Automatic Paper Generation

Some of you may have heard the news that some MIT students got their computer generated Computer Science paper, consisting of grammatically correct mumbo jumbo, accepted as a non-reviewed paper at some conference.

You can create your own paper with their code here. My friend Jai and I were discussing how fun it would be to add to this code the ability to feed it a selection of essay texts, say a collection of essays by Bhabha or Spivak to take two examples that come to mind, so that the nouns and verbs it chooses are roughly in correspondence to the frequency with which they appear in work by these scholars. It would then be interesting to see if readers can actually tell the difference.

If someone ever gets around to adding this feature, I think all you would have to do is 1) employ some kind of textual analysis algorithms that must be out there already (they use it for example in the processing of texts for speech recognition software or in linguistic studies doing word frequency analysis), 2) take the output of analysis and reconstruct the file “system_names.in” which is in the scigen code package you can download from there site. 3) Add some kind of frequency control to the code to make sure the output used nouns/verbs in a similar frequency to the originals.