What is Conversational UX?
Conversation is a structured method of requesting, exchanging, and expanding on information. It is one of the primary ways that humans have evolved to learn about each other and the world around us. Conversational UX is a way of thinking about and designing the interactions between humans and the technology that tries to mimic it.
One common example of a Conversational User Interface (UI) are chatbots on websites that help you find answers in an automated way. In fact, Adobe Experience Manager supports several chatbot integrations. A chatbot can ask and answer questions, in a similar way to a human. Also, think of applications like Alexa, Google Home, and Siri that can interact with spoken commands in natural language.
How is Conversational UX Different from Regular UX?
Fundamentally, when focusing on Conversational UX, Blue Acorn iCi’s Digital Experience team follows the same process as any other design (discovery, modeling, testing, refinement). Just as we’ve developed conventions for structure and behavior for websites and web applications, we’re currently working to develop conventions and best practices for these types of interfaces.
Uncover User Intentions
This is where the team rolls up their sleeves and goes to work with prospective users. The purpose is to identify the totality of interactivity and situations you would like the Conversational UI to handle. This may be a set of specialized intentions (ex. Troubleshoot steps for setting up a Verizon router) or a set of general intentions (ex. “I can’t find my order”, “how do you make a soufflé?”, “What movies are playing near me?”, etc.).
Scriptwriting and Dialogue Management
The building blocks of human conversations are sequences. We start by scripting a story about your potential users. Once we’ve defined what actions they will likely take, we can define “success paths.” Success paths, in this sense, are sequences of events and triggers that will help them accomplish these actions. Once we have those, we work to create multiple alternative branches of dialog.
These alternative branches of dialog take into account the multiple ways that natural speakers set up the context for a conversation; as well as the multiple layers of progressive details that will follow. Each alternative dialog branch will either end in success (user got desired answer), or the confirmation of a negative (UI confirmed it can’t help with their intention).
Dialogue Management Will:
- Outline the shortest route(s) to completion
- Outline alternate paths and decision trees that govern decision-making sequences
- Outline behind-the-scenes decisions the system logic will have to make in order to support the needed dialogue
Planning Levels of Semantic Deconstruction
A deep understanding of your users based on research and analysis will guide the Digital Experience team to better understand how nuanced your conversational UI will need to be in breaking down sentence composition. Natural speakers use various sentence structures to convey the same or similar intentions. For example: “Are there direct flights from Pittsburgh to Raleigh?” vs. “Find all flights from Pittsburgh to Raleigh, and tell me if there are direct flights.”
Typically, the conversational UI will rely on a Natural Language Processing (NLP) Artificial Intelligence (AI) to parse the user’s speech or text. It will inform UI with the Context (who, what, where, why, when), the Utterances (desired outcomes), and any additional progressively disclosed information.
These are typical “conversation markers” that an NLP AI will consider and evaluate:
- Timeline (‘when’, ‘from’, ‘to’)
- Local context (the user’s location or somewhere else)
- Acknowledgments and confirmations (‘yes’, ‘ok’, ‘I don’t understand’)
- Pointers (what/whom does the desired action apply to)
- Transitions (moving from one topic to another, narrowing or expanding of context)
Planning Response Cadence
When working with a Conversational UI, response cadence is especially important; both for spoken and text output. Users expect an immediate response when they ask a question. Response time will vary with both the complexity of scenarios handled by the UI and maturity of the AI handling natural language processing.
To avoid unnatural stumbles and unexpected pauses, we plan for your UI to “stage” the answer in multiple parts. We then have the user react to each part of the answer, one at a time. For example, “There are red, white, blue, and yellow blinds, at 26 inches of width, made from natural materials, that are under $100 per set. Would you like to know more information?” is a complex question that leaves the user unsure of how to proceed in order to progressively disclose more of the sequence.
Offering a staged question that continues an expected conversational cadence: “There are red, white, blue, and yellow blinds that fit your criteria. Which color are you interested in?” explains to the user what questions they can ask to continue the dialog. At that point, the user can make a singular choice and continue the session asking for more detail.
Planning for Memory Affordances
Conversational UIs must have short term memory so they can maintain the context of current conversations. Sometimes, the UI may need to consider context from multiple conversations in order to give more useful answers. What a conversational UI can remember and how it’s allowed to access those memories is crucial for understanding the development complexity of the entire system. Some conversational UI will be paired with a neural net machine learning AI that will ingest memories and attempt to derive communication patterns. This is done to make the conversation more “natural” and add more value to the end user. For example, when you ask a conversational UI to “add paper towels to your shopping cart” the UI may respond with “I added paper towels to your cart. Would you also like to get paper plates like last time?”
What Type of Tool(s) Does the Team Use to Create a Conversational UX?
- Mind mapping tools – Scriptwriting and dialogue management
- Headless device model – Software that uses NLP AI and allows for modeling interactions. Apple, Google and Amazon all have their own SDKs to accomplish this type of modeling
- Screen device model – Same as above, but for text-based conversations
- Test Simulator – Each NLP AI package offers their own test simulator tools (for example: https://developer.amazon.com/docs/devconsole/test-your-skill.html)
What Are the Limitations of the Relevant Technology in This Space?
The efficiency and breadth in capabilities of NLP AI is the largest limitation of conversational UIs. NLP is a monumental undertaking. Companies like Apple, Google, IBM, and Amazon spend significant resources to slowly advance the capabilities of their solutions. As these solutions advance, Conversational UIs will be able to take advantage of more nuanced sentence analysis, probabilistic inference, speech pattern adaptation, and other complex language features that are currently only the domain of advanced linguistics.