Conversational AI specialist Nuance has announced a new project called Pathfinder, which aims to take much of the manual drudgery out of building dialogue models for virtual assistants (VAs). The new data- and AI-driven system combines chat log transcripts with advanced Natural Language Understanding (NLU) technology to boost the 'conversational intelligence' of VAs, resulting in more efficient two-way interactions with customers.
There are three main components of conversational AI, Paul Tepper, principal product manager of AI and machine learning at Nuance, told ZDNet: to understand user requests and generate intents and concepts; to deliver answers, both simple and complex; and to enable complex two-way dialogues.
Today, there's plenty of machine learning and AI involved in the understanding part, but "when you move to answering the question, it starts to fade away a little bit," Tepper said. "For simple FAQs, there's AI involved there, but when you get into more complex answers — maybe you need to dip into a database or talk to an API — you need a lot more people. The next level is dialogue, where a person says something and maybe it's incomplete, or maybe it involves a long business process to complete. The way those kind of tasks are accomplished today in virtual assistants is via specialists — we call them 'conversational designers', who often have a background in sociolinguistics."
SEE: Special report: How to automate the enterprise (free ebook)
Conversational designers work out the business rules and processes, and then design dialogues — basically building a graph or flowchart. "This is a very human process," Tepper said, "it involves a lot of meetings — weeks and weeks of meetings, talking to subject-matter experts across the contact centre. Companies can end up with hundreds, even thousands, of pages of these documents describing their different flows."
With Project Pathfinder, Nuance is aiming to automate this time-consuming, manual and error-prone process. "The end goal, maybe several years down the line, is you dump data in and you get a dialogue system out", said Tepper.
So far, Nuance has built a tool that takes conversational data — live chat and transcribed phone calls from contact centres — and carries out 'intent discovery' where similar chats are grouped together based on similar language. "The new innovation is, once you pick one of these intent patterns, we look at the 'turn level' of the conversation: we try to identify commonalities across the different turns, or exchanges between the contact centre representative and the user," Tepper explained. "From there, we can automatically generate a dialogue graph that a designer can then manipulate — it's more like an AI Photoshop for dialogue, rather than a replacement for a designer," he added.
In the current Pathfinder system, each conversation represents a single path with no branching. "But when you aggregate across all the different chats, you get to see the different ways it could go, and from that we get what we call a 'yarn ball' chart. We can then focus on the highest-volume chats and, in the end, get to a dialogue that was created from data, without having to talk to anybody," said Tepper. This can then be exported to conversational AI tooling, or handed off to higher authority.
Customers who have used Nuance's beta Pathfinder release are also excited about using it for call centre analytics, Tepper said — to examine the structure and flow of conversations, and discover where agents are going off-script, or for training purposes, for example. Financial organisations will also find it useful for compliance purposes, he added — "making sure that agents are saying the right things," as he put it.
As a result of this feedback, Tepper said, the Pathfinder technology may be incorporated into two products later this year: "one that's fully integrated into our VA-building tool suite, and the analytics capability — we're still looking at the opportunity there, because Pathfinder wasn't originally designed to do that."
According to Tepper, this sort of tool should encourage companies that were previously put off by the laborious manual conversation-design process to create virtual assistants. The key, as ever, is data: "a lot of companies have live chat, and more and more are recording contact centre conversations — there's all this data that's just unlocked potential."
PREVIOUS AND RELATED COVERAGE
Nuance spins off automotive segment into new publicly traded company
Nuance said the move will help the company become more focused on its healthcare and enterprise verticals.
Nuance partners with Nexgen on voice recognition tools for first responders
The partnership gives Nuance a way to bring its Dragon Law Enforcement speech recognition product to a wider customer base that includes 911 call centers and EMT agencies.
Nuance sells document imaging business for $400 million
Intelligent automation software maker Kofax is snapping up the product line.
Can virtual assistants plug the healthcare gap?
The "virtual nurse" will see you all the time. Electronic Caregiver is launching what it claims is "the world's first virtual caregiver" at CES in January, though it's by no means the first chat-bot to offer medical help.
These are the practical uses for artificial intelligence in business (TechRepublic)
In order to plug AI into your existing workflow you must first understand and organize master data sets, says Schneider Electric Chief Digital Officer Herve Coureil.
CES 2019: Cortana's gone underground, but Microsoft is not out of the virtual assistant game
Cortana was a no-show at CES 2019 this week. But Microsoft's Virtual Assistant Accelerator, which lets customers build their own personalized digital assistants, racked up a few big wins.
Robots will fix us, whether we like it or not (CNET)
From offering companionship to critiquing our ping pong skills, robots are poised to be there for us.