Sunday, September 12, 2004

consistent copying errors

Neil Uchitel has another strong case against the authenticity of the Bush memo. The only remaining difficulty for those who think it is a fraud is to explain the consistent differences between characters in the memos vs. those in a Word document. Some of the letters in the forged documents seem to be off the baseline a bit and some of the letters appear to be different in the memo than in a Word document. This is easy to explain with mechanical errors on a mechanical typewriter, hard to explain on a laser printer.

I'm only concerned with consistent morphing of the underlying letters. Clearly, the process of copying or faxing is going to produce artifacts, but the question is, can they produce artifacts that are consistent within a figure (seeming to move the whole figure) or between different figures of the same type (changing the shape the same way for each one)? I haven't seen the typographic experts like Neil comment on this, so I thought I'd take a stab at it.

Clearly it is possible for one line to move up or down (especially a horizontal line), depending on where the picture elements are in the printer (or fax), but is it possible for an entire figure to move up or down? I can imagine mechanisms that would allow it: To move up, the figure would have to have real top and bottom lines that both lie about the same distance below a scan line in the device. In the copying process, the scan line just above the correct position would be darkened to represent the figure, and the figure would seem higher.

I tried an experiment where I printed out four lines of "e"s in Word then photocopied the result for four generations. There were persistent changes in the letters within one line, but no movement up or down. In retrospect, I don't see any mechanism that would move the same character on the same line different amounts.

One commenter on Neil's post claims that the capital "M" and "A" and the "8" are clearly different. I went into word, set the point size to sixteen (to match the size in the pdf on the CBS site) and wrote several lines of each letter. Then I printed it out and compared the result on paper to the pdf on screen. I wasn't sure whether there was a real persistent difference or not because the characters in the memo are so fuzzy, but it looked like the "A" in the memo was a bit more symmetrical and the "M" a bit less symmetrical than the printout. Also, in the pdf, the eights have a smaller top loop than bottom loop, but there are two eights right together where one has a much smaller bottom loop than the other.

It's important to realize that Times New Roman at sixteen point is not just a scaled version of Times New Roman at twelve point. If the document was produced at twelve point and magnified by the scanning process, that could change the results. One could try this experiment with a bunch of different points sizes to see if there is an exact match, but I didn't do that. Instead, I just photocopied my printout four times to see if persistent errors were possible. They are. The fourth generation of the printout had some fairly consistent, but very slight differences from the original: some lines were thinner and that made some angle appear slightly larger.

I made two more copies of the fourth generation at a darker setting, then one copy scaled 129% and one copy scaled 78%. This 8th generation shows some persistent differences, though not the same as the memos. The bottom loop of the eight is distinctly triangular, the "A" has thicker serifs at the base lines, the cross bar o the "A" is slanted up to the right, and the "M" is more symmetrical.

So although I didn't reproduce the features of the memo, I convinced myself that copying can produce consistent morphing of letter shapes. (Especially within one line, as the similar experiment with lowercase "e" showed).

Translation from the baseline is still unresolved to my mind.

No comments: