Dimensionality Reduction for non numerical data?

If I have datasets with lot of features, some of them are non numerical (symbolic), is it possible to apply Dimensionality Reduction like with PCA and/or using Feature Extraction like RBM ?