This is a great area for projects because there’s so much free data. Or you could create some of your own data. Projects can be more or less linguistically oriented.

At the simplest level, there’s just throwing something at a new corpus. Like say our latent Dirichlet clustering in some language or domain other than English news. Or even simpler, topic extraction over time.

Or you could create your own named entity corpus using our annotation tool and use it to build a named entity extractor.

For that matter, you could follow my instructions and build a Chinese search engine. Or Thai or Turkish for that matter.

Take a look through the ACL proceedings — there are lots of student papers and poster write-ups in there. I’d aim at getting something you could submit to the ACL student session as a concrete goal. Assuming you’re going to write it in English, that is.

At the most complex level, you could try to take on a whole new problem or develop a new model for an existing problem. That’s much harder, though.