[ninjin]

Visualising multi-dimensional data is indeed, tricky. The best tool I have found so far for word representations similar to LSA has been t-SNE:

http://homepage.tudelft.nl/19j49/t-SNE.html

In some cases what I get isn't all that much better than simply using PCA, but overall t-SNE is superior. Although t-SNE is dreadfully slow... Below is a link to an implementation used for text and I can highly recommend the original paper on t-SNE:

https://github.com/turian/textSNE