The blog post The Unreasonable Effectiveness of Recurrent Neural Networks (http://karpathy.github.io/2015/05/21/rnn-effectiveness/ ) describes a fascinating set of examples of RNN character-level language models, including the following:

• Shakespeare text generation similar to this example
• Wikipedia text generation similar to this example, but based on different training text
• Algebraic geometry (LaTex) text generation similar to this example, but based on different training text
• Linux source code text generation similar to this example, but based on different training text
• Baby names text generation similar to this example, but based on different training text