This is part of archived UD v1 documentation. See http://universaldependencies.org/ for the current version.

home v2/v2 issue tracker

Copula in UD v2

The treatment of copula constructions (non-verbal intransitive predication) is quite diverse in the current version of the treebanks (see table below for the status quo). In order to provide more concrete guidelines and to achieve better consistency cross-lingually and within a single language, we propose the following changes:

• We should be maximally restrictive with respect to which words can be copulas (only one word in most languages)
• The copula word should never be the root, except through promotion (“he is not happy, but she is”)
• When there is more than one possible candidate head, the rules to establish it should be determined on a language-specific basis
• We should add the subtype nsubj:cop to signal that the subject in copula constructions is special, and to partially solve the problem of having to flip dependencies when the predicate is a clause (see below)

Problems with the current copula analysis

The main problem is the lack of standardisation. Leaving aside the Galician example, which appears to be a conversion error, the Spanish treebank has over 229 verbs with the cop relation, where the Swedish treebank has one.

edit

  

    
input
    
brat
    
info
  
Éste quedó sorprendido . \n He was/stayed surprised
cop(sorprendido, quedó)
nsubj(sorprendido, Éste)
{
    "text": "Éste quedó sorprendido . \n He was surprised ",
    "entities": [
        [ "embedded-1-T1", "token", [ [ 0, 4 ] ] ],
        [ "embedded-1-T2", "token", [ [ 5, 10 ] ] ],
        [ "embedded-1-T3", "token", [ [ 11, 22 ] ] ],
        [ "embedded-1-T4", "token", [ [ 23, 24 ] ] ],
        [ "embedded-1-T6", "token", [ [ 27, 29 ] ] ],
        [ "embedded-1-T7", "stayed", [ [ 30, 33 ] ] ],
        [ "embedded-1-T8", "token", [ [ 34, 43 ] ] ]
    ],
    "relations": [
        [ "embedded-1-R1", "cop", [ [ "arg1", "embedded-1-T3" ], [ "arg2", "embedded-1-T2" ] ] ],
        [ "embedded-1-R2", "nsubj", [ [ "arg1", "embedded-1-T3" ], [ "arg2", "embedded-1-T1" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "1" ]
}

edit

  

    
input
    
brat
    
info
  
Han blev överraskad . \n He was/became surprised
nsubj(blev, Han)
xcomp(blev, överraskad)
{
    "text": "Han blev överraskad . \n He was surprised ",
    "entities": [
        [ "embedded-2-T1", "token", [ [ 0, 3 ] ] ],
        [ "embedded-2-T2", "token", [ [ 4, 8 ] ] ],
        [ "embedded-2-T3", "token", [ [ 9, 19 ] ] ],
        [ "embedded-2-T4", "token", [ [ 20, 21 ] ] ],
        [ "embedded-2-T6", "token", [ [ 24, 26 ] ] ],
        [ "embedded-2-T7", "became", [ [ 27, 30 ] ] ],
        [ "embedded-2-T8", "token", [ [ 31, 40 ] ] ]
    ],
    "relations": [
        [ "embedded-2-R1", "nsubj", [ [ "arg1", "embedded-2-T2" ], [ "arg2", "embedded-2-T1" ] ] ],
        [ "embedded-2-R2", "xcomp", [ [ "arg1", "embedded-2-T2" ], [ "arg2", "embedded-2-T3" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "2" ]
}

Treebanks differ in if they treat the PP/case-marked nominal as head, in Swedish it is head, while in Finnish it is dependent:

edit

  

    
input
    
brat
    
info
  
Hon är i huset
nsubj(huset, Hon)
cop(huset, är)
{
    "text": "Hon är i huset ",
    "entities": [
        [ "embedded-3-T1", "token", [ [ 0, 3 ] ] ],
        [ "embedded-3-T2", "token", [ [ 4, 6 ] ] ],
        [ "embedded-3-T3", "token", [ [ 7, 8 ] ] ],
        [ "embedded-3-T4", "token", [ [ 9, 14 ] ] ]
    ],
    "relations": [
        [ "embedded-3-R1", "nsubj", [ [ "arg1", "embedded-3-T4" ], [ "arg2", "embedded-3-T1" ] ] ],
        [ "embedded-3-R2", "cop", [ [ "arg1", "embedded-3-T4" ], [ "arg2", "embedded-3-T2" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "3" ]
}

