Lie-Detection Software Could Scan E-Mail, Text Messages

[Hrach_Techie]

Researchers at Cornell University want to create lie-detection software for e-mail and text messages.

In the past three years, professors and students from several universities have coordinated with counterparts at Cornell to conduct experiments, analyze language, and compile a list of indicators of written deception. They have drawn from 40 years of research in linguistics and lies, including recent work in the context of computer media and reviews of Enron e-mails.

The National Science Foundation awarded the researchers $680,000 last summer to develop software that would sift through text and flag messages in which in the authors had lied.

"There's still an open question of whether that is actually possible or not," said Jeff Hancock, a communications professor and information science faculty member at Cornell. "Our research suggests that it is."

Passive voice, verb tense changes, and even noun or verb selection can suggest a person is lying, he said. Hancock said another indicator of written deception is the decreased use of the word "I," which is most likely an attempt to create distance.

"One of the reasons we think that works as an indicator is that pronoun use is subconscious," he said.

In interactive speech, like instant messaging and some dialogues, liars go into a "persuasive mode" and increase the length of their message by 30% to describe and explain situations, he said. Other factors -- such as individual beliefs about behavior, whether someone is accused of something or interacting with an accuser -- can complicate the process.

"In person, normally people say less because they don't want to be caught," Hancock said during an interview Thursday. "In a monologue, you may say less because you don't want to talk yourself into a trap. It's certainly not a simple thing," Hancock said. "It's a very complex problem. From a theoretical point of view, it's really fascinating."

Researchers in Canada studied thousands of e-mails from Enron for cues, like the use of the first person singular ("I") and negative emotion, and they became interested in about 10 e-mails that appeared to be unique.

"Sure enough, they stood out as problematic and of interest to the legal team," Hancock said.

Over the next three years, the researchers will study linguistic patterns in different situations and train software systems to evaluate communications based on content and context. They will try to train a software system to adapt to subtle differences in communication between friends and co-workers, expressions of fact and opinion, and even specific individuals' lies and truths.

"There are contextual parameters," Hancock said. "We can't assume that the kinds of predictors between two friends would be the same as a person writing in AP [Associated Press] style to a large audience."

Researchers have the advantage of tremendous computational power and decades of previous research. Hancock said researchers will publish the findings and the body of work they're based on. The results could be useful for law enforcement, in corporate settings, for online dating, and possibly among suspicious spouses. Hancock cautioned, however, that lie-detection software is likely to provide probabilities -- not definitive answers -- and be most effective for those with vast amounts of data.

Unlike polygraphs or brain scanning techniques demonstrated recently in Europe, which look for physiological cues, the software would focus on the "building blocks" of lies.

"Humans lie by using words," Hancock said.

http://www.theemiratesnetwork.com/ex...4-0030488344e4

[acid]

doesn't make any sense to me

[Hrach_Techie]

in terms they won't be able to account for all the lie patterns or in other sense?

[acid]

exactly. even if they collect a database of prehistory they won't be able to do that.

[W_z_rd]

By comparing given message to person's previous message perhaps it is possible to make some conclusions... But I know for sure: I can never tell for sure if the person is liyng by reading his message; so cannot computer(s) too.

[Anouk]

A professor of mine told not that long ago:

"A good academician is the one who can turn a bull-**** into a flight ticket"

In this case, I think, it could be applied to these researchers: they got a grant and have to justifiy it... how - does it really matter

[acid]

It is more than ridiculous how people are giving/approving grants to such nonsense projects instead of human disease researches. Perhaps it's some stupid software with artificial intelligence approving them.

[Hrach_Techie]

Guys! Still inexact pattern matching forms of such projects are clearly preferable over nothing as they can at least point out classes of obviously explicit lie patterns. Definitely, it's inevitable that the tool will miss some or the majority of lie patterns but there is an excuse for this as it's a new technology and in the long run the others can at least gradually build upon it

DISCLAIMER: It's just my IMO. I am not involved with the project nor do I have any financial interest in its success