Thursday, September 9

Too Good to be True Department

Notice of Predisposition: I hate W. Okay? Believe me. Ask anybody.
Disclaimer: I don't have any special talents regarding document forensics.

Right. The Boston Globe and 60 Minutes Bush national guard service stories were based on materials allegedly from the personal records of the now-deceased Colonel Jerry B. Killian. It seems quite obvious to me that the August 18th document is a crude forgery.

I say a crude forgery because it was done with MS Word using out-of-the-box defaults for font, font size, margins and other settings. If you type the text into Word (I did, look here) you get an identically-appearing result, minus the scrawled signatures and the (trivially simulated) photocopy/fax artifacts.1

The giveaways are:

  1. Alignment of the first character of the date line with the period in the first paragraph.
  2. All four line breaks in the first paragraph.
  3. Superscript "th" in "187th", fourth line, first paragraph. This is what you get with Word's default settings in the "Autocorrect Options..." menu dialog "Autoformat as you type" tab "Ordinals (1st) with superscript" setting.
  4. As was immediately observed, the whole damn memo was in a proportionately spaced font, pretty freaking good for a typewriter issued to an Air Force Reserve desk in 1973.2

I've just noticed that both Powerline and Ratherbiased pretty much cover the same ground.

Message to Democrats: Given the forger seems to be either very stupid or alternatively, hoping it will be detected, this could easily be a Republican dirty trick. DROP THE TOPIC, NOW

Update (9:20pm): Wow, there's a lot of stuff in the Blogosphere about this. The most convincing, showing an overlaid fax and Word reconstruction like mine is Little Green Footballs.

[1] Formatting details: Five tabs before typing the date line, three returns after. Double returns after "Memo", "Subject" lines, and each paragraph.

[2] Yeah, sure, the IBM Executive series typewriters. Give me a break.

5 comments:

Anonymous said...

I saw the initial story and I thought "Wow! Those are devastating." Then I saw the follow-on and thought "Wow, that sure would be an obvious forgery."

And then I paused and thought -- "Hang on! Maybe a little skepticism is called for here, especially in view of the humbling issue of being wrong at least once on this. Could it be these came from a typewriter?"

Consider the most important issues:

1. An IBM typewriter existed that could produce this.
2. Microsoft Word, if it does nothing else, successfully reproduces the kinds of run-of-the-mill office memos and letters produced by executive secretaries following standard operating procedures from the 1950s through the 1980s.

It might surprise us that a human being in the 1970s would use exactly the same defaults as Microsoft Word has for documents today, but that's how those defaults got there. I wouldn't even be surprised if somewhere deep in the bowels of Word is a comment saying something to the effect of "Make a letter come out exactly like it says on page 15 of the 'Better Typing with the IBM Whizzolectric Proportional 69000' from 1968."

And, lo and behold! a blogger has now identified some tell-tale features of the document that appear to evidence typewriter origins.

http://www.dailykos.com/story/2004/9/10/34914/1603

Note specifically the comment on the character '4' and the floating 'e'.

The last word on this has certainly not been written.

AMS said...

I understand your point: modern word processors descend from high-end typewriters and hence deliberately ape many details of layout, fonts and defaults. I thought of this myself when originally looking at the documents.

But.

Have you looked at any of the overlays or blink-comparitor GIFs that bloggers have been creating? Or better, create them yourself.

There's all sorts of other indications that the documents are forgeries, but the exact match between the fax and the output from Word should be enough for a reasonable observer.

Anonymous said...

I have looked at the pixel overlays and, yes, the letters matched up very well.

But what's the power of this as a test of the hypothesis that the documents were therefore created on Word?

I'm going to assume that the margins and sizes could easily be (and probably are) adjusted to make the text identically centered for most of these comparisons, e.g. the underlying number of pixels in images is meaningless, since that depends on details of scanning that are not available (e.g. the size of the copy worked from, and whether it had ever been blown up or shrunk.)

This then comes down to: how likely is it that a) the font face overlay would be that close? b) that the typist would decide to do line returns, spacing between block elements, and indentation just like Word does it? c) any other giveaways?

My answer is that from all I have read a) is pretty much meaningless, given the standardization of certain proportional spacing fonts, e.g. the null hypothesis gives us this one just as well. As long as you choose to type exctly the same sequence of characters in the same font face, you'll likely be able to overlay two samples and get an excellent match, as long as two different machines implementing the face are fairly faithful implementations.

b) Well, it would take a little luck, but is it really so hard to believe? The human just has to choose to indent to certain common stops and set up margins the same.

c) The evidence here, according to some, is both for and against:

Pro--
Some folks who claim to have used IBM machines of the right type in the 1970s say that apparently the font in the memos more closely matches a common one used on them: something that might be "Bembo" or "Delegate" rather than "Times Roman".

Some folks claim they can see an inconsistenly floating "e", which definitely bespeaks a typewriter.

I can't see all this in a quick glance at their pictures.

Some forensic "experts" have already looked at stuff on the web and said they believe it's some variation of a 70s era IBM typewriter.

Contra--
Other than the coincidence of choosing the exact same indentation and spacing, the best evidence here is that some folks say the way the memos are phrased is wrong. Too candid, worded or spelled slightly wrong, varying from military protocol, and just plain weird text features (e.g. not using standard three letter month abbreviations, weird P.O. Box # 34567, and no letterhead, etc.)

Others disagree and say that an informal memo might be done thus.

Some "experts" say they feel sure it wasn't from a typewriter. So much for the value of the opinions of randomly press-annointed "experts" given the exactly mirror-image expert opinions.

Where does this leave me? With the jury very much still out.

Anonymous said...

Who cares? I care.

First I'd like to comment that I've since read more and it seems I was definitely wrong; it's not the case that a 1970s machine definitely could do this, and it looks overwhelmingly likely that it could not. I'd say the odds are now down to 10000:1 or lower that this is not a forgery. So I've changed my perspective -- the jury is in and guilty is the verdict.

Why should I, or anyone, care?

We should care because it's good to know how badly news organizations mess up even when they try to be careful.

It's good to know how quickly people leap to conclusions for political reasons.

And it's good to know that when people try to inject completely bogus evidence they can be taken to task. I wish there were simililarly concrete evidence re: Kerry and his medals to use in debunking the anti-Kerry Swift Boat Veterans' charges. That's assuming, as I do, that their charges are mainly bunk.

I think it's fine for the public to be reminded, again and again, what a hypocrite and privileged string-puller Bush is. Tainted evidence should be out of bounds for everyone, though.

Anonymous said...

Oh, and concerning "Why should we care?" You're a rabbi, right?

Somebody out there was bearing false witness.