Have a chat - conversational interface & all that

As is the case with so many silver bullets sprayed about by the automatic fire of ITC marketing weaponry, conversational interfaces (CI)  tend to be poorly aimed. It's never going to fix a lame product or service, it will just add a layer of high fashion to a shoddy undergarment unless due diligence is done and the right resources are brought into the project..

Mapping conversations

To fully understand and appreciate the relevance of a CI to your product design there must first exist a detailed, evolving qualitative and quantitative picture of your user community to allow you to form a structural model of their behavioral dynamics. This may provide a means for determining the potential role of conversation; at least in terms of its context and direction. Consideration may also be given to implementation of a Conversation Map or equivalent analytic tool as a means of defining a suitable lexicon for your conversational UI, peculiar to your target community. A Conversation Map is an interface for very large scale conversation (VLSC) analysis that interrogates the content feed and the relationships therein between messages. It then uses the results of the analysis to create a graphical interface for visualization of interconnections..  With the graphical interface, a participant can see the social and semantic relationships that have emerged over the course of the discussion.  The Conversation Map system computes and then graphs out who is "talking" to whom, what they are "talking" about, and the central terms and possible metaphors of the conversation. 

Originally designed by (Sack, 2002) for analysis of Usenet NewsGroup content, Conversation Map as a technology has largely deprecated but the design principles are still relevant today. A new generation of related research findings and actionable codebases is exemplified by the work of (Verspoor, Ofoghi and Robles Granda, 2017). The research output from this trio enables interrogation of the semantic structures of massive network data and with it the visualization of language patterns across large message sets. If you are going to be serious about designing a conversational interface then surely you must invest in determination of a suitable ontology that includes language, behavior and environmental contexts. 

Field guides

By going to the trouble of producing a true behavioral model of your user community and applying advanced ontology analysis and visualization tools to  understand their conversations you will have a comprehensive data set that helps you answer questions of who these people are, what behavioral state they are in , what affects their transition between states, what they talk about and how they talk about it. Having done this home work, you may be in a position to start framing a germane conversational UI strategy.

when the rubber has to hit the road and you actually need to produce something that works, there are very few well written guides to steer you in the right direction.  Two sources in particular are well worth reading.  Google provide an excellent, clear concise introduction to the principles of conversation through a set of straightforward articles . Although, continued insistence upon the use of the empirically deficient notion of personas is frustrating. An excellent read for practitioners comes from the team, @TheRectangles who share a brilliant expose into their CI work.

Of particular importance is the method that Leszek Zawadzki and Alina Prelicz of @TheRectangles explain. It  provides a detailed path  for conversational overview, flow and structure as a means to organise the essential elements of a CI solution:

  • scripting;
  • frame-syntax-
  • message content
  • visual arrays-non-verbal cues
  • avatar personalisation and emotional variants-
  • emojis and animations.

Feeding the output of your conversational mapping and behavioral models into the rigors of conversation design should underpin your CI efforts. It certainly helps you frame a step by step approach to operationalising a chatbot system exactly tailored to your particular problem space rather then regurgitating a generic throwaway. Given the nature of our own advisory and build work we tend to focus on existing validated Likert-scale based questionnaires  which tend to limit the openness and familiarity of the user dialogue. That said, an overly structured conversation such as a chatbot administered scale is a reasonable sandbox for learning.  I'd still urge you to go against the current flow of self-help and learn by Google to tap into the likes of @TheRectangles for collaboration that may ultimately save you time, money and pain.

Effective delivery of a CI involves far more than rehashing templates from vendors. You need to baseline your users in terms of their behavioral states and transitional factors, identify and classify the language they use in the problem domain your app operates in and devise a repeatable documented method for incorporating end to end conversational design with non verbal mechanisms into an integrated UX.

References

Conversation Map: An Interface for Very Large-Scale Conversations. (2000). Journal of Management Information Systems, 17(3), pp.73-92.

Hindriks, F. (2009). Constitutive Rules, Language, and Ontology. Erkenntnis, 71(2), pp.253-275.

Roth, C. and Cointet, J. (2010). Social and semantic coevolution in knowledge networks. Social Networks, 32(1), pp.16-29.

Sack, W. (2002). What Does a Very Large-Scale Conversation Look Like? Artificial Dialectics and the Graphical Summarization of Large Volumes of E-Mail. Leonardo, 35(4), pp.417-426.

Sack, W. (2002). What Does a Very Large-Scale Conversation Look Like? Artificial Dialectics and the Graphical Summarization of Large Volumes of E-Mail. Leonardo, 35(4), pp.417-426.

Verspoor, K., Ofoghi, B. and Robles Granda, M. (2017). CommViz: Visualization of semantic patterns in large social communication networks. Information Visualization, DOI: https://doi.org/10.1177/1473871617693039

Dr Daryl Foy

Dr Daryl Foy is a Behavioural Scientist who specialises in the design of effective health behaviour change apps based on evidence including his own validated models for optimising persistent use. He consults to industry on how-to integrate persuasive design into LEAN product development as well as conversational UI. He can be contacted at dlfoy@mortonlawson.com