edit

  

    
input
    
brat
    
info
  
Se on talossa
nsubj(on, Se)
nmod(on, talossa)
{
    "text": "Se on talossa ",
    "entities": [
        [ "embedded-4-T1", "token", [ [ 0, 2 ] ] ],
        [ "embedded-4-T2", "token", [ [ 3, 5 ] ] ],
        [ "embedded-4-T3", "token", [ [ 6, 13 ] ] ]
    ],
    "relations": [
        [ "embedded-4-R1", "nsubj", [ [ "arg1", "embedded-4-T2" ], [ "arg2", "embedded-4-T1" ] ] ],
        [ "embedded-4-R2", "nmod", [ [ "arg1", "embedded-4-T2" ], [ "arg2", "embedded-4-T3" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "4" ]
}

There are also inconsistencies within a language, for example the existential construction with copula in English:

edit

  

    
input
    
brat
    
info
  
There is a book on the table .
expl(is, There)
nsubj(is, book)
{
    "text": "There is a book on the table . ",
    "entities": [
        [ "embedded-5-T1", "token", [ [ 0, 5 ] ] ],
        [ "embedded-5-T2", "token", [ [ 6, 8 ] ] ],
        [ "embedded-5-T3", "token", [ [ 9, 10 ] ] ],
        [ "embedded-5-T4", "token", [ [ 11, 15 ] ] ],
        [ "embedded-5-T5", "token", [ [ 16, 18 ] ] ],
        [ "embedded-5-T6", "token", [ [ 19, 22 ] ] ],
        [ "embedded-5-T7", "token", [ [ 23, 28 ] ] ],
        [ "embedded-5-T8", "token", [ [ 29, 30 ] ] ]
    ],
    "relations": [
        [ "embedded-5-R1", "expl", [ [ "arg1", "embedded-5-T2" ], [ "arg2", "embedded-5-T1" ] ] ],
        [ "embedded-5-R2", "nsubj", [ [ "arg1", "embedded-5-T2" ], [ "arg2", "embedded-5-T4" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "5" ]
}

Compared to the bare copula:

edit

  

    
input
    
brat
    
info
  
A book is on the table .
nsubj(table, book)
cop(table, is)
{
    "text": "A book is on the table . ",
    "entities": [
        [ "embedded-6-T1", "token", [ [ 0, 1 ] ] ],
        [ "embedded-6-T2", "token", [ [ 2, 6 ] ] ],
        [ "embedded-6-T3", "token", [ [ 7, 9 ] ] ],
        [ "embedded-6-T4", "token", [ [ 10, 12 ] ] ],
        [ "embedded-6-T5", "token", [ [ 13, 16 ] ] ],
        [ "embedded-6-T6", "token", [ [ 17, 22 ] ] ],
        [ "embedded-6-T7", "token", [ [ 23, 24 ] ] ]
    ],
    "relations": [
        [ "embedded-6-R1", "nsubj", [ [ "arg1", "embedded-6-T6" ], [ "arg2", "embedded-6-T2" ] ] ],
        [ "embedded-6-R2", "cop", [ [ "arg1", "embedded-6-T6" ], [ "arg2", "embedded-6-T3" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "6" ]
}

We also do not provide a consistent analysis when one side of the copula is a clause:

edit

  

    
input
    
brat
    
info
  
The important thing is to keep calm
nsubj(is, thing)
ccomp(is, keep)
{
    "text": "The important thing is to keep calm ",
    "entities": [
        [ "embedded-7-T1", "token", [ [ 0, 3 ] ] ],
        [ "embedded-7-T2", "token", [ [ 4, 13 ] ] ],
        [ "embedded-7-T3", "token", [ [ 14, 19 ] ] ],
        [ "embedded-7-T4", "token", [ [ 20, 22 ] ] ],
        [ "embedded-7-T5", "token", [ [ 23, 25 ] ] ],
        [ "embedded-7-T6", "token", [ [ 26, 30 ] ] ],
        [ "embedded-7-T7", "token", [ [ 31, 35 ] ] ]
    ],
    "relations": [
        [ "embedded-7-R1", "nsubj", [ [ "arg1", "embedded-7-T4" ], [ "arg2", "embedded-7-T3" ] ] ],
        [ "embedded-7-R2", "ccomp", [ [ "arg1", "embedded-7-T4" ], [ "arg2", "embedded-7-T6" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "7" ]
}

