Causatives

[mcandito]

(sorry to reopen the causative cases)

cf. issue : https://gitlab.inria.fr/sequoia/deep-sequoia/issues/153#note_28165

We agreed on the following general representation of French causatives:

the combination "faire"+ Vinf is viewed as a complex predicate. We can treat "faire" as an auxiliary, with all other dependents attached to the infinitive (including the causer argument).

There remains a discussion on whether to add suffixes:

• to mark the causer argument (nsubj:caus)
• and to mark which argument is the canonical subject, using a :agent suffix as in the passive obl:agent case:

Paul fait dormir Pierre => nsubj:caus(dormir, Paul), obj:agent(dormir, Pierre) Paul fait présenter l'exposition à Pierre => ambiguous between obl and obl:agent(présenter, Pierre)

MC: I think that

• the :caus suffix is useless, because it is trivial to find the causer once the "faire" auxiliary is specified as such.
• On the other hand, which argument should bear the :agent suffix is much more complicated ==> I am in favor of treating that in the enhanced representation, using directly a @caus and @nsubj suffix

(but I agree that to be coherent, this should also be done for nsubj:pass, namely marking them as "nsubj" in the non-enhanced rep, and as nsubj@obj in enhanced)

We need to decide with other Romance treebank groups

[mcandito]

Discussion Guy / Bruno / Djamé / Marie:

For release 2.1 :

• ok for using :caus and :agent suffixes for causative constructions, in the basic representation

(cf. unclear whether in 2.1 there will be enhanced representation with neutralization of syntactic alternations)