a camera; the analysis will be perceived as totally off topic.

  • The second part is the good analysis. There will be many conversational interfaces and some of them will bring better deals than other ones.

  • The third part; these results can be limited or too much for user. In such cases; particular user behavior shall match with user habits. In such cases; user may want a specific form of results.

  • Fourth part; tone of communication. Some people like jokes and some people are to the point. In such individual level; different personalities shall be designed.

4

In conversational design, you need to learn linguistics to understand how to write proper prompts. Most automated systems are bad because they speak like the way you write. This is because the “designer’ doesn’t have background in designing for voice.

In voice designing, you still use user-centered design process (personas, usability testing, prototyping, etc.). They’re a little different but you have to if you want a good experience. For example, before designing a dialogue, you need to understand your end-user and their expectation behind using your device. We don’t have wireframes - we have sample dialogues.

Using sample dialogs, write scenarios that your user might use and when writing, use conversational approach. For example, in a typical conversational design, you don’t use “may” since “may” is permissive. How many time have you heard, “You may hang up the call if you are finished. To go back to main menu, say ‘main menu’” This is not natural, is it? In a conversation, you don’t go back. So how else would you write this? Maybe…”If that’s all you needed, feel free to hangup. For help with anything, just say ‘help me with something else.”

Another part is contraction. In a conversation, we tend to use contraction (don’t, it’ll, etc.) often. Again, you see this often with automated systems: “It will be created tomorrow”. Replace it with: “No problem. It’ll be done tomorrow”.

Understand that language happens in context. So when designing, keep in mind the context of the user and its device:

User: What’s the weather for today. System: Currently it’s 75 and cloudy. User: And tomorrow? System: Tomorrow there’ll be thunderstorms with a high of 87 and a low 74.

The area of conversational design is too broad for me to cover all here but I think I covered some basics.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.