7.1 Using Grammar for Mimicry

Chapter 6 shows how to hide data and turn it into something that mimics the statistical patterns of a file. If you wanted a piece of text to sound like the New York Times, for instance, you could feed in a large amount of source material from the paper and gather statistical patterns that would make it possible to mimic its output. Ideally, such a function would be a strong technique that could hide data from automatic scanning programs that might use statistical patterns to identify data.

The output of these Huffman-based methods could certainly fool any machine examining data looking for suspicious patterns. The letters would conform to the expected distribution; "e"s would be common, "z"s would be uncommon. If either second- or third-order text was used, then "u"s would follow "q"s and everything would seem to make sense to a computer that was merely checking statistics.

These statistical mimic functions wouldn't fool anyone looking at the grammar. First- or second-order mimicry like that found in Section 6.1 looks incomprehensible. Words start to appear in third-or fourth-order text, but they rarely fall into the basic grammatical structure. Even a wayward grammar checker could flag these a mile away. This chapter describes how to create mimicry that will be grammatically correct and make perfect sense to a human. The algorithms are based upon some of the foundational work done in linguistics that now buttresses much of computer science. The net result is something that...