Open arademaker opened 2 years ago
Yes I think for now, it should be two PPs attaching to the verb. (There is perhaps an argument to be made that "from X to Y" is a special construction reminiscent of coordination, but that is certainly not the standard practice currently.)
In some French corpora we analyzed "de X à Y" as a coordination : http://match.grew.fr/?corpus=SUD_French-Rhapsodie@latest&custom=61961d959f3e2
They are good reasons to do that in French.
1) "de X à Y" can occupy different positions, including the determiner position:
on attend de 10 à 20 personnes 'we expect 10 to 20 people' (de is optional)
2) the order is fixed (French as a freer order than English)
le train de Paris à Londres 'the train from Paris to London' *le train à Londres de Paris 'the train from Paris to London'
3) NCC is possible:
un vent de force 4 le matin à force 5 l'après-midi 'a wind from force 4 in the morning to force 5 in the afternoon'
4) it can be cleft (normally only one constituent can)
_c'est de chez toi à chez moi que c'est le plus court 'it's from your place to mine that it's the shortest'
In some French corpora we analyzed "de X à Y" as a coordination : http://match.grew.fr/?corpus=SUD_French-Rhapsodie@latest&custom=61961d959f3e2
That is SUD. Would the UD equivalent for "aller de la place Notre-Dame à La Nef Chavant" be:
obl(aller, place)
case(place, de)
conj(place, Nef)
case(Nef, à)
or perhaps
obl(aller, place)
cc:preconj(place, de)
conj(place, Nef)
cc(Nef, à)
?
English equivalents of the above examples:
It's worth considering "between X and Y" as well—I suspect that would have a similar analysis, with the differences that (i) "between" is mandatory (*We expect 10 and 20 people); (ii) "and" is readily recognizable as a coordinating conjunction, whereas "to" is mainly a preposition (arguably a conjunction when expressing a range, though).
I think wanting to connect the "from" phrase to the "to" phrase in some contexts (e.g. when there is no verb) could make sense, but I would still expect the normal PP deprels (nmod/obl etc.), and I imagine most users of the data would expect those as well.
We just ran into a similar issue in the upcoming GUM8 data, where a present is described as "From Santa, to Dan" (no other words in the sentence), where I think the simplest option is obl (treating "from Santa as root", and "to Dan" as a modifier of the predication, not of "Santa"). Alternatives we considered were orphan with a reconstructed verb (seems excessive), nmod (seems wrong since it's not really adnominal) and parataxis.
Our UD analysis for "aller de la place Notre-Dame à La Nef Chavant" is:
obl(aller, place) case(place, de) conj(place, Nef) case(Nef, à)
In this example, the two complements can be inverted: "aller à La Nef Chavant de la place Notre-Dame". In the other example, with a figurative sense of ALLER 'go', the inversion is impossible:
ça va vraiment de romans étrangers à quelques romans français contemporains 'it really goes/ranges from foreign novels to some contemporary French novels' *ça va vraiment à quelques romans français contemporains de romans étrangers
For the POS, de 'from' and à 'to' are ADPs. It would be strange to analyze them as CCONJs, even if they play a role in the coordination. These prepositions keep their original meaning (origin vs destination) in this construction. Same thing in English.
We just ran into a similar issue in the upcoming GUM8 data, where a present is described as "From Santa, to Dan" (no other words in the sentence), where I think the simplest option is obl (treating "from Santa as root", and "to Dan" as a modifier of the predication, not of "Santa"). Alternatives we considered were orphan with a reconstructed verb (seems excessive), nmod (seems wrong since it's not really adnominal) and parataxis.
"From Santa, to Dan" or "To Dan, from Santa" is not the "from X to Y" construction we're talking about. It's perhaps a convention of the genre to put these two PPs next to each other—I would call it parataxis or asyndetic coordination (could be paraphrased with "and").
Yeah, I realize it's not the same, I just thought it was similar in that they seem to belong together, though semantically there is an implied predicate whose valency they fill, and we have to connect them somehow due to the environment. I'd still be against calling it a coordination, that's something different again (in a coordination I'd expect both conjuncts to saturate the same single role as a coordinate phrase, and in "I went from La Nef Chavant to Notre Dame" they saturate different roles - source and destination)
Thank you @nschneid , so maybe we can keep this issue open since your query return different analyses, right? One situation where we don't have a common VERB head is
The 22 member countries of the Arab league, from Mauritania to the Gulf States, have a total population of 300 millions, larger than the US and almost as large as the EU before its expansion.
My question was motivated by the Ontonotes example:
Through a series of transitional forms for state capitalism , ranging from low to high levels , and after taking into consideration China 's national conditions , the Party implemented " peaceful redemption " on capitalist owned means of production based on the visions of Marx and Lenin .
the only problem of making [from low] and [to high levels] both obl of 'ranging' is that we lose the connection of 'low' and 'levels' because both low and high are levels.
UniversalDependencies/docs#871
This analysis of
nmod(years, days)
seems wrong to me, am I right? How should we analyze this kind of structure? Other cases that I found take thefrom X
andto Y
as both obl from the verb.