Vectorizing words to use a Machine Learning algorithm

I'm testing an idea to vectorize any secuence of symbols into fixed size vector in Rⁿ using a hierarchical combination of echo state networks. The objective is to classify these sequences as vector (there is a lot of Machine Learning algorithms to use with fixed sized real vectors).

In particular, i'm testing this algorithm with english words, trying to classify them as nouns or adjectives. My dataset is here: http://www.ashley-bovan.co.uk/words/partsofspeech.html

Using a SVM to classify, i'm getting 9% of error, somebody please can point me to relevant papers or results to compare?

Thanks!