Copula constructions in UDv2

For language-specific examples, see below, but here is a summary:

Nominals

The structure wil remain the same, but the relation will be changed to nsubj:cop:

edit

  

    
input
    
brat
    
info
  
Ivan is the best dancer
nsubj:cop(dancer, Ivan)
cop(dancer, is)
{
    "text": "Ivan is the best dancer ",
    "entities": [
        [ "embedded-8-T1", "token", [ [ 0, 4 ] ] ],
        [ "embedded-8-T2", "token", [ [ 5, 7 ] ] ],
        [ "embedded-8-T3", "token", [ [ 8, 11 ] ] ],
        [ "embedded-8-T4", "token", [ [ 12, 16 ] ] ],
        [ "embedded-8-T5", "token", [ [ 17, 23 ] ] ]
    ],
    "relations": [
        [ "embedded-8-R1", "nsubj:cop", [ [ "arg1", "embedded-8-T5" ], [ "arg2", "embedded-8-T1" ] ] ],
        [ "embedded-8-R2", "cop", [ [ "arg1", "embedded-8-T5" ], [ "arg2", "embedded-8-T2" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "8" ]
}

edit

  

    
input
    
brat
    
info
  
Bill is honest
nsubj:cop(honest, Bill)
cop(honest, is)
{
    "text": "Bill is honest ",
    "entities": [
        [ "embedded-9-T1", "token", [ [ 0, 4 ] ] ],
        [ "embedded-9-T2", "token", [ [ 5, 7 ] ] ],
        [ "embedded-9-T3", "token", [ [ 8, 14 ] ] ]
    ],
    "relations": [
        [ "embedded-9-R1", "nsubj:cop", [ [ "arg1", "embedded-9-T3" ], [ "arg2", "embedded-9-T1" ] ] ],
        [ "embedded-9-R2", "cop", [ [ "arg1", "embedded-9-T3" ], [ "arg2", "embedded-9-T2" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "9" ]
}

edit

  

    
input
    
brat
    
info
  
She is in shape
nsubj:cop(shape, She)
cop(shape, is)
{
    "text": "She is in shape ",
    "entities": [
        [ "embedded-10-T1", "token", [ [ 0, 3 ] ] ],
        [ "embedded-10-T2", "token", [ [ 4, 6 ] ] ],
        [ "embedded-10-T3", "token", [ [ 7, 9 ] ] ],
        [ "embedded-10-T4", "token", [ [ 10, 15 ] ] ]
    ],
    "relations": [
        [ "embedded-10-R1", "nsubj:cop", [ [ "arg1", "embedded-10-T4" ], [ "arg2", "embedded-10-T1" ] ] ],
        [ "embedded-10-R2", "cop", [ [ "arg1", "embedded-10-T4" ], [ "arg2", "embedded-10-T2" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "10" ]
}

When there are more than one PP, the head should be the least oblique argument/modifier according to relevant language-specific tests. For example:

edit

  

    
input
    
brat
    
info
  
She was in Prague on Tuesday
nsubj:cop(Prague, She)
{
    "text": "She was in Prague on Tuesday ",
    "entities": [
        [ "embedded-11-T1", "token", [ [ 0, 3 ] ] ],
        [ "embedded-11-T2", "token", [ [ 4, 7 ] ] ],
        [ "embedded-11-T3", "token", [ [ 8, 10 ] ] ],
        [ "embedded-11-T4", "token", [ [ 11, 17 ] ] ],
        [ "embedded-11-T5", "token", [ [ 18, 20 ] ] ],
        [ "embedded-11-T6", "token", [ [ 21, 28 ] ] ]
    ],
    "relations": [
        [ "embedded-11-R1", "nsubj:cop", [ [ "arg1", "embedded-11-T4" ], [ "arg2", "embedded-11-T1" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "11" ]
}

The omission test could be used:

