Digital Deception

A9 Option 1: The Return of Jeff

This title has nothing to do with anything.

I decided to do Part I, and took the Perceptual Style Experiment (“You Are What You See”). I wrote for five minutes about a picture of a water bottle, though for me at least, it wasn’t quite as boring as it sounds. I got to 175 words, which was reliable enough for the data analysis. But I found the analysis to be unreliable in major ways. I’m sure that’s one of the reasons why it says at the top of the page: “Take what is said with a grain of salt.” When I was ready to write and the timer began, I described the features of the water bottle. I talked about the cap, the color and opacity of the bottle, described the label in the middle, how much water was in it … things like that.

I was shocked and horrified by the results of the analysis. It explained that I was completely normal in every category. Not what I was hoping for. I was glad to see, however, that a PHP page with major coding flaws could tell me what it did. Here were the results:

So for one, I found these results to be wrong in almost every way. I spent little time describing the words on the label, just stating that they exist. Maybe this is average, but I’m not too certain. I spent almost the entire time describing the color and opacity of the bottle, label, text, etc… and it still said I was average in every way. Considering this was a percentage of my total text, I highly doubt this was average. I spent almost no words on function, but somehow, I got the average score anyway. Same for tactile sensitivity: I never even described the feel of the bottle. And the most hilarious and telling result to me as the last one: contextual thinking. I never once mentioned the context, but because it picked up the word surrounding, lighting, or something (I had been talking about the logo on the label), it assumed this is what I was doing. Brilliant language analysis.

There are entire CS classes devoted language parsing and analysis, so I was pretty skeptical coming in, considering PHP is not exactly well-equipped for proper analysis techniques. Sure enough, the script does something very simple, and very flawed: it looks for key words. My buddy Killer Cam (aka Cameron Hall) is right: we could make this page and it might even be better. The whole time I was writing, I wanted to analyze the visual aspect of the bottle, but my score ended up being average for everything.

So this relates to deception in a couple of ways. For one, time to construct thoughts and write played a crucial role in the experiment and in all analysis. If you write for an extended period of time, it’s possible to start acting out of character. This is why there was a cap on time for the experiment. More to the point, having more time would allow you to have more time to tailor your message and deceive others. So, it seems to me there is a balance that needs to be struck between thee two factors when it comes to detecting deception through literary (or any) analysis. Moreover, message length (somewhat related to time) needs to reach a certain point to be sufficiently analyszed. Many people barely reach 50 words (the site’s stated minimum word count) in their text communications. Another deceptive thing people can do is selectively choose what we write about in our analysis, which of course, would effect what someone reading our description would think the bottle looks like. Someone below “stole” my point that it’s even possible to describe things in such a way that it might seem like a totally different object.

Finally, there is a fair amount of research about writing styles between various subgroups of people (ie. genders), but not as much about deception. It is believed that linguistics can be used to catch a liar because by investing in lying, your writing style, which is subconscious, changes because you can’t control it. People practicing deception generally use fewer first person pronouns and exclusive words, and more negative emotions and action verbs. Deceptive messages are also more informal and expressive, and are less complex in their syntax and verbiage. All of these things can be used in language analysis to find deceptive messages. However, until we learn to better analyze these things with more advanced tools, analysis will remain fairly imprecise and inaccurate.

posted by Josh P @ 11:50 AM