pronoun duplication

[jheinecke]

In Welsh the pronominal possessor can be repeated using a so-called independent pronoun. So both fy nhŷ and fy nhŷ i mean "my house". The second pronoun is optional, but frequent. Especially in spoken Welsh, there is a third form: tŷ (f)i or (with the mutation normally caused by fy, even in its absence): nhŷ (f)i

1  fy    fy  PRON/dep   ...   2 nmod:poss
2  nhŷ   tŷ  NOUS       ...

1  fy    fy  PRON/dep    ...   2 nmod:poss
2  nhŷ   tŷ  NOUS        ...
3  i     i   PRON/indep  ...   3

2  nhŷ   tŷ  NOUS        ...
3  (f)i  i   PRON/indep  ...   3

What deprel would be appropriate here? nmod:poss, dislocated, compound(:redup) ?
In other issues compound was proposed for other kinds of duplications (#238, #307 and #449), however this is not a dislocation neither a compound.

The same construction is also used for direct objects on verbnouns (cf. #752)

1	Mae	VERB   ...
2	o 	PRON   ...
3	wedi	ADP   ...
4	fy	PRON/dep	5	obj
5	ngweld	NOUN/verbnoun	1	xcomp
6	i	PRON/indep	5	obj (?)

Is the deprel of 6 i also an obj?

[ftyers]

There is also something similar in K'iche'. I would analyse it simply as nmod or nmod:poss with the idea that it's something like "my house (of me)"

[jheinecke]

I agree for the possessive. But what would you to in the case of verbnoun (fy ngweld i)? both, fy and i are obj. I tried to find similar examples in other treebanks, but the closest I found is in French-Spoken lui il vient ("he comes") where il is subject of vient, and lui is dislocated (attached to vient as well). But as mentioned above, the Welsh is is not a dislocation. Currently I use obj as the least bad solution...

[Stormur]

Are the pronouns used with the verbnoun functioning in the same way as with the noun? I mean, could it be sensible to consider a more nominal relation like nmod in this case, too, if the verbnoun is "embedded" in a nominal context? Is the ADP wedi also depending on it?

[jheinecke]

verbnouns behave syntactically as nouns. With the difference of posessives which correspond to the direct object of regular verbs fy nhŷ i "my house of me" fy ngwled i "my seeing of me" (= "my being seen"). wedi is a TAM for anteriority. Mae o wedi fy ngweld i means "He is after my seeing of me" = " He has seen me"
So in case of the non-verbnouns, I think (this time using 3rd person pronouns)

his house of him

1	ei	ei	PRON	dep	P=3|Num=Sg|Gen=M	2	nmod:poss
2	dŷ	tŷ	NOUN	noun	_		0	root
3	e	e	PRON	indep	P=3|Num=Sg|Gen=M	2	nmod

should do (knowing that ei and e have identical reference)

For direct objects I still hesitate to make both pronouns obj

his seeing of him (= seeing him)

1	ei	ei	PRON	dep	P=3|Num=Sg|Gen=M	2	nmod:poss
2	weld	gweld	NOUN	verbnoun	_		0	root
3	e	e	PRON	indep	P=3|Num=Sg|Gen=M	2	nmod

[Stormur]

Maybe, also to avoid the embarrassment of a double, but actually repeated obj, from what you say (a preposition used to convey aspect, the same possessives as for nouns, the verbnoun marked as NOUN....) the whole construction might be treated in a fully nominal way? I do not know, the verbal interpretation (i.e. using obj) almost looks to me like an "analogical reinterpretation", in the sense that in most other European languages "more verbal" forms are used, so we want to see the same here... but it really looks different, like a modal construction (with xcomp) or a non-verbal predicate. In the latter case, the verbnoun could be the root. So would something like this not be acceptable?

1	Mae	VERB    5? cop?  /  0  root
2	o 	PRON    5? nsubj? / 1  nsubj
3	wedi	ADP     5  case  
4	fy	PRON	5  nmod:poss
5	ngweld	NOUN	0  root  / 1  xcomp?
6	i	PRON	5  nmod(:poss)

For the copula part, I followed your translation. By the way, does not the validator complain for a nominal xcomp?
Of course, I understand that the main difficulty is the elusivity of nominal forms of verbs, which are somewhat syntactically ambiguous, which was discussed at length elsewhere...

Sorry for any misunderstandings, I am trying to better understand this (for me) unusual construction! 🙂

[jheinecke]

Currently it is annotated like

1	Mae	VERB    0  root
2	o 	PRON    1  nsubj
3	wedi	AUX     5  aux
4	fy	PRON	5  obj
5	ngweld	NOUN	1  xcomp?
6	i	PRON	5  obj

In fact, in my copy of the Welsh treebank I use nmod:obj instead of obj since with real verbs other pronouns are used. But currently I change them to obj to stress the "object function" of these pronouns.
The validator is OK with that. I followed the Irish treebank which annotates this construction similarly: ... tá sé ag obair (UD_Irish-IDT/ga_idt-ud-train.conllu, sent-id 915

15      tá      bí      VERB    PresInd Mood=Ind|Tense=Pres     1       conj    _       _
16      sé      sé      PRON    Pers    Gender=Masc|Number=Sing|Person=3        15      nsubj   _       _
17      ag      ag      ADP     Simp    _       18      case    _       _
18      obair   obair   NOUN    Noun    VerbForm=Vnoun  15      xcomp   _       _

with the difference that in Welsh wedi is a TAM (other can be in its place like yn, am, ar, heb, newydd) so it is tagged AUX and uses the deprel aux.

[Stormur]

with the difference that in Welsh wedi is a TAM (other can be in its place like yn, am, ar, heb, newydd) so it is tagged AUX and uses the deprel aux.

I saw that in your first post wedi is annotated as an ADP, and now I am wondering if it and all these other elements also work "regularly" as prepositions in Welsh and Irish. This would be one of the decisive arguments for how to treat such constructions, including fy - i...

In fact, in my copy of the Welsh treebank I use nmod:obj instead of obj since with real verbs other pronouns are used. But currently I change them to obj to stress the "object function" of these pronouns.

But if this is the case, then probably there is not much choice as to leave them as nmods, as you are already doing, no? If I understand that pronouns functioning as objects assume other forms.

[jheinecke]

you spotted it right, I made an error in my earlier posting, wedi should be AUX, not ADP. Since my question was about the doubled object, I did not recheck the rest.
So you propose putting all arguments of verbnouns as nmod which they are syntactically? Even though from a semantic point of view it's the object?

[sylvainkahane]

So you propose putting all arguments of verbnouns as nmod which they are syntactically? Even though from a semantic point of view it's the object?

Yes the UD annotation is supposed to be syntactic. For instance, in English, Chomsky will be nmod in the critic of Chomsky , no matter if it is agent or patient.

[dan-zeman]

For direct objects I still hesitate to make both pronouns obj

You cannot have two children of one predicate both attached as obj.

And I, too, would use nmod, following syntax and ignoring semantics.

[jheinecke]

@dan-zeman that's exactyl why I hesitate(d) :-) Since in Welsh the verbnoun is far more frequent (currently 3025 verbnouns against 756 inflected verbs (other than the auxiliary bod ) annotating (semantic) objects as nmod would result many nmod-dependents on a same head (verbnoun) and we would lose information. I think in this case I should create nmod:obj Welsh-specific deprels or another standard is needed (Irish and Scottish-Gaelic also use obj/obl for dependents of verbnouns (tagged NOUN)