Docs: Inflected prepositions (deoc'h)

[jheinecke]

in sentence apertium.vislcg.txt:21:280 (and others) _deoc'h_ (inflected preposition meaning to-you (PL) is annotated as a MWE:

3-4    deoc'h  _       _       _       _       _       _       _       _
3      deoc    da      ADP     pr      _       4       case    _       _
4      h       indirect        PRON    prn     Case=Acc|Number=Plur|Person=2|PronType=Prs      1       obl     _       _

which is not correct, since _c'h_ is considered as a single letter (for the phoneme /x/). I doublt whether inflected prepositions can be analysed as MWE at all, even though historically the were two distinct words (and in other Celtic languages, like Welsh, the inflected forms can be followed by a pronoun: _amdano fo_ "for+3SGMASC him")
For Breton I would propose either an annotation as in sentence apertium.vislcg.txt:22:295

3-4     dezhañ  _       _       _       _       _       _       _       _
3       dezhañ  da      ADP     pr      _       4       case    _       _
4       _       indirect        PRON    prn     Case=Acc|Gender=Masc|Number=Sing|Person=3|PronType=Prs  1       obl     _       _

(but what is the point having a MWE when the second part is empty)
or as in Welsh, Irish and Scottish Gaelic (i.e. ADP with morphological features indicating Number and Person)

Welsh:

8       amdanyn am      ADP     cprep   Number=Plur|Person=3    9       case    _       _
9       nhw     hwy     PRON    indep   Number=Plur|Person=3|PronType=Prs       7       obl     _       SpaceAfter=No

Irish:

24      againn  ag      ADP     Prep    Number=Plur|Person=1    23      obl:prep        _       _

15      orm     ar      ADP     Prep    Number=Sing|Person=1    12      obl:prep        _       SpaceAfter=No

Scottish-Gaelic:

4       dhuibh  do      ADP     Pr2p    Number=Plur|Person=2    3       obl     _       _

[dan-zeman]

Is it still a preposition if there is nothing it is preposed to (because the nominal has been consumed by the preposition)? If it is not analyzed as a multi-word token, shouldn't it be a pronoun rather than a preposition?

[colinbatchelor]

I quite like the pronoun solution but it does lose the information that _orm_, say, is related to the preposition _air_. Currently we record this in the LEMMA field, but with the pronoun solution _orm_ would lemmatise to _mi_ and where would _air_ go? (Other than being obl:air in extended dependencies, I suppose.)

[jheinecke]

Good question. I could be seen as a pronoun with a case-prefix. But there are still case in Breton where the pronoun is not dropped: _deoc'h c'hwi_ "to-2PL you(PL)", _deomp ni_ "to-1PL us". In Welsh and Irish (and Scottish Gaelic) the pronoun is less often dropped. So to avoid tagging deoc'h differently whether the pronoun is present or absent I would go for ADP

[colinbatchelor]

I don't think I've ever seen similar cases in Gaelic: _dhuibh sibh_ 'to-2PL you(PL)' and _dhuinn sinn_ 'to-1PL us' are just wrong, I think. Are those Breton examples emphatic forms, where Gaelic would have _dhuibhse_ and _dhuinne_?

[jheinecke]

I saw some cases in the Irish treebank though. I guess (as often in prodrop languages) they are likely to be emphatic. I'll have to check (don't have the right stuff here right now)

[dan-zeman]

If there are still cases where the pronoun is not dropped, then I think I would go for the MWT analysis. (But I would not leave the form field of the pronoun empty. I would put the full pronoun there. The MWT mechanism was designed so that the surface form does not have to be a concatenation of the individual syntactic words.) That is the only way that allows to show the link to both the preposition and the pronoun, as there will be two lemmas and two UPOS tags. And it also allows to avoid the unusual situation where a nominal consists of nothing else than a preposition.

[jheinecke]

But MWT would be bad if the pronoun followed, (_deoc'h chw'i_) because than you had something like
1-2 deoc'h
1 da
2 chw'i
3 chw'i

I see these prepositions similar to prodrop verbs like in Spanish: _quiero_ "I want" is not analysed as a MWT of _yo_ and _quiero_ neither.

[dan-zeman]

Okay, that makes sense. I did not know that the inflected preposition can occur also with the full pronoun.

[dan-zeman]

I am transferring this issue to the docs repository, as its importance exceeds the Breton treebank and language.

[ftyers]

The first part (e.g. splitting deoc'h as deoc' + h) is a bug in my conversion script and I should fix it.

