OpenNLP

Hello,

I was checking the opennlp.uima project in your repository. That is a great initiative. Congratulations for the effort.

I notice that the code is not as customizable as I expected. For example, one should be able to configure the context generators, since each language should require a different one. The annotators are always using a default context generator. If you like we could talk about a good approach, also we could help you coding that.

Thank you,

William São Paulo University - USP

There is an ongoing discussion for opennlp 1.5 to place the context generators directly in the model to guarantee that the features generated for training are identical to those generated during tagging.

That would be nice.

For now I added a new parameter to the component descriptor specifying the name of a class that would be the Context Generator factory. The implementation of that factory would load the Context Generator, passing the model, dictionaries etc. With that I was able to use the same code for en-US and pt-BR.

Dou you have a target date for OpenNLP 1.5?

Thanks!

Well there are still lots of things to do for 1.5, I hope we can finish most of that over christmas, right now Tom and me are really busy.

If you wanna help out you are very welcome.

Jörn

Hi, Jörn,

I've been using OpenNLP since 2006. It would be a pleasure to help you with the changes for version 1.5. I can work on OpenNLP from 4 to 6 hours on Fridays, is it OK? If yes, you can assign an issue to me to start with. I already checked out the code from CVS and have both OpenNLP and Maxent compiling and passing the tests. A lot easier with Maven!

Hi, yeah, that would be really nice, I spoke with Tom and thats something you could help us with:

https://sourceforge.net/projects/opennlp/forums/forum/9944/topic/3463818

Jörn