One application of age analysis would be judging the seriousness of threats made in wiretapped telephone conversation, Gadde said. For example, knowing that a speaker is a child might reveal that talk about a bomb is probably a hoax.
Audio analysis might also reveal a foreign-born speaker's native language. But Gadde says he knows of no program with that capability.
There are practical limits to what computers can do.
"You could [sort audio into] more than two classes [say, 'children' and 'adults'], in principal," Gadde said. "But I'm not too sure if you can distinguish between a person who's 25 years and someone who's 35 years [old]."
On the Spot
A speaker's choice of words and style of speech reveal additional aspects of his or her identity and status, said the University of Texas' Pennebaker, who studies patterns in written texts.
In an interview or interrogation transcript, he said, automated analysis can spot statements that are probably lies.
"'Junk words'—words like pronouns, prepositions, articles—are words that we've found to be particularly" indicative of lying, Pennebaker said.
By contrast, frequent use of "exception words" (such as "but," "not," "never," and "except") reflects a "complex picture of the world," Pennebaker said. Such nuance often indicates that the speaker has a firm grasp on the account being given, which suggests that the story is likely to be true, he said.
But there's no guarantee of success using that analytical technique or any other, Pennebaker said.
If the public expects absolute accuracy, he said, "it's going to be sorely, sorely disappointed."
"No system is going to be 100 percent accurate," agreed the University of Arizona's Meservy.
Still, he said, software can "augment human intellect" to help investigators recognize "what's going on when people deceive."
Bonus Applications
Some computer approaches to security have additional applications, said Paul Thompson, a computer scientist and text-analysis specialist at Dartmouth College in Hanover, New Hampshire.
Such software could also be used to detect biased writing, he said. "There has been interest in the last few years in trying to detect opinion"—in news stories, for instance—"rather than factual writing," Thompson said.
Text analysis can also examine a ransom note to determine who authored it, or assess whether misleading stock tips sent from multiple email accounts actually originated from the same person.
Big Eyes, Big Ears
Video and audio provide some data that can't be gleaned from text-only communications or transcripts of conversations, Arizona's Meservy said.
For instance, numerous pauses or other "disfluencies" may indicate that a person is "managing a greater cognitive load," he said. The speaker could be halting because he or she is making up the story as he or she goes along.
Meservy and his collaborators are working on video-analysis software that's designed to decode the motions of a subject's head and hands during a face-to-face interview.
The software tracks on-screen "blobs" superimposed on a speaker's body parts, Meservy said. Continually analyzing the paths of the moving blobs can yield clues about the speaker's state of mind, he said.
In theory, such analysis could allow video cameras at security checkpoints to size up individuals before they've even uttered a word. Telltale behaviors often mount while people are waiting in line, Meservy said.
Technical challenges remain, however. Computer programs have difficulty tracking which body parts belong to different individuals when multiple people are visible at once, Meservy noted.
It's little wrinkles like that—so complicated for computers but so simple for people—that will keep humans at the top of the surveillance food chain for the foreseeable future.
Free Email News Updates
Sign up for our Inside National Geographic newsletter. Every two weeks we'll send you our top stories and pictures (see sample).
|
SOURCES AND RELATED WEB SITES
|
