Posted on 05/16/2004 12:05:42 PM PDT by E. Pluribus Unum
An Irish graduate student has uncovered words blacked-out of declassified US military documents using nothing more than a dictionary and text analysis software.
Claire Whelan, a computer science student at Dublin City University was given the problems by her PhD supervisor as a diversion. David Naccache, a cryptographer with Gemplus, challenged her to discover the words missing from two documents: one was a memo to George Bush, and another concerned military modifications to civilian helicopters.
The process is quite straightforward, and according to Naccache, Whelan's success proves that merely blotting words out of declassified documents will not keep the contents secret.
The first task is to identify the font, and font size the missing word was written in. Once that is done, the dictionary search begins for words that fit the space, plus or minus three pixels, Naccache explained.
This process yielded 1,530 possibilities for word blanked out of a sentence in the Bush memo. Then, the text anaysis routine checks for words that would make sense in English. The sentence was: "An Egyptian Islamic Jihad (EIJ) operative told an XXXXXXXX service at the same time that Bin Ladin was planning to exploit the operative's access to the US to mount a terrorist strike." Just 346 words remained on the list at this stage.
The next stage is to involve the brain of the researcher. This eliminated all but seven words: Ugandan, Ukrainian, Egyptian, uninvited, incursive, indebted and unofficial. Naccache plumped for Egyptian, in this case.
Whelan subjected the helicopter memo to the same scrutiny, and the results suggested South Korea was the most likely anonymous supplier of helicopter knowledge to Iraq.
Although the technique is no good for tackling larger sections of text, it does show that officials need to be more careful with their sensitive documents. Naccache argues that the most important conclusion of this work "is that censoring text by blotting out words and re-scanning is not a secure practice".
According to the original report in Nature, intelligence experts may consider changing procedures. ®
And just how do they know the result is correct?
Too simple?
Interesting method. Wonder how many documents are floating around out there that have just a few words blacked-out but the full-text versions are still classified? Must be zillions of 'em.
just a computerized version of a known method
You don't know 100%.
But it is likely in the example to be right by at least 99.9%...
Only a limited number of words will fit the space exactly. From that only a few words will make sense relative to the context. In the article's example only one word remaining fits the context.
Chances of being correct: very high.
Seems like an overly fancy way at guessing to me.
Retype the document with a fixed number of "X"s for all redacted words.
just an enhancement of a method, used for years, to make an educated guess about a censored true-copy typed document.
works fine, so long as acronymics are not a factor.
smart.
Don't you mean mono-spaced fonts?
Does anyone have copies of redacted Clinton files on which to practice?
Perhaps. If a word is x-length with a mono-spaced font (that is, where each character is the same width), you know it's x-letters. With a proportional spaced one, a word x-pixels long is variable on character-length, but might be deduced by adding different combinations of letters that will produce that length.
But if you vary the space between letters randomly (say +/-10 pixels?), that should make it impossible to begin to deduce what the individual characters are. A patch to word processing software would probably do it...
South Korea
North Korea
They look the same length to me - must be a context thing. Maybe it's like interpreting chads.
ping
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.