We have a bit of experience with Stanford NLP but are wondering if others have surpassed it.
We're looking to read email documents to, among other things, identify implicit tasks (and their attributes - owner, priority, date).
I've built my own full stack http://www.alchemyapi.com/ using UIMA[1] and cleartk as a base though.
The option of the different JVM scale out frameworks like akka,hadoop,storm etc have been amazing.
I would recommend NLTK for simpler tasks though.
We're working on an OSS Watson over at[3]:
that uses those components though. It's fairly easy to use different kinds of machine learning models for what you're looking for. Shoot me an email (in my profile) if you have any specific questions.
[1] http://uima.apache.org/ [2] http://cleartk.googlecode.com/ [3] https://github.com/SolrSherlock/