Open tzshi opened 6 years ago
From looking at a few of these examples, it seems like all of them should be annotated like the first two examples that you provide (like, e.g., Prime Minister Iyad Allawi).
Pull requests for fixing inconsistencies are always welcome :)
I never understood why President Obama should be flat
. The main and simple test to decide whether a construction AB is headed is to look whether AB can be replaced only by A or only by B.
Clearly President Obama can be replaced by Obama and not by President:
President Obama came to Paris Obama came to Paris *President came to Paris
So this seems to be the common English noun-noun construction, which is analyzed as compound
in other cases. Same thing for Prime Minister Iyad Allawi, French actor Gaspard Ulliel, Leader Mahmoud Abbas, or Mister Miller.
flat
is only for headless constructions.
in that case, Prime Minister Iyad Allawi, flat is used because both are possible as head [Prime Minister] or [Iyad Allawy], no ?
@dseddah Well, but you still cannot say *"Prime Minister came to the meeting" without adding a determiner.
@sylvainkahane I personally totally agree with your analysis but others don't. See, e.g., https://github.com/UniversalDependencies/docs/issues/324.
@sebschu Thanks for the answer.
UniversalDependencies/docs#324 was about name
, which no longer exists. Now UD has moved to more syntactic criteria and flat
is clearly defined as a relation for headless constructions. Today, nothing justifies to use flat
for President Obama, except for the history of UD.
I agree with @sylvainkahane that in President Obama, Obama is the head in English (because you cannot just remove Obama without adding an article). However, it is not necessarily true about corresponding phrases in other languages. For example, Czech does not have articles, and you can replace the phrase prezident Obama by either prezident or Obama. Therefore, the phrase is headless in Czech, and suitable for the flat
relation.
Unfortunately, people sometimes apply rules from one language to translated phrases in another language without noticing that the syntactic environment has changed. Names and titles are particularly prone to this sort of misguidance because they often look very similar in the two languages. We had the problem with name
and we apparently still have it with flat
, so we should make sure that the documentation makes the cross-language differences very clear, with examples from multiple languages.
Hi all, I just wanted to point out that if we take "President X" out of flat, then the same analysis should be applied to "Mr." and other titles in English. Is that something we want for languages with English-like determiners? I think it would affect Romance and Germanic as well.
Note also that there can be a sequence of titles, in which case I'm not sure about their internal headedness (e.g. German "Frau Professor Doktor X"). In languages with strong case marking I think there's usually agreement (e.g. Polish "pana direktora" - Mr./gen. director/gen.), so it's hard to motivate what's governing what.
In the German example, you could probably do
flat(Frau, Professor)
flat(Frau, Doktor)
nmod(X, Frau)
In Polish, I suppose it should be
flat(pana, direktora)
I think this could be OK, but the grammatical agreement makes me hesitate with nmod as the label. Usually nmod has its case based on its function (e.g. genitive modifiers, an adnominal locative, etc.). For these cases, all of the modifiers agree with the name itself, so they look almost more like adjectives. It's hard to see in German, but I guess you'd just need an n-stem. So if we are talking to a Mr. Raven in a fairy-tale:
Herr Rabe! (vocative=nominative in German) Ich sehe Herrn/acc Raben/acc
I suspect it's the same in most Slavic languages - it's not exactly an nmod like other adnominal phrases.
@amir-zeldes Could it be that German is reusing the adjectival modification construction for titles like Herr? Or are there tests for attributive adjectives that would fail?
It seems that different languages have recruited different core-grammar constructions for the formation of names, so maybe we can make the commonalities explicit with cross-cutting subtypes. E.g. English could have compound:pre-name
whereas German could have amod:pre-name
for titles like Mr./Herr.
Herr is NOUN
, definitely not ADJ
. Therefore, if it is *mod
, then it must be nmod
, nod amod
.
Yes, German adjectives have distinctive morphology, this is definitely a noun
When I looked into cases where a word has both
compound
andflat
children, I see some seemingly-inconsistent annotations.Examples from Grew: http://match.grew.fr/?custom=5ace8a92b2c1f&corpus=UD_English
We have for example: ... Prime <-[compound]- Minister -[flat]-> Iyad Allawi ... ... U.S. <-[compound]- president -[flat]-> George Bush ...
and we also have: ... Leader <-[compound]- Mahmoud -[flat]-> Abbas ... ... adviser <-[compound]- Karen -[flat]-> Hughes ...