Linguistically-aware shell-code

It's not often that I read something involving the intersection of infosec and computational linguistics. But this post by Breck Baldwin popped up in my blogroll tonight:

http://lingpipe-blog.com/2010/03/09/language-model-generated-injection-attacks-cooldisturbing-lingpipe-application/

Some researchers at Johns Hopkins figured out how to generate shellcode that, while perhaps nonsensical, is reasonably close to grammatical English, enough so that shellcode filters won't detect it.

Did I miss this when it first came out? It looks like it was released in November.