Phrase model

[dan-zeman]

A new model of phrase/dependency structure that can handle coordination, prepositional phrases etc. properly during tree transformations and normalization. At the moment it is used only in HamleDT::Udep. It should replace the Treex::Core::Coordination and Treex::Core::Cloud classes in future.

[martinpopel]

Thanks Dan for a lot of work.

• Should I wait with merging this into master until TC::Coordination and TC::Cloud are replaced?
• Can you add some tests? (They will also serve as a usage info.)
• Is there a reason why TC::Phrase::depends_on() has a different name (and slower recursive implementation) than TC::Node::is_descendant_of()?
• And what about TC::Phrase::dependents() and TC::Node::get_descendants() vs. TC::Node::get_children()?

[dan-zeman]

• Replacing TC::Coordination and TC::Cloud is probably not going to happen fast and I'm afraid many Harmonize blocks will have to be adapted and tested. I reasonably trust what the current implementation does with Prague-to-UD conversion. That is, I can use this code to produce the next release of the UD treebanks and I'm not going to need the current code in the master branch any more. So I think it can be merged now, I will keep the phrase branch and create another pull request once the other half is done. But if you don't like it, I can as well live in the phrase branch and ignore the master branch until it is done :-)
• Tests: OK, I'll look into it.
• depends_on vs. is_descendant_of: no, that's random. I will rename it (and look into the implementation too).
• TC::Phrase::dependents() is different from TC::Node::get_descendants() semantically, not just in name. But I could add an alias TC::Phrase::get_children() for TC::Phrase::children(). (And then perhaps for all the other similar methods, too.)