Ellipsis in coordination below the word level (with hyphens) in Swedish

[AleksandrsBerdicevskis]

Not sure if the title is clear enough (is there perhaps a more precise term?), but what I am interested in is how to treat the Swedish sentence Man får olika perspektiv på hur andra samhällen fungerar levnads- och beteendemässigt.
From the existing treebanks, @harisont and I gather that levnads should be the head of the coordinated constuction and it should be treated as a form of the noun levnad (and not adverb levnadsmässigt).

Two questions:

1. Why not add "Case=Gen" to the features?
2. In most such cases, the conjuncts belong to the same POS, but here, they don't, which leads to a NOUN under advmod, which is invalid. How should we deal with that? (Possible systematic solutions: always choose the full word as the head, or always assign the full word's POS+features to all the conjuncts, or allow advmod+NOUN. Possible ad hoc quick fix: I guess obl instead of advmod is not entirely unthinkable?).

[dan-zeman]

If the first conjunct is NOUN, the incoming relation has to be obl rather than advmod. (If you also do enhanced dependencies, you could still have incoming enhanced advmod to beteendemäsigt.)

Treating the first conjunct as a form of adverb levnadsmässigt is definitely an option (and I guess I might even prefer it if I were doing it) but if the other Swedish treebanks already use the other option, it is desirable that the new one does not diverge :-)

[jnivre]

I would have to check what the current practice is. Most truncated compounds are noun-noun compounds, where the problem does not arise. But I think there is some merit to regarding "levnads-" as a form of "levnadsmässigt" to circumvent the problem. One should then probably also assign it the lemma "levnadsmässigt".

The linguistically correct analysis (I think) is that the actual coordination is "levnads- och beteende(-)", which then forms a compound with "mässigt", but this is currently beyond the expressive capacity of UD.

[LarsAhrenberg]

I support @jnivre's analysis.

[AleksandrsBerdicevskis]

I support it, too (this is exactly what Eukalyptus does).

[jnivre]

Excellent! Issue resolved?

[AleksandrsBerdicevskis]

If you and Lars are willing to modify the existing treebanks, then yes!

[jnivre]

At least I am, but I can't promise that it will happen before the next data freeze. I have started on the list of things we previously agreed on, so we will see how far I get.

[LarsAhrenberg]

Same here.