George Flanagin

Ishtar

For the past few decades I have been fascinated with nonsense speech. It probably started when I first read Alice's Adventures in Wonderland, or perhaps when my parents looked at me and said 'I do not understand you.' I have forgotten when I first encountered Lorem Ipsum, but it got me to consider how to have nonsense text based on something other than degenerate faux-Latin. So I wrote the original version in C++. (The source code has been lost, and mercifully so.)

Other than being one of the least successful movies, Ishtar is the name of my text travesty generator. Ishtar transforms text into non-sensical but similar text that resembles the input to any degree desired. Garbage in; trash talk out.

Along came Python, and I needed to learn a lot about it in a hurry. So I did what all experienced programmers do: I rewrote a program whose design was already well known to me, and Ishtar 2.0 was the result. At times, the output of Ishtar 2.0 seems rather insightful, and Donald Trump continues to provide source material. His fascinatingly small vocabulary intrigues me, and Ishtar's algorithm is ideally suited to drive his teleprompter. For example one of those is Ishtar's regurgitation and one is Don's original.

1. Well — well, I don't like it, and maybe it will take longer than Dwight D. Eisenhower, who helped win World War II. And we spent probably, if you add it up now, $4 trillion, maybe more than just — in my opinion is that they wanted to do, but for traveling and for other things, you know, I felt, again, knowing the people said, 'pay.'
2. Well — well, I don't know, first call or second call, they will be a very powerful call, that's going to be a great relationship. They'll be great allies, they always have been. And I think zero will change on that score. There has never been a better ally, and I think nothing will change on that score.

Not bad, eh?

Ishtar 3.0

As much as possible, I am trying to do a little serious work, which is why this repo is private. If you are interested in doing something with Ishtar, please send your ssh key to me+ishtar@georgeflanagin.com, and I will see about putting your key on the ring.

Development at this time is going toward understanding sentence patterns and parts of speech analysis by writing code that exploits the NLTK library. Ishtar 3.0 currently understands parts of speech, vocabulary, stems, and function v. content words. Text can be 'fingerprinted' by applying Fast Fourier Transforms and Discrete Cosine Transforms to the distributions of parts of speech, and pandas has come in very handy.

The development of Ishtar 3.0 is taking place in a private repo so that I don't make a complete fool of myself, and more importantly so that no one accidentally downloads a program that will gobble up memory, corrupt the disc, or who knows what else. It runs very well on University of Richmond's 36 Xeon machine with 128GB of RAM and SSDs.

== 30 ==