Text generation
http://pad.constantvzw.org/p/text_generation
you can work on different levels:
* easy grammar algorithms
* recurrent neural networks, can be applied to text http://googleresearch.blogspot.be/2015/06/inceptionism-going-deeper-into-neural.html
learn to talk: http://karpathy.github.io/2015/05/21/rnn-effectiveness/
* take existing code and see how we can use them
ex : http://botpoet.com/
Possible interests:
- legal language = ritualistic / how to engage with it in machinic way
- write your own Shakespeare, experiment with types like in some literature
- automated twitter commenting in certain style (f.ex. antipatternalist comments to sexists)
- look at what computer does with language precisely
- reverse the spam
tools : pattern
possible to combine with text mining
algolit : experiment using markov chains (in an analog way, using dice)
<<<
fifty shades in C, automatic erotic texts
recursive neural networks
incremental means of traning neural networks to produce texts from totally random to increasingly well formed text.
the code validates how much text is text-like... how much it is shakespear-like.. so its
supervised vs. unsupervised learning
neural networks are actually quite old (Perceptron 1960's)
Markov chains, example of generative calvin & hobbes. (Calvin and Markov)
http://www.joshmillard.com/markov/calvin/
Using large-scale brain simulations for machine learning and A.I.
http://googleblog.blogspot.be/2012/06/using-large-scale-brain-simulations-for.html
https://static.googleusercontent.com/media/research.google.com/en//archive/unsupervised_icml2012.pdf
**********************************************
presentation (20th August)
notes/
- the ritualistic reality of language situations
- look at different text generators and to what they do
- - many times the language is so realistic (human) that is difficoult to see a difference (perhaps through the abuse of stereotypes)
- e.g.
- - a google algorithm that finds in a text also what is not there (different informations that are not direclty expressed) or (the algorithm learns how to speak in the stile of shakespeare - considering spaces in between words)
- - a computer learning to play a computer game (Marl/O - Machine learning from videogames - youtube)
uses:
a kind of 'commedia dell'arte': tipes of bots playing a commedia on twitter...
play with administration on a computer generator, playing with the regularity of the language
the tools are the starting point, you are not asked to know how to code! play around and explore these tools in different ways
- but the code is the 'grammar', you need to look at how these things are constructed!
- looking for the political implications of administrative-institutional language, ways to hack them
- looking vs reading - understanding parodies
- connection with data mining topic of the Training Common Sense track, related to text mining
Continues in [[text gen]]