his blog about using a grammatical framework for generating texts. He also mentions Reasonator and indicates that its support for language is limited.
When you look at the code for generated texts, it is indeed a programmers job and not so much something a translator does. What it does is parses existing data and generates strings of text. The generated text for Marilyn Monroe for instance has it that she is an "actor".
"Lets talk man and horse" is a Dutch proverb and it is exactly what we will not have in generated texts. It is indeed grammar that we need to concentrate on. I am sure that many people will be EAGER to work on generated text using a grammatical framework.
Once it is shown to work in Reasonator, the next step is obvious; generating the texts for consumption in Wikipedias that want to share in the sum of available knowledge . This is best done by caching results because of the flexibility it offers. When this is seen as problematic; it is even easier to generate articles using bots.
Niklas, what category of topics tickles your fancy?