• She was in Prague
• *She was on Tuesday

Only in cases where no tests apply should we resort to general heuristics such as “closest to the copula” and so on:

edit

  

    
input
    
brat
    
info
  
She was in Prague with her friends
nsubj:cop(Prague, She)
{
    "text": "She was in Prague with her friends ",
    "entities": [
        [ "embedded-12-T1", "token", [ [ 0, 3 ] ] ],
        [ "embedded-12-T2", "token", [ [ 4, 7 ] ] ],
        [ "embedded-12-T3", "token", [ [ 8, 10 ] ] ],
        [ "embedded-12-T4", "token", [ [ 11, 17 ] ] ],
        [ "embedded-12-T5", "token", [ [ 18, 22 ] ] ],
        [ "embedded-12-T6", "token", [ [ 23, 26 ] ] ],
        [ "embedded-12-T7", "token", [ [ 27, 34 ] ] ]
    ],
    "relations": [
        [ "embedded-12-R1", "nsubj:cop", [ [ "arg1", "embedded-12-T4" ], [ "arg2", "embedded-12-T1" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "12" ]
}

and:

edit

  

    
input
    
brat
    
info
  
She was with her friends in Prague
nsubj:cop(friends, She)
{
    "text": "She was with her friends in Prague ",
    "entities": [
        [ "embedded-13-T1", "token", [ [ 0, 3 ] ] ],
        [ "embedded-13-T2", "token", [ [ 4, 7 ] ] ],
        [ "embedded-13-T3", "token", [ [ 8, 12 ] ] ],
        [ "embedded-13-T4", "token", [ [ 13, 16 ] ] ],
        [ "embedded-13-T5", "token", [ [ 17, 24 ] ] ],
        [ "embedded-13-T6", "token", [ [ 25, 27 ] ] ],
        [ "embedded-13-T7", "token", [ [ 28, 34 ] ] ]
    ],
    "relations": [
        [ "embedded-13-R1", "nsubj:cop", [ [ "arg1", "embedded-13-T5" ], [ "arg2", "embedded-13-T1" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "13" ]
}

Clausals

When there is a clausal predicate, then we make the head of that the head of the whole copula clause:

edit

  

    
input
    
brat
    
info
  
The important thing is to keep calm
nsubj:cop(keep, thing)
cop(keep, is)
xcomp(keep, calm)
{
    "text": "The important thing is to keep calm ",
    "entities": [
        [ "embedded-14-T1", "token", [ [ 0, 3 ] ] ],
        [ "embedded-14-T2", "token", [ [ 4, 13 ] ] ],
        [ "embedded-14-T3", "token", [ [ 14, 19 ] ] ],
        [ "embedded-14-T4", "token", [ [ 20, 22 ] ] ],
        [ "embedded-14-T5", "token", [ [ 23, 25 ] ] ],
        [ "embedded-14-T6", "token", [ [ 26, 30 ] ] ],
        [ "embedded-14-T7", "token", [ [ 31, 35 ] ] ]
    ],
    "relations": [
        [ "embedded-14-R1", "nsubj:cop", [ [ "arg1", "embedded-14-T6" ], [ "arg2", "embedded-14-T3" ] ] ],
        [ "embedded-14-R2", "cop", [ [ "arg1", "embedded-14-T6" ], [ "arg2", "embedded-14-T4" ] ] ],
        [ "embedded-14-R3", "xcomp", [ [ "arg1", "embedded-14-T6" ], [ "arg2", "embedded-14-T7" ] ] ]
    ],
    "attributes": [],
    "error": false,
    "sentlabels": [ "14" ]
}

We distinguish copula subjects from non-copula subjects, so that when there is a clausal we do not get a double subject:

The main thing is that the device works
nsubj:cop(works, thing)
nsubj(works, device)
cop(works, is)

To discuss

However, we still get duplication of the cop relation where you have a copula on the right:

To be free is to be capable of thinking one's own thoughts
nsubj:cop(capable, free)
cop(capable, is-4)
cop(capable, be-6)
cop(free, be-2)

And in the case of having an expressed subject, we get two subjects for the main predicate:

The problem is that she is not happy .
nsubj:cop(happy, problem)
nsubj:cop(happy, she)
cop(happy, is-3)
cop(happy, is-6)

Language-specific examples

For the purposes of demonstrating the new classification system a number of examples have been prepared for a range of UD languages. The examples are in English, but where they are ambiguous in a given language multiple variants will be given.

1. She is a student
2. I am a student
3. She was a student
4. I was a student
5. She is happy
6. I am happy
7. She is in shape
8. She is in the house
9. I am in the house
10. She was in the house
11. There is a house in the village
12. The house is in the village
13. There was a house in the village
14. The house was in the village

English

The English analysis more or less follows the analysis in the UD_English treebank, with the addition of the relation nsubj:cop for subjects of copula constructions. There is a difference however with how (11) and (13) are treated.

