When a book is written in a way that you are better off reading research papers in the library, that is exactly what you should do. This book has no didactic value, and fails at defining the problem before attempting to explain it. The author does not attempt to put any examples together and the presented data is minimal to non-existent. Strikingly obtuse is the chapter on language learning. The chapter is inconsequential and self-defeating: "since language learning is one of the simplest cases of the kinds of learning that recurrent networks are tasked with, we must infer that learning in recurrent networks is difficult in general." Not surprisingly, this book is also poorly indexed.

To end in a positive note, I did like the cover!