English Interrogative-WHInfo rules

[nschneid]

These are 46 WH items found by Amir's old rules, but not the current rules: https://universal.grew.fr/?custom=65fb5e4b3fe93

Examples:

• What?
• What about brain damage?
• What's their personality like?
• What is it they say in yoga class?
• What are your thoughts about Rick Santorum’s views on gay rights?

[nschneid]

@WesScivetti

[nschneid]

I made some adjustments to the rules for WH-interrogatives, in particular adding rules to match WH words as predicates. Let's see if that fixes the discrepancy.

[amir-zeldes]

The data in UD was scripted using depedit, rather than grew, so there may be subtle differences somehow. I'm still not sure how to bake the grew rules into the GUM build bot easily, I wish there was a python library implementing the grew transformations...

[s-herrera]

There is a Python binding, that requires to install OCamL. If you want to use it, I recommend using it as it is done in the script of this project. The .grs files we use are not really valid for grew. I have tried to make this difference by adding the extension .ucxn.grs.

[amir-zeldes]

Thanks for pointing that out - I will have to think about how this could work, since it would add an entire additional programming language as a dependency for building GUM...

[nschneid]

If we're keeping this repo alive as a hub, maybe GUM can ship its pre-Cxn-annotated data to this repo, and the script calling Grew can automatically add the UCxn layer? I.e. treat UCxn as a service called by the GUM build bot (or other treebanks).

[amir-zeldes]

If it was implemented as a REST API or similar that would be an option (we need the build to be runnable on other people's machines just by cloning the git repo, but demanding an Internet connection for the build should be fine)