Is NLP relation extraction a hard topic?
At the Masters level you should not be looking for something easy; instead, you should be looking for something manageable. You should be looking for something useful, and something interesting. There are Masters theses that are on par with PhD theses.
Automatic summarization is by far the most cutting edge, and potentially useful. Relationship extraction may, or may not, be required for summarization. Word-sense disambiguation is at least thorny, if not a quagmire.
For details, see my quick and dirty webpages: