Hello, yes I did a lot of testing to make sure that 1.7.0 and 1.7.1 are as good as any other OpenNLP release.
The OpenNLP eval tests use all the data we currently support to ensure it all works. If you have some specific corpora you need to be sure OpenNLP performs well on we can try to add them to the evaluation tests. Looks like to update we need to change the version and remove the maxent dependencies. The maxent part is not integrated into opennlp- tools. The existing models dating back to 1.5.0 are supported and should work just as good. In some rare cases we fixed obvious bugs which can affect the performance but this should be really minor. In case you have the data I would like to encourage you to retrain with 1.7.1, and otherwise you are also fine with the 1.5.x models. Also I can dedicate some time to ensure the corpora you are using are supported by the OpenNLP formats package. Your POS Tagger readme is mentioning GENIA, do you still use that? The POS Tagger these days can also be trained directly on the Penn Tree Bank, but you could also use OntoNotes. Should I sent you a patch to update your poms? I also noticed you are not using the opennlp-uima integration, is that not working for cTAKES? We could try to find a solution which works for you without duplicating our effort. Jörn On Wed, 2017-01-25 at 18:43 +0000, Finan, Sean wrote: > Hi Jorn, > > Great idea! Have you done any testing with the latest release? I > think that ctakes has 3 dependencies on opennlp libraries. We should > try to get our integration with that update (and others) solidified > for the upcoming release. The boston children's hospital nlp team > just gained a new member who will be devoting a significant amount of > time to a 3.2.3 (or 3.3.0) release of ctakes. He is James Masanz, > one of the original developers of ctakes, and we are extremely happy > to have him onboard. > > Thank you for the notification, > Sean > > > > > -----Original Message----- > From: Joern Kottmann [mailto:[email protected]] > Sent: Tuesday, January 24, 2017 4:52 AM > To: [email protected] > Subject: Update to OpenNLP 1.7.1 > > Hello all, > > the OpenNLP team just released a new version and I would like to > propose to update cTAKES to it. > > The release should be fully backward compatible. If you don't mind I > am happy to send you a patch to make it easier to migrate. > > The OpenNLP project also ships a UIMA integration which as far as I > can tell could be used by cTAKES directly. > > BR, > Jörn
