A limited command set could impose a larger cognitive load on the user, as they will have to remember the exact syntax / command structure that they would need to use for the application. If it’s for features that they don’t use that often, I would expect that the cognitive load would be higher.

Here’s a Wired article discussing your point about conversations w/ a person in-car vs on the phone.

http://blog.wired.com/cars/2008/12/new-study-confi.html

I would expect that a speech app would have more characteristics of a phone conversation vs an in-car one. That is, unless the automation gets hooked up to the sensors in the car to be able to detect sudden braking or something like that to shut it up in an emergency. Of course, that brings up other issues.

I am by no means a psychologist, nor have I done any of my own research into the subject. I should note though that the conversations I’ve had with experts, particularly from AAA, seem to suggest that the danger in using a phone in the car, or the radio, or any other device for that matter, stems from the cognitive load they impose and the fact that those devices do not respond to road conditions. I’m not sure if a limited vocabulary would contribute to that, but intuitively I feel like that makes sense.

I wanted to add, though, that when I spoke to Fairley Mahlum, the director of communications for the AAA Foundation for Traffic Safety, she suggested that the difference with talking to someone on the phone versus in the car next to you is that the person next to you has some idea of what’s going on around the vehicle. They know when an accident is about to happen and can react. Someone on the phone has a very limited idea, and our technology for that matter has no to very little idea as well.

I imagine that the more we see studies conducted around these issues the more we’ll understand the nature of the problem more precisely.

It’s not.

We’re getting to a point where embedded speech rec can accommodate larger grammars, but there’s still a pretty long lag between when new technology is developed and when it’s deployed in hardware. For an automobile, it can be up to 5 years.

I agree with Emmett’s analysis, BTW.

I prefer to remain anonymous because I used to work at Nuance.

Why is talking to a person next to you in the car so different? Partly because you are not forced to use an arbitrary and small vocabulary. Years later, I still say to my cell’s voice dialer “Call Chris at the office” when it can only understand “Call Chris at work”. I make the mistake most dialing “hands free” while driving.

Another, and very BIG issue is that today’s voice apps are not really “there with you”. (see http://ejtalk.com/wordpress/?p=10)

As an after thought, let’s not forget that there are just a lot of voice interfaces that are just … well … after thoughts.