Chatbots Must Be Able To Manage User Digression In Conversations
Fundamental principles of digression can be gleaned from how Kore AI has implemented conversation digression in their framework.
Let me start here…
Intents are the frontline of digital assistants (chatbots, voicebots) and lately there has been much innovation in terms of intents.
Innovation came in the form of granular management of intents, structure was added to intents in terms of nested or sub-intents. And lastly, management of intents were improved with merging intents, splitting intents, and more…
Multiple intent recognition is common place, and often disambiguation is implemented to assist the user in choosing the most appropriate intent.
All these developments, can be categorised under NLU Design.
From a conversation design perspective, the handling of intents have been very linear with resolution paths running in series, not parallel…if you like.
Conversation design is still very much stuck in the following sequence:
- Select and assign one single intent to a section of dialog flow
- Taking the conversation associated with that intent to its logical conclusion
- And only once the conclusion is reached, the conversation flow steps into initiating or addressing another or following intent
This is typical of a state machine process.
Running parallel intents in a conversation
We demand from systems and programs to run multiple tasks in parallel (multiple threads)…but somehow this is not demanded from the Conversation Design process.
When we as humans have a conversation, there can be multiple intents raised in a short succession…and we often address those intents in a very parallel fashion, pausing and continuing on intents spontaneously.
Virtual Assistants (chatbots, voicebots) must be able to accommodate and manage this multi-threading of intents by allowing users to pause a dialog, start and complete another task, and return to the original task.
This should be done in the following way…
- Supporting important contextual data
- Maintaining conversational continuity
- With granular control
How Is Kore AI Approaching Digression?
Below is an example conversation with a Kore AI chatbot…in this example the bot responses are simple dialog cards with links.
The user starts the conversation by asking to rebook travel arrangements, then digresses into baggage allowance.
The bot puts the rebook travel intent on hold, and subsequently return to it once the digression task is completed.
Highly Granular Digression Settings
With Kore AI digression can be managed on three levels:
1️⃣ Flow Node Level
2️⃣ Dialog Task Level
3️⃣ Virtual Assistant Level
Digression, or interruptions as Kore AI refer to it, can be set on a node level. Digression configurations on a node level enjoys the highest priority.
Dialog level digression has a higher priority over the Virtual Assistant level settings. And again, it has a lower priority than node level digression settings.
The Virtual Assistant Level digression settings can be overridden at the task level as discussed above.
Kore AI has four digression sequences:
- Hold the current task and resume back once the new task is completed.
- Discard the current task and switch to the new task.
- Switch to a new task without any notification to the user and discard the current task.
- Continue with the current task and add a new task to the follow-up task list.
As seen below…
Kore AI has also implemented the notion of nested digression, this is the number of tasks on hold which can be set.
The chatbot/voicebot keeps the maximum number of tasks on hold and ignores additional tasks once the threshold is reached.
Below you can see the interface where the on hold quantity can be set, very astutely Kore AI advises that too many tasks on hold can introduce unwanted complexity to the conversation.
Most probably there are certain customer journeys which are deemed as important enough to keep on hold. Whilst other journeys can be deemed as disposable and not qualifying for digression.
FAQs & Small Talk…
FAQs and small talk are especially susceptible to digression. As seen below, FAQs digression can be set in accordance to the rest of the conversation.
Also below, small talk can be suppressed or executed…
However, it is important for Conversation Design as a discipline to keep in step with NLU Design innovation…. in order to contribute to an exceptional user experience.
I’m currently the Chief Evangelist @ HumanFirst. I explore and write about all things at the intersection of AI and language; ranging from LLMs, Chatbots, Voicebots, Development Frameworks, Data-Centric latent spaces and more.
No-code tooling for NLU
The complete productivity suite to transform natural language into business insights and AI training data
Get an email whenever Cobus Greyling publishes.
Get an email whenever Cobus Greyling publishes. By signing up, you will create a Medium account if you don’t already…
Eliza Language Technology Community — Language Technology: Conversational AI, NLP/NLP, CCAI…
ELIZA — Where language technology enthusiasts unite.
This Is How Cognigy Implemented Digression
Users Should Not Be Constrained To Fixed Dialog Paths
Your Chatbot Conversations Must Allow For Digression
Don’t Trap Users In Constrained Dialog Paths
Actions In IBM Watson Assistant Now Support Digression
And Digression Is Integral To Having A Natural Conversation