UMBC: Your research focuses on improving speech recognition systems. What are some of the problems facing these systems?

AS: The technology itself has improved dramatically in recent years, but the nature of human speech means that we still have what are referred to as "recognition errors." This is when the user says one thing and the system recognizes it as something different. The errors that are generated are often very unexpected. Sometimes the words sound alike, but not always.

UMBC: What is your approach to correcting this?

AS: Right now, many algorithms work off of the sounds the user produces as well as very simplistic models of what the English language looks like. There are a lot of researchers working to make speech recognition algorithms more accurate. For example, some researchers are looking to integrate more complex models of what the English language looks like. But we're not taking that approach. What we're doing is trying to help people take the erroneous information produced by the system and correct it. The main focus of our work has been on dictation-oriented tasks.

UMBC: Can you illustrate with an example?

AS: Envision using speech to write an email or produce a paper or write a letter. You are speaking, the computer is recognizing what you are saying and putting it into Microsoft Word for you. What you get is a document where a lot of words are correct, but there are still a fair number of words that are incorrect. So a big emphasis for us is helping people find and correct those words.

UMBC: How significant is the problem?

AS: There are really several pieces to the problem. The first one is to find the words that are wrong and it turns out from our research, this doesn't seem to be a major problem. People are pretty good at finding words that are incorrect in what they've written. Once you've found the word, you need to be able to navigate to the word. If you envision that you are using a mouse, you would just click on the word you want to fix. But if you have limited or no use of your hands and arms, perhaps from a spinal cord injury, using the mouse is not easy and may not even be feasible. That's why we've been looking at how you do that using speech. There are a bunch of existing solutions out there, but it turns out they're very ineffective.

There are a couple of ways they work. For example, if you want to correct the word Friday, you would say "select Friday." In theory, that sounds like a good alternative. The problem is, once again, you have recognition errors so quite frequently, the command itself fails and you get what we call "cascading errors," one error leading to another error, to another and this snowballs to the point that a user can spend a great deal of time just trying to correct what started off as one simple error. Another common approach allows the user to say "move up two lines," or, "move right four words," and again you run into the problem that the commands can fail and require more work to correct.

We found when we first started working at this that people spent two-thirds of their time correcting the errors and one-third of their time dictating. That's not very effective.

UMBC: What approach have you taken to fix these issues?

AS: We've gone through and identified where people were having difficulties and why, things that were not working well as they went through the process, and we designed new ways of navigating to correct the words. In the earliest study, they [subjects] were spending two-thirds of their time correcting errors and about half of that was navigation, and the other half was fixing the word itself. After we implemented our new solutions, we've been able to reduce the amount of time they spend fixing errors to one-third of the total time spent. The time spent navigating to the error (the mouse-like activity) was reduced from about 33 percent of their time to just 12 percent.

UMBC: You worked with people with spinal cord injuries; what other groups will benefit from this research?

AS: Anybody in a situation where his or her hands or arms are not available to work with the keyboard for some reason. Spinal cord injuries were the initial population, but you could envision this being used by people with repetitive stress injuries, which is a very large and growing population. At certain stages, people with Lou Gehrig's disease could also benefit. We're also looking to take these ideas to even broader audiences, including mobile applications. If you think about many mobile technologies, you don't have your standard mouse and keyboard and you need other ways to complete those tasks.

UMBC: What spawned your interest in applying high-tech capabilities to such a human need?

AS: It was the very end of my undergraduate studies. I took a course in designing computer systems for people. I found it an interesting, challenging topic. Adding the person into the equation makes things even more complicated, because the technology will do exactly what you tell it to do, and it will do it the same way every time you ask it to complete a task. But when you put a person into the equation, five different people will accomplish the task five different ways. People are a lot less predictable. Personally, I find working with the combination of people and technology more challenging and more interesting than just working with the technology itself.

UMBC: You are also researching "situationally-induced impairments and disabilities" related to the use of multiple mobile devices. Are we becoming victims of technology?

AS: I prefer not to think of it as being a victim, but we're certainly becoming dependent. More and more people are carrying electronic devices with them and, in many situations, more than one device. We're interested in looking at how the environment you're in, or the activities you're engaged in, make the use of the technology more challenging. In some cases, the devices simply don't work or won't work effectively in that environment. How many people use PDAs to record little notes? As they're walking from one location to another, if you need to record an idea, most people have to stop, pull out the device, carefully enter the information and then continue. We're looking for ways we can design the technology to fit into what you're trying to do more effectively. Instead of interfering with your primary goal, which is to get across campus, we want to develop solutions that let you continue with your primary goal and take this secondary, technology-oriented task of recording a piece of information and make it a lot easier. We find that users often perform a task in a certain way because of the technology. From our perspective, the technology should conform to your way of doing things, instead of you conforming to the technology.

UMBC: It must be more rewarding to work so closely with people who may one day benefit from commercial applications of your research.

AS: It is. I like the focus on people. I like the focus on addressing the needs of individuals and organizations as opposed to just coming up with a technology in isolation. I came to UMBC because, in this department, all of the faculty are really focused on the human side of technology. I found that to be a really refreshing perspective.

To learn more about the Department of Information Systems at UMBC, please read Dr. Sears' welcome at http://www.is.umbc.edu/chairs_message.asp.