(1)

She is a student
nsubj:cop(student, She)
cop(student, is)

(2)

I am a student
nsubj:cop(student, I)
cop(student, am)

(3)

She was a student
nsubj:cop(student, She)
cop(student, was)

(4)

I was a student
nsubj:cop(student, I)
cop(student, was)

(5)

She is happy
nsubj:cop(happy, She)
cop(happy, is)

(6)

I am happy
nsubj:cop(happy, I)
cop(happy, am)

(7)

She is in shape
nsubj:cop(shape, She)
cop(shape, is)

(8)

She is in the house
nsubj:cop(house, She)
cop(house, is)

(9)

I am in the house
nsubj:cop(house, I)
cop(house, am)

(10)

She was in the house
nsubj:cop(house, She)
cop(house, was)

(11)

There is a house in the village
expl(village, There)
nsubj:cop(village, house)
cop(village, is)

(12)

The house is in the village
nsubj:cop(village, house)
cop(village, is)

(13)

There was a house in the village
expl(village, There)
nsubj:cop(village, house)
cop(village, was)

(14)

The house was in the village
nsubj:cop(village, house)
cop(village, was)

Swedish

(1)

Hon är en student
nsubj:cop(student, Hon)
cop(student, är)

(2)

Jag är en student
nsubj:cop(student, Jag)
cop(student, är)

(3)

Hon var en student
nsubj:cop(student, Hon)
cop(student, var)

(4)

Jag var en student
nsubj:cop(student, Jag)
cop(student, var)

(5)

Hon är glad
nsubj:cop(glad, Hon)
cop(glad, är)

(6)

Jag är glad
nsubj:cop(glad, Jag)
cop(glad, är)

(7)

Example needed

(8)

Hon är i huset
nsubj:cop(huset, Hon)
cop(huset, är)

(9)

Jag är i huset
nsubj:cop(huset, Jag)
cop(huset, är)

(10)

Hon var i huset
nsubj:cop(huset, Hon)
cop(huset, var)

(11)

Existential constructions in Swedish do not use the copula verb.

Det finns et hus i byn
expl(finns, Det)
nsubj(finns, hus)
nmod(finns, byn)

(12)

Huset är i byn
nsubj:cop(byn, Huset)
cop(byn, är)

(13)

Det fanns et hus i byn
expl(fanns, Det)
nsubj(fanns, hus)
nmod(fanns, byn)

(14)

Huset var i byn
nsubj:cop(byn, Huset)
cop(byn, var)

Spanish

The UD_Spanish treebank has very many verbs classified as copula. We propose reducing it to the single verb “ser”.

(1)

Ella es estudiante
nsubj:cop(estudiante, Ella)
cop(estudiante, es)

(2)

Yo soy estudiante
nsubj:cop(estudiante, Yo)
cop(estudiante, soy)

(3)

Ella fue estudiante
nsubj:cop(estudiante, Ella)
cop(estudiante, fue)

(4)

Yo fui estudiante
nsubj:cop(estudiante, Yo)
cop(estudiante, fui)

(5)

In Spanish you can say either Soy feliz “I am happy” or Estoy feliz “I am happy”/”I feel happy”. In the following examples, the subject pronouns are expressed to illustrate the difference in relation for the subject. They may equally well be dropped.

Ella es feliz
nsubj:cop(feliz, Ella)
cop(feliz, es)

Ella está feliz
nsubj(está, Ella)
xcomp(está, feliz)

(6)

Yo soy feliz
nsubj:cop(feliz, Yo)
cop(feliz, soy)

Yo estoy feliz
nsubj(estoy, Yo)
xcomp(estoy, feliz)

(7)

Instead of “in shape” we’ll use “de puta madre” which means “really great”,

Esta canción es de puta madre
nsubj:cop(madre, canción)
cop(madre, es)

Esta canción está de puta madre
nsubj(está, canción)
xcomp(está, madre)

(8)

In Spanish location/position uses the verb estar and not ser.

