Update United to be VERB, as per discussion at https://github.com/Uni…

[AngledLuffa]

Update United to be VERB, as per discussion at #480

[AngledLuffa]

@nschneid and/or @amir-zeldes

[amir-zeldes]

Yes for me, I think this is in line with tagging non-nominal components of names based on their underlying morphological categories. For me "united" is a verbal passive participle, transparently derived from a verbal lemma (even if used adjectivally in context, i.e. deprel amod). I think maybe @nschneid wanted to think about this some more?

[nschneid]

I think somebody should do a full review of the passive participle/adjective distinction in EWT to make it consistent.

[amir-zeldes]

I can extract a list of forms whose tags don't match between GUM and EWT pretty easily, but a manual review of all cases seems unfeasible, no? Another approach would be to use a model trained on either corpus to predict on the other and the do some adjudication, but this also feels like it would be a very big manual effort. If we limit it to the top K cases it might be possible, but if by full review you mean all cases I really don't think we can do it.

[nschneid]

By full review I didn't necessarily mean annotating every case but looking broadly at the range of lexical items. It seems like a good project for a student RA.

[AngledLuffa]

Looking at the first few, it's already getting kind of grim

For example, there's

respected - currently ADJ in EWT. VERB in GUM. sure, you can respect someone
involved - you can also involve someone. involved_ADJ in Baath activities in EWT vs a more egregious example involved_VERB efforts by ... in GUM

Even the IP and ICDC have abandoned the neighbourhood, and those are trained and armed ...

trained and armed - again, you can train or arm people
both ADJ in EWT
trained is VERB in GUM: trained to walk in a straight line
armed though is ADJ: armed_ADJ conflict. i suppose that isn't a person. there's also armed camp though

scared: both treat it as ADJ, although you can definitely scare people
I was a little scared_ADJ in GUM
and before anyone says transitive vs intransitive, it is also possible to say I do not scare easily

500 Iraqis are reported_VERB wounded_ADJ in EWT vs thousands of wounded_VERB soldiers in GUM

so I suppose that we'd want to really come up with some overall guidelines / tests for VERB vs ADJ. certainly that would be necessary before throwing models or undergrad backhoes at the problem. in the meantime, it's a little difficult that United is treated differently, seeing as how common it is in both treebanks

random aside, Google's explanation of "undergrad backhoe": Backhoe operator positions typically do not require a bachelor's degree. You can become a backhoe operator with a high school diploma or equivalent

It's true. I got a (small) construction vehicle stuck in mud before I had a high school diploma, and someone with a high school diploma used an bigger vehicle to pull it out.

[amir-zeldes]

a more egregious example involved_VERB efforts by

This is a finite verb (VBD), so that example is not the same construction

trained to walk in a straight line ... armed_ADJ conflict

These both look right to me. If you are trained to walk in a straight line, then you've been trained by someone, so it's a passive participle. And armed conflict is not a conflict that has been armed by someone (the combatants were probably armed by someone, which I would also consider to be passive, but the conflict is just armed). In other words, it fails the paraphrase test: an armed conflict is not a conflict that has been armed by someone.

I was a little scared_ADJ

This is the only one I'm hesitant about - of course, we can be scared by someone, but it's pretty odd to paraphrase that way. ENCOW puts the proportion of "scared by" to "scared of" at just under 10%. If 90% of cases express the source of the fear (if it's even expressed) without using a by-phrase, that makes me think that to a large degree this is already a lexicalized adjective. But if I had a sentence with "scared by", I would probably annotated it as passive all the same, so that would create a discrepancy.

500 Iraqis are reported_VERB wounded_ADJ

I would prefer to analyze cases like these as passives (somebody wounded them - it's morphologically and semantically transparent, easy to add a by-phrase, etc.)