Passive sentences with 'cael'

[ftyers]

How should we annotate sentences of the type Mi gafodd o ei ladd ym Mangor. "He got killed in Bangor" (lit. "He got his killing in Bangor").

Proposal (1) is:

This states that:

• The root of the sentence is gafodd, a normal VERB
• ladd is a clausal complement of gafodd where the subject is determined by the higher clause
• Mi is a verbal "particle" which expresses some TAMVE category
• ei is an object pronoun of ladd.

Does anyone have any thoughts on this? Particularly:

• Should we really consider ei an object here ?
• Is the object of the complement clause ei always the same as the subject of the higher clause o ?
  • If so, perhaps expl might be a better analysis?
• Might it be better to treat this as an auxiliary construction, where ladd is the main verb ?

[dan-zeman]

Why xcomp(gafodd, ladd)? The subject of ladd corresponds neither to the subject nor to the object of gafodd, or does it?

[jnivre]

I agree with @dan-zeman. This doesn't look like xcomp to me. What kind of verb form is "ladd"?

[ftyers]

@jnivre it's a verbal noun/infinitive. ei could be seen as a determiner or pronoun.

@dan-zeman afaik, the subject of gafodd must be the object of ladd, @jheinecke, @donnekgit is this right?

[jnivre]

But xcomp is used only when the implicit subject must be identical to an argument in the higher clause, not when an explicit object is required to be co-referential. So if it is going to be analysed as a complement clause, it seems ccomp must be used (even though the co-reference constraint is not captured). The alternative would be some kind of complex predicate analysis, but I don't know if this is relevant at all.

[dan-zeman]

So the tree given by @ftyers above seems almost OK to me, with two exceptions:

• Replace xcomp by ccomp
• The UPOS tag of the particle mi should be AUX instead of PART

[tsamardzic]

I don't know much about the syntax of Welsh, but this cannot be ccomp because the verb in the subordinated clause is non-finite, and therefore cannot have its own syntactic subject. This definitely looks like a complex/decomposed predicate to me, which means that the two verbs share arguments and the lexical one (ladd) is xcomp.

The confusing factor here is the fact that the construction is passive/reflexive, introducing a co-reference between o and ei, which might have nothing to do with the construction itself. I think it will be helpful to consider another potential literal translation to English:

He got himself killed in Bangor.

The unexpressed passive subject of killed is the same as the one of got. The co-reference between he and himself is expressed with the reflexive form (which might not be used in Welsh) and is sort of accidental in this case: it would exist even if the predicate were not decomposed as in He saw himself in Bangor).

[jnivre]

First of all, ccomp neither has to be finite nor has to have its own subject. See the documentation at: http://universaldependencies.org/u/dep/ccomp.html. An example of a non-finite ccomp without an overt subject is: "The boss said to start digging". The point is that, in this example, the implicit subject of "digging" is not constrained to be an argument in the higher clause. Therefore, it is ccomp not xcomp.

But back to the Welsh example. I also thought about the construction "he got himself killed in English", but the difference is that (according to the UD analysis) "himself" is an argument (the object) of "got", not of "killed" (except implicitly through the control relation). And the assumption here seems to be that "ei" is an argument of "ladd", not "gafodd", but perhaps this is open to discussion. If you adopt a small clause analysis of the English example (which UD does not), then I suppose you would say that "himself" is the subject of "killed" (with exceptional case-marking).

[tlynn747]

Hi Fran

I'm a bit confused - a gloss line might help a bit!

But my two cents worth:

If ei can mean "his", then ladd is a verbal noun, not a gerund. obj(gafodd, ladd) , nmod:poss(ladd, ei) I'm not clear on why Mangor is attached the ladd? Should it not be an adverbial dependant of the matrix verb? I also don't think it's an oblique agent.... nmod(gafodd, Mangor)

Teresa

2018-05-21 11:42 GMT+01:00 Francis Tyers notifications@github.com:

How should we annotate sentences of the type Mi gafodd o ei ladd ym Mangor. "He got killed in Bangor" (lit. "He got his killing in Bangor").

Proposal (1) is:

[image: captura de 2018-05-21 12-32-55] https://user-images.githubusercontent.com/449545/40303462-1f212060-5cf3-11e8-98b6-0fdd9944b659.png

This states that:

• The root of the sentence is gafodd, a normal VERB
• ladd is a clausal complement of gafodd where the subject is determined by the higher clause
• Mi is a verbal "particle" which expresses some TAMVE category
• ei is an object pronoun of ladd.

