Description

We recently upgraded Lucene's dependency on OpenNLP to 2.5. This upgrade offers a new part-of-speech tagging model that works across more languages. The update maintained backward compatibility with the old Penn model by hardcoding it in the token filter.

We should expose the UD model as an option.

I'd like to work on this, so please assign it to me.

@mawiesne, you're the OpenNLP expert, please let me know about potential pitfalls. I don't know OpenNLP, but this seems fun.