search

UniversalDependencies / UD_English-EWT

English data
Creative Commons Attribution Share Alike 4.0 International
197 stars 41 forks source link

"other than" meaning 'apart from' #431

Open nschneid opened 9 months ago

nschneid commented 9 months ago

https://universal.grew.fr/?custom=650f8f5e94c4a

nschneid commented 9 months ago
  1. I have other ideas than this one
  2. I have ideas other than this one
  3. Other than this one, I have more ideas

Arguably, "other" is an ADJ and functions as amod in (1) and (2). But it's not really attributive in (3)—it functions more like case (a deadjectival connective).

(The part after "than" may also be clausal: "I have other ideas than repaving the driveway", etc.)

Related: "apart from", "except for"—but "apart" and "except" aren't ADJs

Better analogies: "separate/distinct from"? Or "more than": "More than for the money, he did it for the prestige."

amir-zeldes commented 9 months ago

I agree it's just a potentially extraposed ADJ/amod in 1.-2. The preposed version is adverbial as a whole, and ON mixes JJ and RB here. I would be open to having other/RB in 3., but I think it's good to keep a consistent structure for its dependents, so I would say it's still the head (not case) in 3., and takes an obl/advcl child with "than" as case/mark.

nschneid commented 9 months ago

I dunno, I think it really behaves like case in (3) at least. Its function is to mark a sort of concessive or setting-aside adjunct whose main content is "this one". It lacks the 'additional' meaning present in (1). (Note that for (3) to be a paraphrase I had to add "more". "Other than this one, I have ideas" would not quite work.) In terms of how it marks a nominal, can't think of a salient difference from "except for" (it's just that "except" can't be an adjective). "Other" cannot appear on its own, as a typical advmod in a clause might. The combination "other" + "than" is entrenched and thus arguably fixed along the lines of "prior to". (We never resolved whether "except for" would be better as fixed: see UniversalDependencies/docs#795.)

If you don't like fixed, another option is to allow ADJs as case like we do for VERBs, and have it be sister to "than".

I hear you that there's a cline of grammaticalization here (rather than a sharp reanalysis) and thus it would be nice to keep the structure similar, but I think the content-heads policy prevents us from doing that as far as items grammaticalizing as case/mark are concerned. (By contrast, in PTB or CGEL, prepositions would be heads as well so the structure wouldn't have to change.) IOW, the content-heads policy prevents continuity of structure as items progress from content to functional.

amir-zeldes commented 9 months ago

the content-heads policy prevents us from doing that

I think that presupposes that "other" is not a content head :) I understand what you are saying about grammaticalization messing with consistency, but I don't think it's necessary in this case. "Other than this one" really does mean that there are other ideas (it is pretty literally connected to the meaning of 'other'), and it has the regular argstr of other + than. If you treat it as functional, would you also treat "similarly to" as case? I think it's pretty similar, and we could decide that it's a complex case/mark, but I don't see what we would gain by doing that. It's meaning, like "other", is pretty close to a preposition, since they are both relational adjectives.

nschneid commented 9 months ago

"Similarly" can occur on its own as an adjunct in a clause.

"Other (than)" can't, and also can't be premodified when it's marking an adjunct:

I do think the meaning is a bit different (albeit related). The adjunct use doesn't have to occur with an explicit reference to something of which there are multiple:

amir-zeldes commented 9 months ago

"Other (than)" can't

I think that's just because "other" doesn't take "ly" morphologically. It's corresponding adverbial form is "otherwise", which we also treat as advmod and not mark:

So the form "similar" corresponds to "other" in not being used alone in this position (?? similar, bats have fur). The 'multiple reference' thing I think just has to do with the meaning of 'other' compared to 'similar', and I don't think that should change the morphosyntactic classification - they are both relational, and both express a relatum argument.

nschneid commented 9 months ago

How would you parse

Surely "Similar" is not an amod?

nschneid commented 9 months ago

Also, "prior to" and "such as" are in the fixed list. Is "other than" different in some crucial way?

amir-zeldes commented 9 months ago

Similar to your distaste for coffee, I can't stand cilantro.

I think you'd need to decide whether you consider it a reduced clause (advcl) or just a short-form adverb and treat it like "similarly", but either way I wouldn't want it to be a preposition. It is pretty transparently the usual meaning and argstr of "similar".

"prior to" and "such as" are in the fixed list

If I had made that list, prior to would not be in it - it's just an adverb with an argument. "such as" is quite far from it's original meaning and distro (people don't really use "as" by itself for comparison of NP), so I would support that being fixed. More generally though, I think we should keep fixed as small as possible, it's already overly lax for my taste.

nschneid commented 9 months ago

I agree that "similar to" is a weaker case for fixed (for one I think it can be modified: "Very similar to your distaste for coffee,..."), but I'm puzzled by how it seems to behave like "similarly" here.

There's also things like "Reminiscent of" and "Contrary to". Such expressions are sometimes "detached" from a particular predicand and act more like discourse connectives.

I thought "Mindful/conscious/cognizant of" might have similar properties, acting as a speaker-oriented adverbial, but glancing at COCA, it seems these are usually followed by a predicand: "Mindful of the league's difficulties, Toliver has a singular mission..."

Regarding the fixed list in general: its composition seems to be a bit ad hoc, I think we really need some better principles for what should be in vs. out. Maybe "prior to" should be removed, but I would be more comfortable if there were a clear test. This seems like something the UniDive MWE folks should be able to help us with.

amir-zeldes commented 9 months ago

Yeah, I don't disagree with any of that, those are all valid points.

On a practical level it seems to me that whatever we do, there will be items on the fence, so my preference has always been to follow the maxim "pick one bad annotation scheme and stick to it". What I really don't like are frequent changes, and given that the fixed list has been around for UD English for a long while, I'm happy to live with its quirks just like I live with PTB quirks, but with the confidence that it's stable and I have a good idea of what's in there - then I can compare versions through time, compare different datasets "not built here" etc. For English upos, for example, I have very little confidence I can compare any two datasets out there in detail, but I'm more confident about the fixed list because it's been so stable across corpus/parser iterations.

nschneid commented 9 months ago

There was an older issue on this: #275

nschneid commented 9 months ago

It occurred to me to check what PTB does. They treat "other" as an adjective heading an ADVP:

image

It doesn't feel like a separate clause to me (unlike resultative secondary predicates, for example).

It turns out that an ADJ is allowed to attach as advmod, and this happens in the data, though a lot of these look like tag or tree errors: EWT, GUM

So what about

A few of the EWT examples have "other" as ADV/RB, but it's not consistent.

amir-zeldes commented 8 months ago

I guess I could live with that, though I almost wonder if we shouldn't tag other as RB then. Then again, PTB doesn't, so...