Does anyone have any thoughts on this? Particularly:

• Should we really consider ei an object here ?
• Is the object of the complement clause ei always the same as the subject of the higher clause o ?
• Might it be better to treat this as an auxiliary construction, where ladd is the main verb ?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/UniversalDependencies/UD_Welsh-CCG/issues/1, or mute the thread https://github.com/notifications/unsubscribe-auth/AGyMg_D_-GePCyZo7VZTML86Y8JuFNheks5t0pojgaJpZM4UGyhd .

-- Slán agus Beannacht

[donnekgit]

Gloss line (autoglosser.org.uk): Cafodd#get.V.3S.PAST (f)o#he.PRON.M.3S ei#his.ADJ.POSS.M.3S ladd#kill.V.INFIN+sm ym#in.PREP Mangor#Bangor.NAME.F.SG+nm .

This is the standard passive construction in Welsh (for non-impersonals).

Might it be better to treat this as an auxiliary construction, where lladd is the main verb ?

I think so. cael is used as a modal auxiliary in non-passives too, eg cael gofyn = get to ask, be allowed to ask, BUT it cannot be conjugated in this meaning. Once it is conjugated it is passive (at least, in the idiolects I have access to...), and then it must have a possessive pronoun agreeing with the subject of cael. That is: cafodd ofyn (he got to ask) is not allowed - you'd have to say something like llwyddodd i ofyn (he succeeded in asking) cafodd ei ofyn (he was asked) is OK.

Where cael is not conjugated, but is in a periphrastic tense with bod (be), the possessive (with mutation as appropriate) is the only indicator of passivity: mae o'n / mae'n cael gofyn (he's getting to ask) mae o'n / mae'n cael ei ofyn (he's being asked) In some periphrastic tenses using prepositions, cael can be omitted: mae o wedi cael ei ofyn = mae o wedi ei ofyn (he's been asked) mae o heb gael ei ofyn = mae o heb ei ofyn (he hasn't been asked) But not in others: mae o am gael ei ofyn (he's going to be asked) BUT mae o am ei ofyn (he's going to ask him)

Should we really consider ei an object here ?

Can a passive verb have an object? Functionally, the person to whom the action is done is the object of the action, but grammatically they are the subject, no? Ei" is a possessive pronoun. Compare: mi gafodd [o] ei lyfr ym Mangor (he got his book in Bangor) mi gafodd [o] ei ladd ym Mangor* (he was/got killed in Bangor) [lit. got his killing]

Is the object of the complement clause ei always the same as the subject of the higher clause o ?

In this passive construction, yes. (A possibly rash definitive answer!)

He got himself killed in Bangor.

The problem with trying to generate "universal" grammar is that we end up looking at everything through the prism of English ... :-) He got himself killed and He got killed are actually different in meaning in English, so I'm not sure this example sheds much light on the Welsh - I'd probably use llwyddo (succeed) or something to get this meaning.

And the assumption here seems to be that "ei" is an argument of "ladd", not "gafodd", but perhaps this is open to discussion.

I'd agree.

The UPOS tag of the particle mi should be AUX instead of PART

No, mi is an affirmative particle, not an auxiliary. Fe is equivalent (depends on dialect). A negative one is ni[d], and an interrogative one is a. What tag is being used for Irish a, ní, d[h]', etc?

I'm not clear on why Mangor is attached the *ladd?

For me it's a locative phrase qualifying ladd. I don't know how UD tags that ...

CAVEAT: The above is based on my experience, but other speakers may allow other things. :-)

[tsamardzic]

Ah, sorry to have overlooked the non-finite ccomp mentioned by @jnivre (has to be finite in my language).

I tried to find some default approach to causative constructions in the guidelines and in the issues, but didn't find any specific instructions. On the other hand, there has been some discussion on light and serial verbs. Perhaps this example can be analysed along these lines? The verb ladd is clearly a mixed category: it has the content of a verb, but syntactically behaves as a noun (complements a verb, is modified by an adjective). These are exactly the characteristics of nominal complements of light verbs.

[jheinecke]

Hello, it all depends whether lladd(here mutated to ladd) is seen as a noun or a verb. Technically it is a verbnoun (and not an infinitive). But as a (verb)noun it can not have nsubj or obj relations in UD. I saw that verbnouns in the Irish treebank have the UPOS VERB, so if we do so in Welsh too, I propose to have lladd as the root and cael (gafodd) has an auxiliary. I response to @tlynn747 : ym Mangor should be attached to ladd since this is the main verb (if we see verbnouns as verbs. I agree with @donnekgit that mi is a preverbal particle and not an auxiliary (it can be replaced by a interrogative (a) or negative (na) particle).