The second part, splitting or not splitting I think is more open to interpretation. In Breton, they are typically pretty compositional so I've analysed them as contractions of preposition and pronoun, much like in Spanish contigo. In Breton they are not emphatic forms.

One of the reasons I don't like the ADP analysis is that the end result of this contraction (in UD) is a pronoun, not an adposition, e.g. the head of case(PRON, ADP) is the PRON. Of course then the question would be "do we encode the preposition that is contracted then as a feature?" ... Prep=Da, that might be a bit ugly.

Anyway, I'm glad this has come up and I think it is useful to think about.

[ftyers]

But MWT would be bad if the pronoun followed, (_deoc'h chw'i_) because than you had something like
1-2 deoc'h
1 da
2 chw'i
3 chw'i

I see these prepositions similar to prodrop verbs like in Spanish: _quiero_ "I want" is not analysed as a MWT of _yo_ and _quiero_ neither.

Can you give an example of this, is it like "clitic" doubling?

[jheinecke]

Since in the majority of cases in Breton, the pronoun is not repeated after an inflected preposition,
I'm not 100% sure whether the presence of the pronoun is emphatic or not, but there is evidence. Some examples

• _Gloar deoc’h-c’hwi, maro evidom_ "glory to you, [who] died fo us"
• _Bonjour deoc'h c'hwi plac'hig_ "bonjour to you, little girl"
• _Deoc'h-c'hwi bremañ_
• _Ganeoc'h-c'hwi pechañs e c'hellimp ranniñ ..._ "With you we hope to share ..."
• _Hiziv c'hoazh e vez savet ganeomp-ni asambl_

Jules Gros (1984) Le Trésor du Breton Parlé, p.120 mentions these cases as emphatic:

• _Din-me a roi anezañ da gentañ_ "c'est à moi que tu le donneras d'abord"
• _Dit-te na lavarin ket_ "[à] toi, je ne te le dirai pas"

I wonder whether there are important difference between speakers coming from different dialect background, since the "standard" of Breton is debatable (cf. Iwan Wmffre (2007), Breton Orthographies and Dialects: the Twentieth-century Orthography War in Brittany, vols.1–2, Bern: Peter Lang (Series: Contemporary Studies in Descriptive Linguistics) (xxviii + 782 pp)

To resume, we have two cases in at least Welsh, Breton and Irish

• inflected preposition + pronoun
• inflected preposition

If we chose an MWE annotation, we would have two pronouns if the pronoun is not dropped. In this case, how can we attach the second (independent) pronoun ?

e.g. Welsh _amdani hi_ "about her":

1-2 amdani  _   _
1   am  am  ADP 2   case
2   hi  hi  PRON    0   root
3   hi  hi  PRON    ?   ?

But since in prodrop languages (like Spanish) we do not analyse _quiero_ as

1-2 quiero  _   _
1   yo  yo  PRON    2   nsubj
2   quiero  querer  VERB    0   root

so why should we do the following:

1-2 amdani  _   _
1   am  am  ADP 2   case
2   hi  hi  PRON    0   root

Of, course, in the Spanish case, the dependent (_yo_, nsubj) is dropped, in the case of Celtic
the head (_hi_) is dropped, so it is really the same thing.

In Arabic-PADT and Hewbrew the prepositions with pronominal clitics are annotated as MWE.
But these prepositions are not inflected as the Celtic ones but simple prepositions with clitics and in both languages no additional pronoun can follow:

7-8 فيه _   _
7   في  فِي ADP     8   case
8   ه   هُوَ    PRON    6   obl

11-12   بينها   _   _
11  بين بَينَ   ADP     12  case
12  ها  هُوَ    PRON    10  obl

2-3 לו  _   _
2   ל_  ל   ADP 3   case
3   _הוא    הוא PRON    1   obl

and Amharic

1-2 ከኔ  _   _   _   _   _   _   _   _
1   ከ   ከ   ADP ADP _   2   case    _   Translit=kä|LTranslit=kä
2   እኔ  እኔ  PRON    PRON    _   7   nmod    _   Translit='əne|LTranslit='əne

But in any case I agree at least for the Celtic languages the inflected pronouns should be annotated in a similar way.
Currently only the Breton treebank annotates inflected pronouns as MWEs.

