Cryptography Reference
In-Depth Information
How to Use It?
Thecodeforthemimicsystemisavailablefromthe
author.. Or you can just go to
spammimic.com
if you want your
message to look like spam.
Further Work
There are a number of avenues to pursue in this
arena. A theory that gives a stronger estimates of the brute force
necessary to recognize a language would be nice. It would be
good to have a strong estimate of just how many strings from
a language must be uncovered before someone can begin to
make sense of it. If someone could program the entropy es-
timates from [Way95a] or come up with better ones, then we
could experiment with them to see howwell they assess the dif-
ficulty of attack.
It would also be nice to have an automatic way of scanning texts
and creating a grammar that could be used by the system here.
There are many basic constructs from language that are used
again and again. If something could be distilled from the raw
feed on the Net, then it could be pushed directly into a program
thatcouldsendoutmessages. Thiscouldleadtotrulyauto-
mated broadcast systems. One part would scan newsgroups or
the net for source text that could lead to grammars. The other
would broadcast messages using them. I imagine that it could
lead to some truly bizarre AI experiences. You could set up two
machines that babble to each other, mimicking the Net but re-
ally exchanging valuable information.
Further Reading
•
A number of papers from Purdue's large group extend
these grammar techniques with a comprehensive database
of synonyms for words. Their work suggests that the
most ambiguous words where possible because more spe-
cific words may not be easily swapped into a sentence.
[ARC
+
01, ARH
+
02, TTA06, AMRN00, ARH
+
03, TTA06]
•
CuneytM.Taskiran,UmutTopkara,MercanTopkaraand
Edward J. Delp developed a tool for finding text generated
by methods like the grammar-based tool described in this
chapter. They compare the statistics gathered from ana-
lyzing when words are found adjacent to each other with
their general occurrence in the general language and use
it to train statistical classifiers. [TTTD06]
•
The BLEU score, a rough statistical comparison of the
phrase length, measures the quality of automated transla-
