expletive

[kimgerdes]

The current definition of expletives states

Expletives "are nominals that appear in an argument position of a predicate but which do not themselves satisfy any of the semantic roles of the predicate. "

This definition is nonoperational.
What is an argument that does not have a semantic role?

the Wikipedia definition seems better:

A syntactic expletive is a word that performs a syntactic role but contributes nothing to meaning.

It is still questionable whether expletive is a syntactic dependency relation and should thus be part of the functional tagset.

It seems clear that our definitions should exclude Idiomatic "arguments" like "ice" in "break the ice" (currently they seem to fall in the definition) or

la moutarde me monte au nez
the mustard to_me climbs to_the nose
i'm getting angry

Now, if pronouns are similarly part of an idiomatic structure, should they be more "expletive"?
For example the pronouns in

s'en aller 'leave'
Il s'en va de chez lui.
He leaves from his home. 

'se' and 'en' are obligatory.

* Il va de chez lui
* Il se va de chez lui
* Il en va de chez lui

se souvenir: Il se souvient bien de ça.
remember: He remembers that well.

il y a: Il y a un lapin dans le jardin.
locative be: there is a rabbit in the garden.

it rains.

Just like the nouns in the idiomatic expression the meaning is non-compositionally constructed from the different words of the expression.

So if we call "ice" an object, we don't we call "se" in "se souvenir" an object and "en" an iobj in "s'en aller" and "y" an advmod in "il y a "? If we call "moutarde" a subject, why do we want to call "it" a subject in "it rains" or "il" in "il y a"?

So should the category "pronoun" be part of the definition of the expl relation?

Even this seems to be insufficient to clearly distinguish expletives from other (semi-) idiomatic constructions. Take

ça craint
it sucks/this sucks

as an example. Is this an expletive? It's a pronoun but does it participate in the meaning? If it refers to a specific situation, it's a referential pronoun.
Weather pronouns are similar :

il/ça pleut dans mon salon.
it rains in my living room.

In our observations of the current state of the English and French treebanks we noticed an important heterogeneity of the expl relation.

The first example on the general expl page is http://universaldependencies.org/u/dep/expl.html
There is a ghost in the room
We wonder why "is" is not a copula here?
Isn't "is" a copula in the following sentence?
A ghost is in the room.
We don't really understand the conclusion of #170

Even the only example given in the French expl page contains a referential c' / it which is clearly not an expletive under any definition.
http://universaldependencies.org/fr/dep/expl.html

C'est la seule manière de réussir
It's the only way to succeed.

Note also example 3 where a euphonic "t" does not fill any syntactic position and example 4 on this page where an "expl" pronoun is modified by a relative clause.

The usage of expl to refer to impersonal pronouns in quasi-subject and quasi-object constructions leads to contradictions, too.

c'est chouette que Pierre vienne (c'=expl)
it is great that Peter comes

Que Pierre vienne, c'est chouette (c'=nsubj)
That Peter comes, It's great

Similar problems arise with cleft and in particular pseudo-cleft sentences which we don't discuss here. See also UniversalDependencies/UD_Danish-DDT#13

In our Spoken French treebank, we plan to use expl only as a subrelation for impersonal constructions, as in

Il est arrivé trois personnes.
three people have arrived.
arrivé -subj:expl-> il
arrivé -nsubj:quasi-> personnes

(because this is a redistribution similar to the use of nsubj:pass for the passive)
and to otherwise keep nominal and pronominal idiomatic arguments as such (nsubj, obj, ...)

il <nsubj- pleut
it rains

la moutarde me monte au nez.
moutarde <nsubj- monte

il se <obj- souvient.
he remembers.

[jnivre]

It is possible that the definition can be improved, but I would still like to defend it against the alternative. Note that the definition does not talk about an argument without a semantic role but about a "nominal in argument position" without a semantic role. This essentially restricts the notion of expletive to subject and object positions (if we interpret "argument" as "core argument"). Restricting it to pronouns (or other function words like English "there") seems like a good idea, and perhaps "non-referential" is better than lacking a semantic role.

[kimgerdes]

Thanks, Joakim, for your answer!
Let me put this into yes/no questions, to understand better what we agree on.

• So do we agree that "Peter" in "Peter seems to sleep" is a nominal in an argument position of "seems"?

• Do we agree that "Peter" does not satisfy any of the semantic roles of the predicate "seems"?

• Do we agree that the two following sentences have the same "is" (that should be either a copula or something else - i still haven't understood the udv2 decision)?

