Imagine you’re lost in Dublin. You’ve made it to the Guinness Storehouse at St. James’s Gate Brewery, but have no idea how to get back your hotel. You don’t even remember what road it’s on, so street-by-street directions are useless. Now imagine turning to your cell phone for landmark-by-landmark directions, because you remember exactly where all the big churches you passed are.
Kristina Striegnitz, an assistant professor of computer science, has helped develop a computer program that could do exactly this on phones of the not-too-distant-future.
The program is a natural language generation system, and she hopes that someday, it will evolve into a flexible and complex descendant of the simplistic systems currently used to run automated information lines.
“An example of a very simple natural language generation system is one Amtrak uses,” Striegnitz said. “Amtrak has a system you can call on the phone and it gives you train information.”
“You talk to a computer and you say, ‘I want to go from here to there,’ and it gives you an answer – a natural language answer in English,” she continued. “It’s very simple because we know the types of things the system has to say, because we know what people are going to be asking for.”
The system she created with the help of teammates like Filip Majda is not quite as straightforward, and is meant for more complex use. Majda, a recent exchange student at Union, is now pursuing a master's degree in the Czech Republic.
“Researchers are looking at systems that are more flexible and used for situations where we can’t list all the sentences the systems will need to say,” Striegnitz explained.
She envisions such systems being used to give surgeons virtual training or to provide more helpful directions for navigating on foot.
“I could imagine somebody having this on a smart phone while walking around a city,” Striegnitz said. “People use landmarks much more than street names when giving directions, so this program might say, ‘Go to the tall tower and then turn right.’”
But in order for the program to be reliable, it has to be tested and evaluated and used by real people. Bringing people into her lab, Striegnitz discovered, was challenging. There were scheduling conflicts and paying people for their participation got expensive.
So, she helped create a Web site, GIVE (Generating Instructions in Virtual Environments) to give users a chance to test the program.
“The idea behind GIVE was to set up a scenario to have real evaluations of the system through human users,” Striegnitz said. “They can access it from their desktop, rather than come in to the lab..
“In addition to our system, four other systems from other universities are running as well,” she added. “You are randomly associated with one system when you log on.”
All systems, however, are designed to enable human users to complete a 3-D treasure hunt in a virtual world where they don’t know the rules.
“Because you don’t know how the world works, we provide software that will tell you what to do to find the trophy,” the GIVE site said. “As you play, the system generates English instructions that are meant to help you solve the puzzle.”
As participants play, the GIVE server analyzes how well and how quickly they follow directions. The collected data helps the researchers fine-tune the program.
To participate, visit www.give-challenge.org. The evaluation period runs through January.
“We want to get as many people as possible to try it, so we get as much data as possible about how well our system functions,” Striegnitz said.