Ella está en la casa
nsubj(está, Ella)
nmod(está, casa)

Note that in Catalan, this would be “Ella és a la casa”, using the ser verb, not the estar verb. This would be analysed as:

Ella és a la casa
nsubj:cop(casa, Ella)
cop(casa, és)

(9)

Yo estoy en la casa
nsubj(estoy, Yo)
nmod(estoy, casa)

(10)

Ella estaba en la casa
nsubj(estaba, Ella)
nmod(estaba, casa)

(11)

Existential constructions in Spanish do not use the copula verb.

Hay una casa en el pueblo
obj:dir(Hay, casa)
nmod(Hay, pueblo)

(12)

La casa está en el pueblo
nsubj(está, casa)
nmod(está, pueblo)

(13)

Había una casa en el pueblo
obj:dir(Había, casa)
nmod(Había, pueblo)

(14)

La casa estaba en el pueblo
nsubj(estaba, casa)
nmod(estaba, pueblo)

Russian

In Russian, there is no copula verb in the present tense. In the future and past tenses, the verb быть “be” is used. Note that when the copula verb is used, the complement can be either in nominative or instrumental case. When it is instrumental it is is category of and when it is nominative it is more like has quality of. We propose using the same structure for both.

(1)

Она студентка
nsubj:cop(студентка, Она)

(2)

Я студентка
nsubj:cop(студентка, Я)

(3)

Она была студентка
nsubj:cop(студентка, Она)
cop(студентка, была)

Она была студенткой
nsubj:cop(студенткой, Она)
cop(студенткой, была)

(4)

Я была студентка
nsubj:cop(студентка, Я)
cop(студентка, была)

Я была студенткой
nsubj:cop(студенткой, Я)
cop(студенткой, была)

(5)

The same goes with adjectival uses:

Она счастлива
nsubj:cop(счастлива, Она)

(6)

Я счастлива
nsubj:cop(счастлива, Я)

(7)

Instead of “in shape”, we’ll use “в курсе” which means “on the ball”

Она в курсе
nsubj:cop(курсе, Она)

(8)

In Russian, there is no verb used for locative predication in the present tense.

Она в дому
nsubj:cop(дому, Она)

(9)

As with (8),

Я в дому
nsubj:cop(дому, Я)

(10)

In the past tense we have the verb and we make it a dependent:

Она была в дому
nsubj:cop(дому, Она)
cop(дому, была)

(11)

In Russian, in the present tense, existential constructions use “есть” which is sometimes described as a “predicative”:

Есть дом в деревне
nsubj:cop(деревне, дом)
cop(деревне, Есть)

(12)

Дом в деревне
nsubj:cop(деревне, Дом)

(13)

In the past tense (and future tense), the verb быть is employed. Syntactically (13) and (14) are equivalent in Russian aside from the word order considerations.

Был дом в деревне
nsubj:cop(деревне, дом)
cop(деревне, Был)

(14)

Дом был в деревне
nsubj:cop(деревне, Дом)
cop(деревне, был)

Finnish

In Finnish the copula verb is olla “to be”. Its complement is typically in the nominative, although it may also be in the essive case -nA.

(1)

Se on opiskelija
nsubj:cop(opiskelija, Se)
cop(opiskelija, on)

(2)

Mä oon opiskelija
nsubj:cop(opiskelija, Mä)
cop(opiskelija, oon)

(3)

Se oli opiskelija
nsubj:cop(opiskelija, Se)
cop(opiskelija, oli)

(4)

Mä olin opiskelija
nsubj:cop(opiskelija, Mä)
cop(opiskelija, olin)

(5)

Se on iloinen
nsubj:cop(iloinen, Se)
cop(iloinen, on)

(6)

Mä oon iloinen
nsubj:cop(iloinen, Mä)
cop(iloinen, oon)

(7)

Se on kunnossa
nsubj:cop(kunnossa, Se)
cop(kunnossa, on)

(8)

Se on talossa
nsubj:cop(talossa, Se)
cop(talossa, on)

(9)

Mä oon talossa
nsubj:cop(talossa, Mä)
cop(talossa, oon)

(10)

Se oli talossa
nsubj:cop(talossa, Se)
cop(talossa, oli)

(11)

In Finnish, existential and non-existential are identical aside from word order.