BTW. the english translation is nice but syntactically to far from the Welsh in order to projet the relations ontot the Welsh example)

[jheinecke]

I forgot the case with redoubled object mi gafodd o ei ladd o where the object ei is redoubled after the verbnoun with o. Since we cannot have two objects, which one is the "correct" object ? In non "passive" sentences we can see this everywhere two Mae hi'n ei garu o (she loves him)
Mae hi'n caru Siôn (she loves Siôn)

with an non-pronominal object, the object follows the verbnoun, pronominal objects can be in either place

[jnivre]

Just one clarification: using the "aux" relation does not imply that it is an auxiliary VERB. The "aux" relation can (and should) be used also for nonverbal TAME particles.

[tlynn747]

Hi Johannes, all

I agree with Kevin's statement: The problem with trying to generate "universal" grammar is that we end up looking at everything through the prism of English

I can give some responses with respect to Irish here as the Irish analysis tries hard to apply the UD analysis while retaining the true analysis of the language.

Verbal Nouns In response to Johannes: I saw that verbnouns in the Irish treebank have the UPOS VERB

Verbal nouns in Irish do not have the UPOS VERB.... They are tagged as NOUN, and their features indicate their verbal nature. I've taken a couple of examples from the Irish treebank:

Verbal nouns can be used as infinitive when preceded by the infinitive particle 'a'. a imirt 'to play' 23 imirt imirt NOUN Noun VerbForm=Inf 13 xcomp Caithfidh sé an cluiche a imirt 'He has to play the game'

Or in progressive aspectual phrases to indicate the equivalent of the English present participle. ag imirt 'playing' 35 imirt imirt NOUN Noun VerbForm=Vnoun 31 xcomp Tá sé ag imirt - he is playing (lit he is at playing)

xcomp is used in both of these examples, as they share a subject with a higher verb. I can't find any example in the Irish treebank where a possessive precedes the verbal noun. But of course, it is possible, and I would label the attachment as 'obj' and thus 'ei' as a possessive modifier of the noun.

In response to @tlynn747 https://github.com/tlynn747 : ym Mangor should be attached to ladd since this is the main verb (if we see verbnouns as verbs).

Given Fran's analysis, and based on this: mi gafodd [o] ei lyfr ym Mangor (he got his book in Bangor) mi gafodd [o] ei ladd ym Mangor (he was/got killed in Bangor) [lit. got his killing]

I would argue that gafodd is the main verb. Possibly these could be analysed as you would a light verb. e.g. "Make a decision" - while "make" doesn't carry the semantics of the entire verb, it's the head of the verb phrase. Thus "Make a decision in the morning".... "morning" would be a dependent of "make", not "decision".

To answer Kevin: What tag is being used for Irish a, ní, d[h]', etc? We use PART for these. We only use AUX for the copula (was formerly tagged as VERB until v2).

Proposed analysis is: obj(gafodd, ladd) nmod:poss(ladd, ei)

mi UPOS=PART (like Irish verb particles, ní, nach, d(o), an) ladd UPOS=NOUN (including verbal features in the morphological features column)

Passives This may not be relevant for the Welsh analysis - but worth putting out there in response to Kevin's question "Can a passive verb have an object? "

We don't have passives in Irish. We have impersonal (autonomous) verbs, so they don't follow the usual rule of an English passive verb, where an object is raised to subject position. Instead, there is an understood subject in the impersonal form, and the nominal argument is the 'object'. [This can be proven by the case of pronouns (only words in Irish that mark for Accusative case) in such examples.]

Láimhseáltar an ghné seo 'this aspect is handled'

1 Láimhseáltar láimhseál VERB PresInd Mood=Ind|Tense=Pres|Voice=Auto 0 root 2 an an DET Art Definite=Def|Number=Sing|PronType=Art 3 det 3 ghné gné NOUN Noun Case=NomAcc|Definite=Def|Gender=Fem|Number=Sing 1 obj 4 seo seo DET Det PronType=Dem 3 det

Initially our parser got confused with the lack of subject in such examples- but once we introduced the feature Voice=Auto, the parser predicted the object correctly. (discussed in this paper https://researchers.mq.edu.au/en/publications/morphological-features-of-the-irish-universal-dependency-treebank )

Happy to have a Skype discussion (as per Fran's suggestion) at some stage to compare treebank analyses. Our documentation for the UD treebank is complete (POS, Dep labels, Features) - but needs a couple of tweaks for v2 before it can be migrated to the new site. We're working on it at the moment. That may help also when it's available. And if not, while based on a different annotation scheme, discussions of all syntactic aspects of Irish wrt to dependency parsing can be found in my thesis: http://doras.dcu.ie/21014/1/Teresa_PhDThesis_final.pdf

