February 2, 2002

11:36 AM

No more false IDs on Metafilter!Now researchers in Italy have developed a program that can spot enough subtle differences between two authors' works to attribute authorship.
posted by rushmc (14 comments total)

It turns out that SDB and skallas were actually the same poster, and that MiguelCardoso is the Unibomber! Who'd have thunk it?!
posted by rushmc at 11:38 AM on February 2, 2002

Oh my god! It says I'm pregnant! Wait... I'm male...
posted by SpecialK at 1:26 PM on February 2, 2002

very cool, but without more specifics it looks more like an evolutionary application of string processing rather than some revolution in the field. Probably requires *much* more text than is in a typical MeFi post.
posted by Llama-Lime at 2:55 PM on February 2, 2002

I wonder what Don Foster, the literary scholar and detective, makes of it. It probably does a lot of the legwork but you'd still need a keen eye to actually make attributions.

::::rush mc::::currently pretending to be
posted by MiguelCardoso at 3:22 PM on February 2, 2002

>Probably requires *much* more text than is in a typical MeFi post.

Filtering something written in text rather than rewritten into text from manuscripts long lost is actually (I would think) easier.

For example, one could check when you miss capitals at the start of a sentence (not a knock, just an observation).

Just as a manuscript has a distinct writing style by the author, typing has a distinct typed in style. For example: Do you always put two spaces after a period? Always two after a colon or question mark? One or two after a comma? Do you tend to use etc... or etc? Do you push the ellipsis key or use three separate periods? Em-dash n-dash or hyphen? Do you capitalize after an em-dash? Periods in your acronyms?

Etc etc ... … , — And–so–on. TYVM, H.A.N.D.
posted by shepd at 3:38 PM on February 2, 2002

"Do you push the ellipsis key or use three separate periods?"

Wait a minute, there's an ellipsis key? Why doesn't anyone tell me these things?
posted by mr_crash_davis at 3:46 PM on February 2, 2002

I hate it when I type three periods into Word and it's converted into an ellipse. If I wanted an ellipse, I woulda damn well put an ellipse there.
posted by ZachsMind at 5:07 PM on February 2, 2002

Wait a minute, there's an ellipsis key? Why doesn't anyone tell me these things?

I was just about to ask the same thing...give it up, shepd!
posted by rushmc at 5:25 PM on February 2, 2002

That article's a bit off, from what I can tell. This one actually tells you what the "revolutionary program" is, and the truth may surprise you.
posted by gleuschk at 5:30 PM on February 2, 2002

Cool it uses gzip! Gzip rocks. But bzip is better.

Oh, and the ellipsis key is alt+0133 (depending on your character set). I guess its just a holdback from the old BBS days. :-)
posted by shepd at 5:48 PM on February 2, 2002

or option-semicolon on a mac.
posted by machaus at 6:41 PM on February 2, 2002

Works…! (use numpad)
posted by EngineBeak at 6:53 PM on February 2, 2002

we need an online lie detector now? why? we already have ip tracers and such... really, i don't think that someone with 2,3, or 4 different user names somewhere matters. i have more then one aol screen name, and i don't consider it a "crime". though i don't pretend to be other people on it either. whichever name i'm on, it's always me. james. i don't let my other personalities online at all. they can't even have their own weblogs or sites. aren't i a horrible domanate personality to my other personalities?
posted by Kafei at 7:18 PM on February 2, 2002

Filtering something written in text rather than rewritten into text from manuscripts long lost is actually (I would think) easier.

That is very true, typographical considerations might add or subtract a few bits from the entropy. However, I still say that given the large number of posters, and the typical length of a post, this solitary method would not be enough. This could definitely be part of a larger system, where each algorithm get its own vote…

Still, I'm wondering why this got published in Nature. Bayes classifiers have been around for a long time (pre 1991?), and work on precisely the same principle that is described in the Nature article.
posted by Llama-Lime at 8:34 PM on February 2, 2002