Kylässä on talo
cop(Kylässä, on)
nsubj:cop(Kylässä, talo)

(12)

Talo on kylässä
cop(kylässä, on)
nsubj:cop(kylässä, Talo)

(13)

Kylässä oli talo
cop(Kylässä, oli)
nsubj:cop(Kylässä, talo)

(14)

Talo oli kylässä
cop(kylässä, oli)
nsubj:cop(kylässä, Talo)

Turkish

In Turkish, there are two copula verbs, i- and ol-. The “true” copula is i- which is defective, only having a limited number of tense forms (aorist and past), and cliticising. When a copula is needed in another tense, ol- is employed. However, if there is a form of i- then the equivalent form of ol- takes on the meaning “become”.

(1)

In the present tense, third person singular aorist non-formal then there is no overt suffix for third person singular. Unlike Russian, where the copula verb does not appear in any part of the present tense paradigm, in Turkish it appears in all persons except third person. This means that it is more like the nominative case in the paradigm (which also has a -Ø suffix, than like the Russian copula).

In the following examples the hyphen is used to separate cliticised syntactic words.

O öğrenci -Ø
nsubj:cop(öğrenci, O)
cop(öğrenci, -Ø)

(2)

Ben öğrenci -yim
nsubj:cop(öğrenci, Ben)
cop(öğrenci, -yim)

(3)

O öğrenci -ydi
nsubj:cop(öğrenci, O)
cop(öğrenci, -ydi)

The copula verb here can also be written separately instead of cliticised in more formal styles,

O öğrenci idi
nsubj:cop(öğrenci, O)
cop(öğrenci, idi)

(4)

Ben öğrenci -ydim
nsubj:cop(öğrenci, Ben)
cop(öğrenci, -ydim)

(5)

O mutlu -Ø
nsubj:cop(mutlu, O)
cop(mutlu, -Ø)

(6)

Ben mutlu -yum
nsubj:cop(mutlu, Ben)
cop(mutlu, -yum)

(7)

Example needed

(8)

O evde
nsubj:cop(evde, O)

(9)

Ben evde -yim
nsubj:cop(evde, Ben)
cop(evde, -yim)

(10)

O evde -ydi
nsubj:cop(evde, O)
cop(evde, -ydi)

(11)

In Turkish (and indeed in most Turkic languages), existence is a syntactically different, using an adjective var “existent”, and so gets a different structure.

Köyde ev var
nsubj:cop(var, ev)
nmod(var, Köyde)

(12)

Ev köyde
nsubj:cop(köyde, Ev)

(13)

Köyde ev var -dı
nsubj:cop(var, ev)
cop(var, -dı)
nmod(var, Köyde)

(14)

Ev köyde -ydi
nsubj:cop(köyde, Ev)
cop(köyde, -ydi)

Irish

Irish has a difference between a copula verb “is” and what is called a substantive verb “bí”. Only the copula verb receives the cop relation. The substantive verb is head and takes an argument with xcomp. Teresa’s thesis has an in depth description of the treatment of the copula in Irish.

(1)

Example needed

(2)

Example needed

(3)

Ba dhalta í
cop(dhalta, Ba)

Bhí sí ina dalta
xcomp:pred(Bhí, dalta)

(4)

Not applicable.

(5)

Not applicable.

(6)

Example needed

(7)

Example needed

(8)

Example needed

(9)

Example needed

(10)

Example needed

(11)

Example needed

(12)

Example needed

(13)

Example needed

(14)

Example needed

Status quo

The languages in UD with the tokens which have the cop relation. If we adopt the above recommendations, the vast majority will need converting.