There is a ghost in the room.
A ghost is in the room.

• Do we agree that the "c'" should have the same syntactic role in the following sentences?
  c'est chouette que Pierre vienne (currently c'=expl)
  it is great that Peter comes
  Que Pierre vienne, c'est chouette (currently c'=nsubj)
  That Peter comes, It's great

• Do we agree that the only French example is actually not an expletive and the C' should be annotated as nsubj (of manière)?

C'est la seule manière de réussir
It's the only way to succeed.

It is important to point out that this is not a personal problem of us not understanding the guide, but it's a problem of the different treebank annotators and transformers who are having difficulties applying the expl rules. Here, for example, you find a list that shows which function is given to the "se" in the different French corpora, depending on the governing verb: https://gerdes.fr/papiers/2017/segovs.tsv
You can see for example that the "se" argument of the verb "agir" is annotated either as expl or as obj, depending on the corpus: fr annotates "se" as "obj" and sequioia as expl.
The expl / obj distinction makes the annotation of the pronoun "se" extremely difficult and the current annotation of "se" is unexploitable.

[dan-zeman]

Hi Kim, my answers to your questions would be:
yes - no - yes - yes(nsubj) - yes

[jnivre]

For me, the first and last are a clear "yes". The others need more discussion, but I don't have time right now.

[kimgerdes]

Hi guys,
we believe in yes yes yes yes yes

Second point:

• Do we agree that "Peter" does not satisfy any of the semantic roles of the predicate "seems"?
  The standard definition of raising verbs such as "seem" is that

raising predicate/verb appears with a syntactic argument that is not its semantic argument
https://en.wikipedia.org/wiki/Raising_(linguistics)

Third point:
ok, great. we agree that "there" is an expletive and that the rest of the analysis should not change depending on the presence of "there".

Fourth point:
if we agree on what Dan said, all quasi subject and quasi object constructions would jump out of the expl domain. and example 5 on the expl page should be analyzed differently http://universaldependencies.org/u/dep/expl.html

Fifth point: we will have to agree on where actual expletives exist in French. For the moment, we would like to restrain the expl relation to impersonal constructions and exclude it for core arguments of idiomatic expressions (break the ice, it rains, it's good to x, do it yourself, to hell with it, to make it impossible to x, se souvenir 'remember', ...)

[jnivre]

I am fine with "yes" to question 2 after the clarification.

The reason I am hesitating about 3 is that this is language-specific. For Swedish, the closest correspondence to "there is a ghost in the room" would use a verb of existence that is definitely not a copula. Whether the verb is the same or not in English would have to be settled using language-specific criteria (but in accordance with the universal guidelines for nonverbal constructions).

The reason I am hesitating about 4 is similar. In Swedish, the two most similar constructions would not get the same analysis. The first has an expletive, which is required on syntactic grounds and cannot be omitted. The second, has a presumptive pronoun, which can be omitted. But maybe the corresponding tests for French say that they are the same.

I completely agree with your fifth point concerning the example, but I would prefer a wider definition of expletive that also includes "it rains", "it's good to x" and "to make it impossible to x" (but none of the others), because I think this would be more consistent with my understanding of expletive.

[amir-zeldes]

Thanks for this interesting discussion! I wanted to point out that the there is discussion also needs to include the 'pure existential' example

There is a ghost (maybe found more often with God, or in negation)

If we treat There is a ghost in the room as a copula construction and stick with the same analysis for the sentence above, then it becomes something like "Something is a ghost" (subject omitted), whereas we really mean "a ghost exists". In many languages, existentials like these are more similar to presentatives (something like 'voila, a ghost!'), which are not generally analyzed as copulas (I'm thinking of Russian vot/jest', Hebrew hine/yesh, and Coptic is much the same as Hebrew).

On the break the ice example, I would be hesitant to annotate idiomatic lexical items as expl - although I see the semantic motivation in this example, I can imagine it will be difficult to decide in many real life scenarios, where we are not sure of the degree of idiomaticity. Is "take the plunge" an expletive, or maybe there really is some sort of plunge involved? Etc.

[jnivre]

This is precisely my point about "there is a ghost". In many languages, these go together with existentials, rather than copulas. English is tricky because they use the same verb for both. But in French, I guess they would be different:

Il y an fantôme dans la chambre.
Un fantôme est dans la chambre.

Although I seem to remember that in older varieties of French you can use "il est" instead of "il y a".

Using "expl" for "break the ice" is completely out, and Kim has a point that we need to change the definition of "expl" to make this clear.