Cornell researchers developing lie-detection software
By Darren Murph 
posted Mar 30th 2007 7:51PM
posted Mar 30th 2007 7:51PM
While most of the world simply takes what everyone else says at face value, there's always been a dark market for inconspicuous lie-detecting gadgetry for the overly paranoid, but researchers at Cornell University are hoping to take lie-detection to the next level by carefully analyzing emails and SMS messages for fibs. In a three-year effort to "compile a list of indicators of written deception," the team drew from some "40 years of research in linguistics and lies, including recent work in the context of computer media and reviews of Enron emails." By carefully inspecting word choices, verb tenses, and a variety of other textual factors, the software can purportedly use "contextual parameters" to spot lies, and they hope to market the goods to police agencies, upset spouses, and of course, corporate ethics committees.So if you're ever-so-suspicious significant other (or mischievous youngster) has just recently put down the Skype headset in favor of pounding out emails, rest assured, help is on the way.[Via TechDirt, image via Cornell]
I'm betting it will be completely unreliable, and cause way more problems then it solves.
Sweet I cant wait to implement an anti lie detecting algorithm and sell it to cheating spouses. I reckon my user base would be larger anyway.
Bogus.
Much like 1960s work by Robert Rosenthal, you can used "stats" to show a better than chance rating on deception detection but this is just an electronic means of much of the work that was done towards contextual analysis over the last 30 years.
No new ideas, just simply spinning the lie-facts in another realm of analysis. Just a pencil pusher with a computer.
Someone look at the context of the question that is posed and the environment in which it was presented and I'll start to pay attention.
Just figures .... spun in different light.
Anyone else think that Ziva David could override this quite easliy????
Whatever! You cannot tell if someone is lie by reading what they type! What if I just happen to type in a way that the machine picks up as a lie?
then you better hope that your significant other doesn't believe in that technology
My wha? It's really common sense here, what if someone naturally types in a way to make the machine think that the person typing is lying?
Can it detect spam?
I think this is quite possible. Just like you can tell what someone is feeling or thinking from how they change their typing styles. But it would be terribly innacurate and nigh worthless other than for proving people's writing habits change based on their mood... and we already know that.
I wouldn't be so quick to discredit this study. No one claims the results of this application are definitive; at best it'll give you a percent chance of whether or not the writer is lying. The results are also based on a whole lot of linguistic analysis that neither you nor I quite understand. I doubt it's just a number count of how many pronouns or certain adjectives are used.
Consider an analysis of code that tries to detect whether or not the coder has a firm grasp of certain programming principles (e.g. whether or not the coder utilizes OOP design principles effectively). While the code may be syntactically correct, there certainly are trends and indicators to show that the coder doesn't quite know what he/she's doing. Linguistics is based on a grammar a lot like code is, and I'm sure a similar analysis can be made about language.
Of course you can "trick" the program, but the intent is that the writer does not know he/she's being watched and is writing casually. You can also "trick" a therapist into thinking you're bipolar. That shouldn't at all discredit the therapist's abilities to diagnose patients.