The Irish treebank tags the inflected prepositions as ADP with morphological features for Number and Person (and Gender in the case of 3SG), if depending from a verb the deprel is obl:prep

3   liom    le  ADP Prep    Number=Sing|Person=1    1   obl:prep

10  dúinn   do  ADP Prep    Number=Plur|Person=1    5   obl:prep

Welsh is annotated similarly to Irish (except the deprel is only prep):

15  gennym  gan     ADP     cprep   Number=Plur|Person=1    14      obl

Unlike Irish and Scottish Gaelic, In Welsh (and to a lesser extend in Breton) the pronoun can follow the inflected preposition

8   wrthot  wrth    ADP     cprep   Number=Sing|Person=2    9       case
9   ti      ti      PRON    indep   Number=Sing|Person=2|PronType=Prs       5       obl

Scottish Gaelic again follows the Irish and Welsh scheme:

4   aca aig ADP Pr3p    Number=Plur|Person=3    2   obl

2   dhomh   do  ADP Pr1s    Number=Sing|Person=1    1   obl

[nschneid]

Of, course, in the Spanish case, the dependent (_yo_, nsubj) is dropped, in the case of Celtic
the head (_hi_) is dropped, so it is really the same thing.

With pro-drop it is straightforward to omit the pronoun because verbs are heads. But adpositions normally are not heads, which makes things awkward for Celtic and Semitic.

How's this as as summary of options:

1. When there is an inflected preposition + separate pronoun, the options are
   
   
   
   1. Treat it as a PP, i.e., a case relation headed by the separate pronoun
   
   
   2. Treat the separate pronoun as a special emphatic modifier (e.g. whatever we do for you __yourself__)
   
   
2. How to analyze the preposition?
   
   
   
   1. Multiword token analysis (ADP+PRON), like in Arabic and Hebrew
      
      
      
      
      
      • Straightforward with 1(ii): separate pronoun is modifier of internal pronoun
      
      
      
      • With 1(i), we have doubling of the pronoun—does it form a phrase, and if so, which is pronoun the head of the other?
      
      
      
   
   
   2. Single-word ADP token with pronominal features, no internal case relation
      
      
      
      
      
      • Always with 1(ii), or when there is no separate pronoun with 1(i), ADP is promoted to obl as it has no overt object
      
      
      
   
   
   3. (goes with 1(i)): (ii) if there is no separate pronoun, (i) otherwise
   
   

I don't love 2(iii) because it makes tokenization context-dependent.

Yet another option is to treat the inflected preposition instead as a case-marked pronoun, with the case specified as a morphological feature (as @ftyers noted). But I take it this construction applies to prepositions more widely than just (say) Accusative, Dative, and Genitive prepositions. I would prefer to keep feature values as abstract morphosyntactic categories rather than lexical items.

It seems important to decide whether to go with the "emphatic modifier" analysis (always, or in some dialects). Do non-prepositional pronouns allow such emphasis?

[amir-zeldes]

+1 for options 1(ii) and 2(i), which is also what we do in Coptic. Coptic 'double' cases are also movable/sometimes floating quantifiers (things like "for-you all-of-you"); we've used 1(ii) partly because of that, so it seems like the best option if you have similar constructions.

[jheinecke]

1(ii) and 2(i) might work for Breton, where the presence of the overt pronoun is more likely to be emphatic, but in Welsh it is definitely not emphatic. It's more a question of style. In more formal text, the overt pronoun is more likely to be dropped, in less formal text it is not dropped (apart form some fixed expressions). So even though I think 1(ii) and 2(i) is the best options, The annotation should not imply any emphasis. So how an phrase like _aros amdanat ti_ "waiting for you(SG)" could be annotated:

1) attaching the overt pronoun to the MWE pronoun:

1   aros    VERB    ....    0   root
2-3 amdanat _   ....    _   _
2   am  ADP ....    3   case
3   ti  PRON    ....    1   obl
4   ti  PRON    ....    3   nmod    // overt pronoun

or 2) attaching the overt pronoun as (second) obl to the verb:

1   aros    VERB    ....    0   root
2-3 amdanat _   ....    _   _
2   am  ADP ....    3   case
3   ti  PRON    ....    1   obl
4   ti  PRON    ....    1   obl // overt pronoun

@nschneid I could find examples of _you yourself_ in the English treebanks, so what exactly you thought for this case. @amir-zeldes could you point me to an exemple in the Coptic treebank ? Thanks a lot!

