bugfix to default normalization in dictionary tagger

[spyysalo]

NER::SentenceTagger::set_normalize_type() was being called prior to looking up the default normalization from the dict data.

[spyysalo]

@priancho : please note that you also got 0b04aa0c6079955ff47b5c285dbc4b585a744adb, which is unrelated to the original pull request description. Sorry about that. That latter commit just adds your code to expose the CRFsuite label bias parameter on the CLI (and your related comments), but you might want to keep this out (or commented out) in master as it adds a dependency on the latest (upcoming?) version of CRFsuite.

[priancho]

Hi, I think that it is not necessary to undo this commit because there is no explanation about the label bias function in src/nersuite/main.cpp :-) and I can add it after Okazaki-san releases a new version of CRFSuite.

[spyysalo]

OK, thanks for the info! On a related note, I'd like to release a tagger at some point that uses the label bias feature of CRFsuite. What instructions should I give to users about installation, i.e. where can they get the CRFsuite version that implements this?

[priancho]

I will ask Okazaki-san whether he has time to incorporate my codes into a new version of CRFsuite. But it may take quite time because recently Okazaki-san is too busy to manage his softwares.

If you need to release it soon before he updates CRFSuite, I think that you can release the modified CRFSuite (that you are currently using) because CRFSuite is based on the modified BSD license. And, of course, it will be nice of you, if you notify him before doing it :-)

I will reply as soon as I got a reply from him.

[priancho]

Okazaki-san said he will release a new version in this weekend :-)

[spyysalo]

Great, thanks to you both!

[aehank]

I'm having problems compiling NERsuite as this error keeps coming up: "nersuite.cpp:154: error: ‘class CRFSuite::Tagger’ has no member named ‘set_bias’"

I looked it up and there is no set_bias member in the CRFSuite class Tagger. Do you know what the fix for this is?

[spyysalo]

@aehank : this relates to the label bias support in CRFsuite. It appear this feature may not have made it into a released in a stable version yet, but @priancho's version https://github.com/priancho/crfsuite includes support. I'd suggest to use this version to compile against recent versions of NERsuite.

@priancho : do you have any information about the CRFsuite release schedule?

[aehank]

@spyysalo Thank you very much! I will give that one a try.

[priancho]

@spyysalo I am sorry about the delay of the release of a new CRFsuite. The weight bias functionality in the forked CRFsuite works correctly, but it is not well integrated into the CRFsuite code. For official release, I need to revise it following the CRFsuite system architecture. So, the integration of this functionality into the official CRFsuite release may not be soon :-0