Use of 'det' (determiners)

[nschneid]

Looking at the dev set, there are some suspicious uses of the det relation:

29  מ  _   ADP ADP _   30  det _   _
4 אותו _ PRON PRON Case=Acc|Gender=Masc|Number=Sing|Person=3|PronType=Prs 5 det _ _
11  כנראה  _   ADV ADV _   12  det _   _
30  גם    _   ADV ADV _   31  det _   _

Also, what is the principle for the division between det and det:quant for numbers/quantifier words? I see שליש and משהו are det:quant, but כל, אחד, כמה, etc. are just det.

[yoavg]

The suspicious det cases are most likely mistakes in the underlying corpus (although גם may not be, I need to see this in context).

Re det:quant, we currently mark it only for partitives in construction similar to "part of the problem", "quarter of the group", etc. So אחד would be marked as det:quant for אחד מהילדים but not for ילד אחד and not for אחד הילדים. This is not ideal, but this is just how it is. :(
This may be improved in future UD versions, but we do not have the capacity to fix this in the current release.

[nschneid]

Re det:quant, we currently mark it only for partitives in construction similar to "part of the problem", "quarter of the group", etc.

The English UD conventions make no distinction between partitives and ordinary PPs: e.g., examples with "some of". Then, for enhanced++ UD, "some of (the)" is converted to an MWE, and the "some" is attached to the heavy noun with a det:qmod edge to signal the partitive relation. (I'm not wild about including the article in the MWE, but we can leave that argument for another day.)

In Hebrew, is the information that it's partitive coming from the source treebanks? If so, it makes sense to record it somehow in the basic UD parse. What about matching the English structure but refining the relations to nmod:partitive and case:partitive?