[amir-zeldes]

Sure, there's lots of them all over the place. We basically have two analyses, depending on whether it's just a dislocated mention of the same pronoun (not adding an argument/adjunct) or if it's adding some flavor, like a quantifier or other modifier (which we don't see as just dislocation, since it has a separate function):

• Floating "on-them all-of-them" - note "on-them" is ejn + ntoou which fuses into ejo:u, and "all-of-them" is a possessed quantifier, lit. "their-all", which is te:r + ntoou = te:rou. Both are dependents of the verb (token 51) in this example: https://github.com/UniversalDependencies/UD_Coptic-Scriptorium/blob/master/cop_scriptorium-ud-dev.conllu#L6313-L6318
• Dislocated full pronoun on top of a fused PP pronoun: ntof is the extra full pronoun 'he' (emphatic or topicalized), and then in this example we also get "to him", which is the mention mandated by the argument structure of "come upon", so "him (ntof), they came upon him (ero+f)", where erof = e+ntof "to him". Example: https://raw.githubusercontent.com/UniversalDependencies/UD_Coptic-Scriptorium/master/cop_scriptorium-ud-train.conllu (look at lines 27121-27130, github is not letting me hyperlink it)

[nschneid]

In EWT the emphatic reflexives like "they themselves" are nmod:npmod (though the npmod subtype is problematic in general and should probably be revised).

[jheinecke]

Thanks for the English and Coptic examples. Hower I couldn't find a solution which fits the Celtic languages. Would expl fit?

Welsh _arhosodd amdanat [ti]_ "He waited for you(SG)"

1   arhosodd    VERB    0   root
2-3 amdanat     _   _   _
2   am      ADP case    3
3   ti      PRON    obl 1
[4  ti      PRON    expl    1]

The advantage is that the structure of the (remaining) phrase is identical with or without overt pronoun.
The only case where this does not work are be a isolated inflected pronouns + overt pronoun:
_ganddo fo!_ "with him"

1-2 ganddo  _   _   _
1   gan ADP case    1
2   fo  PRON    root    0
3   fo  PRON    expl    ??

Back to Breton:
@ftyers More examples of inflected prepositions + pronoun are also mentioned in the _Atlas linguistique de la Basse-Bretagne_(ALBB, http://sbahuaud.free.fr/ALBB/), sheets 106, 107, 208-214, 159, 320 especially for the Leon dialect. At least some dialects have the pronoun frequently (see also Francis Favereau (1997): _Grammaire du breton contemporain_, § 766).

[amir-zeldes]

I think dislocated fits both of the cases above, unless I'm missing some nuance about the meaning of the construction. It doesn't add an argument, so the extra pronoun is a postposed realization which does not itself fulfill one of the core grammatical relations in the sentence, as in the relation description: https://universaldependencies.org/u/dep/dislocated.html

[ftyers]

I quite like dislocated for this as well. I think it is (or should be?) also used for "clitic" doubling in Romance languages, e.g. a mí me gusta, le dio a ella etc. Although it seems that this isn't exactly followed at least in the Spanish treebanks, https://github.com/UniversalDependencies/UD_Spanish-GSD/issues/6 https://github.com/UniversalDependencies/UD_Spanish-AnCora/issues/2 and a post from 2014 (has it been that long!) suggests using expl for this: https://github.com/UniversalDependencies/docs/issues/129

[jnivre]

I completely agree. Using dislocated + a core relation is the preferred solution for non-obligatory cases of "clitic doubling".

[sylvainkahane]

I'm not sure it is comparable to clitic doubling. In case of clitic doubling, the clitic can double any NP. In this case, as Johannes confirmed me, the "doubling" occurs only with pronouns. In other words we have only three possible configurations:

ADP NOUN
ADP=PRON ("=" means cliticization)
ADP=PRON1 PRON2

"ADP=PRON NOUN" is not allowed. Moreover "ADP=PRON1 PRON2" forms a phrase and we cannot consider that ADP=PRON1 and PRON2 occupy two different positions.

It seems to be a very local phenomenon (what we call microsyntax or even morphosyntax) compared to clitic doubling which generally comes from macrosyntactic phenomena such as dislocation. I think we can consider that's a reduplication of the pronoun due to the weakness of the pronoun realization as a clitic. What I suggest is:

PRON1 -case-> ADP
PRON1 -compound:redup-> PRON2

[ftyers]

(An aside @jheinecke I've fixed the stuff with c'h and in passing fixed some other weird tokenisation stuff: https://github.com/UniversalDependencies/UD_Breton-KEB/commit/4daa27b974e67991ff40d4e402e70a01fbe78d05, thanks for the bug report!)