Integrate new field to add unlimited morphological or syntactic features

[Cocophotos]

Hi André,

As Djamé told you in a mail this weekend, I've made some changes on the TurboSemanticParser (great work, BTW, I like it a lot) to cope with our syntactic features.

I've just added a morph features field separated by a pipe and added some new feature templates and new features in SemanticFeatureTemplates.h SemanticFeatures.cpp and of course, I've changed SemanticReader.cpp a bit.

Normally all my changes are tagged with my fullname, so it's pretty easy to track them and to reverse them if needed.

I've made my changes in the 2.1 version of your parser, but for sake of completeness, I'm sending you a pull request on the 2.3 version. I know that this version compiles, but I let git and Github take care of the merging (normally it's fine).

Thanks again for this great parser.

Regards,

Corentin

[andre-martins]

Thanks! I will merge this code once I get some time. So the only change is a new column in the CONLL files with morpho-syntactic information? (Formatted the same way as in the dependency parsing CONLL-X files?)

I will need to create a flag --read_morphosyntactic_information which is false by default, otherwise this change will break the semantic parser on the SemEval files... Right?

André

2015-03-30 10:25 GMT+01:00 Corentin Ribeyre [email protected]:

Hi André,

As Djamé told you in a mail this weekend, I've made some changes on the TurboSemanticParser (great work, BTW, I like it a lot) to cope with our syntactic features.

I've just added a morph features field separated by a pipe and added some new feature templates and new features in SemanticFeatureTemplates.h SemanticFeatures.cpp and of course, I've changed SemanticReader.cpp a bit.

Normally all my changes are tagged with my fullname, so it's pretty easy to track them and to reverse them if needed.

I've made my changes in the 2.1 version of your parser, but for sake of completeness, I'm sending you a pull request on the 2.3 version. I know that this version compiles, but I let git and Github take care of the merging (normally it's fine).

Thanks again for this great parser.

Regards,

Corentin

You can view, comment on, or merge this pull request online at:

https://github.com/andre-martins/TurboParser/pull/11 Commit Summary

• Changed to add morphological features (used for syntactic features)
• Update version

File Changes

• M install_deps.sh https://github.com/andre-martins/TurboParser/pull/11/files#diff-0 (2)
• M semeval2014_data/scripts/augment_with_companion_data.py https://github.com/andre-martins/TurboParser/pull/11/files#diff-1 (1)
• M src/semantic_parser/SemanticFeatureTemplates.h https://github.com/andre-martins/TurboParser/pull/11/files#diff-2 (10)
• M src/semantic_parser/SemanticFeatures.cpp https://github.com/andre-martins/TurboParser/pull/11/files#diff-3 (2974)
• M src/semantic_parser/SemanticReader.cpp https://github.com/andre-martins/TurboParser/pull/11/files#diff-4 (330)

Patch Links:

• https://github.com/andre-martins/TurboParser/pull/11.patch
• https://github.com/andre-martins/TurboParser/pull/11.diff

— Reply to this email directly or view it on GitHub https://github.com/andre-martins/TurboParser/pull/11.

[Cocophotos]

Hi,

The new features are taking into account only if the syntactic flags is set to true.

I also modified the scripts in semeval2014_data to output the augmented version with the new field, so that it's transparent. Of course, it was before your release of the semeval2015_data.

So, yeah a new flag would be better. I can take care of it, if you want. After all, I was the one making that changes and I don't want you to have extra work for it :-)

Finally, yes I only needed to add our features set into your parser, so using a field such as the morph field in a CoNLL-X format gave me enough to test and remove features without making hard changes to the parser (I didn't have time to enter completely into the code base, I've tried doing more deep changes, but I recall that they broke the parser with a segfault, when the external pruner was trained... So I let go for now).

Regards,

[dseddah]

Hi Andre, hi Corentin, I know it's an old thread but I need to reproduce some results. So have these changes have been merged at the end?

Best, Djamé