Three Ways Your Speech Recognition System is Like Your Dog
By Ken Schafer
verb. To ascribe human form or attributes to (an animal, plant, material object, etc.)
Every dog owner is guilty of it. We pay so much attention to our pets that we think of them as human. We’re convinced that our canine companions understand our moods and we end up treating them like people. That’s why we are irrationally disappointed when our dogs do something we don’t like: use the expensive rug as a toilet, say, or insist on sleeping in the bed.
Truth be told, the issue isn’t with the dog – dogs are dogs. They have all the charm and limitations of their species. Our issue is really one of expectation. When we expect the dog to be more human than it is, we will invariably be disappointed.
The same is true of speech recognition systems. The thing is, like our dogs, they seem almost human. We can speak to them. Some of them speak back. And they understand us pretty well … most of the time.
But like dogs, our speech recognition technology (SRT) has limitations. Learning to manage our expectations is key to proper system utilization. Here are three ways SRT falls short, and corresponding suggestions for expectation management:
SRT does not deduce.
Like your dog, speech recognition systems don’t really think, at least not the way we do. Sure, they manage pattern recognition algorithms, but they are not capable of the intuitive leaps that are the defining characteristic of true thought.
Suggestion: Recognize that the phrase “Why doesn’t it just ______” is a waste of energy. All technology has built-in limitations. Accept it and move on.
SRT does not decide.
Like your dog, SRT does not know when it has made a wrong decision. Pattern recognition algorithms don’t work all the time. By definition, SRT cannot correct its own misinterpretations – or your mistakes.
Suggestion: Develop a good QA program to make sure you are using your SRT at acceptable accuracy levels. There is no such thing as 100% accuracy all the time.
SRT does not dream.
My dogs have been happy to follow me anywhere. But they have never had ideas of their own; I am the idea person in the relationship. In a similar way, incredible amounts of data flow through our documentation computers. Unlocking the relationships between various data elements could tell us a lot (i.e., trends in population health). But our systems will never initiate that kind of data mining; we need to make that happen.
Suggestion: Look for help from analytics vendors who have developed formulas to unlock data relationships. Best practices in this field are emerging.
Anthropomorphizing is part of being human. It’s why we like dogs so much and (to a lesser degree) cats. But when we expect our documentation software to act human simply because it appears to understand speech, we fail to maximize our SRT and – even worse – run the risk of allowing critical documentation errors.
Want to let me know why your dog is more human than some humans? Email me at this contact page
Posted by: SpeechCheckBack to Top