Is Natural Language Right for You?

NEW YORK (SpeechTEK 2010) — Natural language has always invoked more questions than answers, said Patrick Nguyen, chief technology officer at Voxify, who introduced what was to be a candid discussion about natural language solutions for interactive voice response (IVR) systems. 

David Attwater, senior scientist at EIG, addressed the potential pitfalls of using natural language—and how to overcome some of them—as well as the reasons a company might want to use natural language.

You want to be careful not to implement natural language just for aesthetic reasons, he cautioned, saying he knew clients who were very happy with natural language and others who ended up going back to the most basic touchtone. 

However, Attwater said one of the most surprising findings is that natural language was successful in stopping callers from simply zeroing out of an IVR. “Customers who zero out in applications essentially say, ‘I want to speak to an agent,’ and if you say to them, ‘Actually we can get you to someone, but it would help if you can tell us why you’re calling,’ at that point they strongly engage,” Attwater explained. “It’s a great moment to actually get the benefits of call steering to the right destination and therefore not double-handle calls, which is a particularly powerful use of natural language.” 

Attwater noted a lack of surprise about how implementing natural language didn’t change users’ likelihood of engaging with the machine. Users were also prone to “under-specify” their call reasons, expecting—and rightly so—that the machine wouldn’t be able to handle complex language, he said. 

However, natural language can really delight customers; it takes only one or two turns to get through to what they want to do, Attwater added. “You can also lead with menus to get your big bang for the buck on calling reasons.”

Overall, Attwater asserted he is pro-natural language, and said you can get great customer service even if a small amount of self-service is sacrificed. 

Following Attwater, Phillip Hunter, senior user experience designer at Microsoft Tellme, took the stage. “I’m going to take less of a data approach,” he said, adding he wanted to analyze the failures and successes and what could be learned from both. 

“Many failures usually start without a proper appreciation of the challenges ahead. Looking back at the wins, we can tell success is really about perseverance,” he argued.

Like Attwater, Hunter also cautioned against implementing a natural language IVR without careful consideration, despite the growing popularity of voice interfaces on mobile devices and in the car. “Opportunity isn’t just keeping up with the hype or building an application because someone else has one. You have to look at what will be meaningful and effective for your customers,” he said.

Hunter then gave helpful tips to the audience about how to implement a good natural language IVR, saying it should be looked at holistically. “You want a proven design pattern to save yourself time,” he argued. 

He also stressed the importance of simplicity and grammars. “You have to get the words right,” he asserted. 

The first 10 seconds of an IVR are crucial, Hunter added. “Especially for natural language, remember that users can’t play if they won’t stay,” he noted.

He and Attwater both agreed that tuning is another critical component of a good, workable IVR. In concluding, Hunter said completion rates are the most important pieces of data to examine. Other data is important, he said, “but nothing will tell you how well you’re doing or what your problems are like task completion rates.”

He also urged audience members who are thinking about implementing natural language to move “confidently, but cautiously.”