Machine translation cracks 18th century occult cipher

I just read the original paper (<a href="http://aclweb.org/anthology-new/W/W11/W11-1202.pdf" rel="nofollow">http://aclweb.org/anthology-new/W/W11/W11-1202.pdf</a>).<p>As far as I can tell, the article by theregister is complete nonsense: the researchers suspected a simple substitution cipher and used statistical analysis of the distribution of (co-)occurrences of the text symbols. Once they conjectured that the source language was German, they were able to use letter frequency information from modern German texts to derive likely mappings between the symbols and letters from the German alphabet.<p>This has nothing to do with machine translation. MT does not typically operate on the level of letters. The only time the researchers mention machine translation in their paper is on page 6, where they piped the partly deciphered text into a German-to-English translation software to see where their mapping was incorrect.<p>The remarkable fact is that apparently none of the researches speaks German, so they used such software to check if it could make sense of the sequences of letters deciphered so far. They might as well have used an old-fashioned dictionary to achieve the same result. On page 7 they talk about finally talking to native speakers of German (quite late in the endavour I think), to help them with the remaining problems.<p>The register article gives the impression that the researchers sort of pasted a transcripted version of the original document into an MT system which then magically produced a "translation" into German autonomously. That is not at all what happened.

The Register article links to three other sources. Here they are in clicky form.<p>(<a href="http://www.wired.com/wiredscience/2011/10/copiale-cipher-crack" rel="nofollow">http://www.wired.com/wiredscience/2011/10/copiale-cipher-cra...</a>)<p>(<a href="http://www.eurekalert.org/pub_releases/2011-10/uosc-csc102411.php" rel="nofollow">http://www.eurekalert.org/pub_releases/2011-10/uosc-csc10241...</a>) &#62;<i>To break the Copiale Cipher, Knight and colleagues Beáta Megyesi and Christiane Schaefer of Uppsala University in Sweden tracked down the original manuscript, which was found in the East Berlin Academy after the Cold War and is now in a private collection. They then transcribed a machine-readable version of the text, using a computer program created by Knight to help quantify the co-occurrences of certain symbols and other patterns.</i><p>(<a href="http://stp.lingfil.uu.se/%7Ebea/copiale/" rel="nofollow">http://stp.lingfil.uu.se/%7Ebea/copiale/</a>) &#62;<i>The “Copiale Cipher” is a 105 pages manuscript containing all in all around 75 000 characters. Beautifully bound in green and gold brocade paper, written on high quality paper with two different watermarks, the manuscript can be dated back to 1760-1780. Apart from what is obviously an owner's mark (“Philipp 1866”) and a note in the end of the last page (“Copiales 3”), the manuscript is completely encoded. The cipher employed consists of 90 different characters, comprising all from Roman and Greek letters, to diacritics and abstract symbols. Catchwords (preview fragments) of one to three or four characters are written at the bottom of left–hand pages.</i>

A note on the content of the decrypted book (105 pages), which according to the Register,<p><i>"[...] has been revealed as the rituals and political thoughts of a German secret society, with a strange fascination for eye surgery and ophthalmology."</i><p>However, the WikiPedia entry [1] makes this ritual sound more like a fraternity/student joke:<p><i>"[...] an initiation ritual in which the candidate is asked to read a blank piece of paper, and on confessing inability to do so, is given eyeglasses and asked to try again, and then again after washing the eyes with a cloth, followed by an "operation" in which a single eyebrow hair is plucked."</i><p>[1] <a href="http://en.wikipedia.org/wiki/Copiale_Cipher" rel="nofollow">http://en.wikipedia.org/wiki/Copiale_Cipher</a>