Teresa

2018-05-22 8:53 GMT+01:00 Johannes Heinecke notifications@github.com:

Hello, it all depends whether lladd(here mutated to ladd) is seen as a noun or a verb. Technically it is a verbnoun (and not an infinitive). But as a (verb)noun it can not have nsubj or obj relations in UD. I saw that verbnouns in the Irish treebank have the UPOS VERB, so if we do so in Welsh too, I propose to have lladd as the root and cael (gafodd) has an auxiliary. I response to @tlynn747 https://github.com/tlynn747 : ym Mangor should be attached to ladd since this is the main verb (if we see verbnouns as verbs. I agree with @donnekgit https://github.com/donnekgit that mi is a preverbal particle and not an auxiliary (it can be replaced by a interrogative (a) or negative (na) particle).

BTW. the english translation is nice but syntactically to far from the Welsh in order to projet the relations ontot the Welsh example)

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/UniversalDependencies/UD_Welsh-CCG/issues/1#issuecomment-390897076, or mute the thread https://github.com/notifications/unsubscribe-auth/AGyMg4_GhbhfzLXGltILPlwRVWB0NFBAks5t08QFgaJpZM4UGyhd .

-- Slán agus Beannacht

[ftyers]

This is great discussion!

• I've made a separate issue for the NOUN/VERB for verbal nouns, see #2.
• I've also made a separate issue for the double object construction, see #3.

In addition, @tlynn747 @donnekgit, would you mind if I changed some of the formatting in your posts ? (but not the content obviously).

[tlynn747]

go for it...

2018-05-22 12:42 GMT+01:00 Francis Tyers notifications@github.com:

This is great discussion!

• I've made a separate issue for the NOUN/VERB for verbal nouns, see #2 https://github.com/UniversalDependencies/UD_Welsh-CCG/issues/2.
• I've also made a separate issue for the double object construction, see #3 https://github.com/UniversalDependencies/UD_Welsh-CCG/issues/3 .

In addition, @tlynn747 https://github.com/tlynn747 @donnekgit https://github.com/donnekgit, would you mind if I changed some of the formatting in your posts ? (but not the content obviously).

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/UniversalDependencies/UD_Welsh-CCG/issues/1#issuecomment-390960162, or mute the thread https://github.com/notifications/unsubscribe-auth/AGyMg7awPAMUqVR4Q6aZS4NoxOudFL7_ks5t0_mZgaJpZM4UGyhd .

-- Slán agus Beannacht

[ftyers]

@tlynn747 actually it seems like editting email replies with Markdown isn't possible. No worries then :)

[donnekgit]

mi gafodd o ei ladd o

I think this is less likely to occur than: mi gafodd o ei ladd (more formal) gafodd ei ladd (more conversational) In other words, the "sandwich" possessive (see #3) is unlikely to occur in a passive, probably because it's clear who is being referred to.

where the object ei is redoubled after the verbnoun with o. Since we cannot have two objects, which one is the "correct" object ?

The sandwich possessive should probably be considered as a discontinuous whole.

In non "passive" sentences we can see this everywhere two Mae hi'n ei garu o (she loves him) Mae hi'n caru Siôn (she loves Siôn) with an non-pronominal object, the object follows the verbnoun, pronominal objects can be in either place

This is because the English pronominal object of the action instantiates in Welsh as a pronominal possessor of the action, which is mostly realised in the sandwich form: mae hi'n ei garu o lit. she is in/at his loving (compare non-standard English "she be a-lovin' of 'im") Where the object is a noun rather than a pronoun, it instantiates after the action: mae hi'n caru Siôn mae hi'n caru'r ddrama (she loves the play)

[jnivre]

UD currently does not allow multiple instances of core argument relations (with the same predicate). Maybe this is something that needs to be reconsidered for future versions, but right now this limits the possibilities to the following:

1. Mark one as core and the other as "dislocated". This is used, for example, in French for cases like:

Jean il ne vient pas Lui il ne vient pas

Here "il" is "nsubj" and "Jean/lui" is "dislocated".

2. Mark one as core and the other as expl. This is recommended for cases of "true" clitic doubling, where the clitic is obligatory regardless of whether the other realisation is nominal or pronominal. I understand this is found in Greek. It sounds to me like it could be relevant here as well, except that is seems hard to tell which is the doubled one. :)

3. Mark the pair as a "fixed" (but discontinuous) expression, that is, mark the leftmost one as core and attach the rightmost one to it with a "fixed" relation. This is used for sandwich prepositions (or circumpositions) in Swedish, for example, but I am not sure I want to recommend it for core arguments (and it could in any case only be used for the pronominal case).