"One another" and NumForm

[amir-zeldes]

The current analysis of "one another" is:

one/CD/NUM <-det- another

This is fine, but I've just added NumForm to GUM, made "one" be NumForm=Word and I get a validation error:

Feature NumForm is not permitted with UPOS DET in language [en].

Possible solutions:

1. Allow NumForm on DET - I'm not crazy about this because this warning actually caught a legitimate error elsewhere in the corpus
2. Not use NumForm on this case - a bit annoying, since it's as much "one" as any other use of "one"
3. Not tag "one" as CD - the PRP generic "one" doesn't get NumForm, since it's not considered a number, so this would be consistent with other uses of pronominal "one"
4. Add some sort of mechanism exempting specific multiword expressions or subgraphs in the validator

I think I like 3 best, but any opinions? Adding @nschneid and @dan-zeman since it relates to the validator.

[dan-zeman]

I like 3, too. But I find 2 as appealing. I don't think you would ever use Arabic or Roman numerals for "one" in "one another" (now disregarding social media where "4U" is used instead of "for you"), would you? In that case it is not "as much one as any other use of one".

As for 4, the validator can (since yesterday) check language-UPOS-feature-value combinations but not more than that. I would like to add more flexibility in defining language-specific rules in the future but it won't be soon.

[amir-zeldes]

Thanks Dan - I think if we say that this word is xpos=CD/upos=NUM, then it should have NumForm like all other cases. If you wanted to know which uses of the number "one" are always spelled out in English, you could then discover that this is one of them. If by contrast we don't really think this is a number, then we should just retag it (but then I would like that to apply to EWT as well, which is why I open the issue in this repo).

For reference, PTB has this "one" four times as CD, and twice as NN (I guess by analogy to "anyone"?)

[dan-zeman]

Yes. I think that it should either be tagged NUM CD NumType=Card (and allow all values of NumForm), or it should be a pronoun with all consequences for features and dependencies.

[amir-zeldes]

Yes, I'm fine with either of these, but something needs to be changed either way:

• If we choose NUM: deprel needs to be changed to nummod for one in "one another"
• If we choose PRON: POS tags need to be changed for the same cases

I can do either, but would like consensus with EWT!

[amir-zeldes]

Actually on second thought there is yet another option that I almost find better - change the deprel and dominance to fixed, maybe even with reflexive pronoun features on the first token. What do you think?

[dan-zeman]

fixed does not sound bad but you (and the EWT team) still have to decide on the tags. I think I lean towards PRON now, but not reflexive.

[amir-zeldes]

If I were doing this all over, I would probably choose PRON, but since PTB has it as CD (at least as a majority), that might motivate us to stick with NUM. Either way, if it's fixed, the individual constituent POS tags would not be very important IMO.

[nschneid]

I would have assumed that "one" in "one another" is PRON. Note that there are other contexts where "one" can be pronominal meaning "someone", which in UD is PRON (versus NN in PTB). I.e. UD takes a broader view of pronouns than PTB.

As for the deprel, fixed sounds good to me.

Related: "each other"

[amir-zeldes]

OK, I went with fixed for the now released GUM V7 - thanks everyone! I think it would be great if EWT adopts this analysis as well. For now I left the POS alone, since PTB already has CD and I don't like messing with existing standards (who knows how many people have already copied that). I also thought briefly what would happen if we have something like this Dickinson poem:

The Brain—is wider than the Sky—
For—put them side by side—
**The one the other** will contain
...

In this case it would be inconvenient for "one" to be PRON, since it has an article, but this is really the transparent source of the reflexive "one another". So maybe CD is not such a bad idea to begin with...

Leaving this open as a reminder to consider using fixed for EWT as well!

[nschneid]

Oh boy, yes, "one" can have noun-like modifiers ("Do you want the tall one on the left or the short one on the right?") in this anaphoric sense. CGEL calls this a pro-nominal, as opposed to the personal pronoun where it refers to a generic individual, similar to "someone" ("One is forced to conclude that...").

But on the particular topic of this thread, CGEL (p. 428) considers "one another" and "each other" to be multiword reciprocal pronouns:

So this supports fixed, and I suppose it wouldn't hurt to tag both the words as PRON.

[nschneid]

Documented for fixed in UniversalDependencies/docs@ff6acd6