Treebank	Unique cop	Top-5 lemmas[POS] with cop relation
UD-Galician	1112	121/de[ADP], 40/necesario[ADJ], 38/como[PRON], 24/posible[ADJ], 23/importante[ADJ]
UD-Dutch	253	2491/ben[AUX], 283/word[AUX], 91/vind[VERB], 73/blijf[AUX], 67/maak[VERB]
UD-Spanish	229	5136/ser[VERB], 353/estar[VERB], 78/llamado[VERB], 66/encontrar[VERB], 48/hacer[VERB]
UD-Arabic	216	384/كَان[VERB], 75/لَيس[VERB], 31/عَدّ[VERB], 27/اِعتَبَر[VERB], 25/زَال[VERB]
UD-Portuguese	135	2120/ser[VERB], 370/estar[VERB], 176/como[ADV], 91/ficar[VERB], 38/parecer[VERB]
UD-French	99	4878/être[VERB], 232/devenir[VERB], 91/appeler[VERB], 70/nommer[VERB], 51/rester[VERB]
UD-Greek	67	531/είμαι[VERB], 86/αποτελώ[VERB], 34/θεωρώ[VERB], 27/γίνομαι[VERB], 20/καθίσταμαι[VERB]
UD-Catalan	57	3609/ser[AUX], 810/estar[VERB], 722/ser[VERB], 136/cop[NOUN], 53/semblar[VERB]
UD-Polish	18	764/być[VERB], 98/to[VERB], 42/być[AUX], 17/stać[VERB], 12/stawać[VERB]
UD-Basque	15	1993/izan[VERB], 266/egon[VERB], 124/ukan[VERB], 31/izan[AUX], 20/ibili[VERB]
UD-German	11	4698/-[VERB], 86/-[NOUN], 31/-[ADJ], 27/-[ADP], 23/-[PROPN]
UD-Estonian	9	3373/olema[VERB], 37/ole[VERB], 29/tunduma[VERB], 5/paistma[VERB], 4/näima[VERB]
UD-Czech	6	20480/být[VERB], 110/bývat[VERB], 3/stát[VERB], 3/bývávat[VERB], 1/moci[VERB]
UD-Hungarian	6	92/van[VERB], 61/lesz[VERB], 11/lehet[VERB], 3/marad[VERB], 1/hoz[VERB]
UD-Bulgarian	5	1940/съм[VERB], 3/съм[AUX], 1/стана[VERB], 1/разпространявам-(се)[VERB], 1/докосна-(се)[VERB]
UD-Buryat	5	70/байха[VERB], 22/болохо[VERB], 2/ябаха[VERB], 2/үнгэхэ[VERB], 2/байха[AUX]
UD-Croatian	5	1236/biti[AUX], 1/željeti[VERB], 1/težiti[VERB], 1/davati[VERB], 1/bivati[VERB]
UD-English	4	5593/be[VERB], 8/`s[VERB], 5/be[AUX], 1/’[VERB]
UD-Kazakh	4	131/е[VERB], 42/бол[VERB], 1/тұр[VERB], 1/атан[VERB]
UD-Uyghur	4	66/-[VERB], 4/-[NOUN], 3/-[ADJ], 1/-[PART]
UD-Hindi	3	3014/है[VERB], 497/था[VERB], 1/बशर्ते[SCONJ]
UD-Irish	3	369/is[VERB], 3/is[PART], 1/má[SCONJ]
UD-Russian	3	538/-[VERB], 5/-[NOUN], 1/-[ADP]
UD-Russian-SynTagRus	3	4457/БЫТЬ[AUX], 622/ЭТО[NOUN], 4/ВОТ[PART]
UD-Chinese	2	1795/-[VERB], 8/-[ADJ]
UD-Coptic	2	30/ⲡⲉ[PART], 2/ⲡ[DET]
UD-Danish	2	1576/være[AUX], 185/blive[AUX]
UD-Hebrew	2	387/-[VERB], 7/-[PRON]
UD-Persian	2	4662/-[VERB], 3/-[ADJ]
UD-Turkish	2	751/i[AUX], 113/değil[VERB]
UD-Faroese	1	1081/vera[VERB]
UD-Finnish	1	3279/olla[VERB]
UD-Indonesian	1	1055/-[VERB]
UD-Italian	1	2767/essere[VERB]
UD_Norwegian	1	7217/være[VERB]
UD-Slovenian	1	2820/biti[VERB]
UD-Swedish	1	1629/vara[VERB]
UD-Tamil	1	1/முயல்[VERB]

UD-internal references

• http://universaldependencies.org/u/dep/cop.html
• https://github.com/UniversalDependencies/docs/issues/329
• http://universaldependencies.org/2015-08-23-uppsala/copula.html
• https://github.com/UniversalDependencies/docs/issues/256

Further reading

For wider cross-linguistic applicability, it is well worth looking at the following book:

• Stassen, L. (1997), Intransitive predication. Oxford: OUP

The following publications have also been cited:

• Hengeveld, K. (1992), Non-verbal Predication. Berlin & NewYork: Mouton de Gruyter.
• Katz, A. (1996) Cyclical Grammaticalization and the Cognitive Link between Pronoun and Copula. PhD Thesis, Rice University.
• Pustet, R. (2003), Copulas. Universals in the Categorization of the Lexicon. Oxford: OUP.

© 2014 Universal Dependencies contributors. Site powered by Annodoc and brat

.