How to annotate "(to be) X years old" / "(att vara) X år gammal"... ?

[harisont]

I noticed that the three Swedish treebanks do three different things with the expression "(att vara) X år gammal":

• Swedish-Talbanken: år <-obl- gammal
• Swedish-LinES: år <-nmod- gammal
• Swedish-PUD: år -amod-> gammal

Since two of these treebanks have a parellel English version, I also checked:

• English-LinES: years <-compound- old
• English-PUD: years <-obl:unmarked- old (like English-EWT)

There might even be other (PUD) languages that do something similar and are worth checking.

So my question is, what to do with "(to be) X years old", "(att vara) X år gammal" and similar?

I'm pretty sure that år -amod-> gammal is wrong and that "gammal" should be the head, but I'm not sure about the dependency label.

[jnivre]

Thanks for noting this inconsistency. In my view, "gammal" should definitely be the head, so the PUD annotation is probably just a parsing error that hasn't been corrected. (Swedish-PUD was parsed automatically, then manually corrected.) And the Talbanken and LinES analysis agree on this, but disagree on the label. If we accept that "gammal" is not a noun phrase in this context, then the label should be "obl", not "nmod". This also agrees with the English-PUD analysis. (A compound analysis does not look convincing to me.)

[amir-zeldes]

Yup, it's an adverbially used NP, but not an adverb, and it modifies an adjective, so obl (and English uses the subtype :unmarked to distinguish such adverbial obliques from prepositional phrases).

[LarsAhrenberg]

Will fix LinES. Thanks for noting.

[harisont]

Thanks for the answer! obl makes sense to me too.
If that helps, I'm happy to open a pull request on Swedish-PUD.

[jnivre]

I have fixed the PUD error now, but thanks anyway. I will close the issue as well, since we seem to all agree on what should be done.