bguil / UD-French-discussion

Discussions sur l'harmonisation des Treebanks du français au format UD
0 stars 1 forks source link

Reflexive clitic "se" #9

Open mcandito opened 6 years ago

mcandito commented 6 years ago

(discussion Pise MCDM, SK, MC)

NB:We need to consult other Romance people

Suppose we consider the following cases, we need to decide whether to distinguish cases at the non-enhanced level, or whether to postpone this at the enhanced level

Problem: frontiers are very very difficult to draw, and classification depends on context (for same lemma, can be true reflexive or other case)

MC: some authors distinguish "autocausatives" : e.g. se promener : not really a true reflexive, so would be better to treat like inherent reflexives

NB: in Sequoia and FTB: we do have a fine-frained classification For French UD: I just had a look at the first examples of "se", the current expl / obj distinction is really really noisy

I would be in favor of using a unique label expl for now

What shall we do for version 2.1 ??

sylvainkahane commented 6 years ago

Nous on n'a pas l’information dans Rhapsodie (tous nos clitiques pronominaux sont comp) et donc pour les clitiques pronominaux qui ne peuvent pas être désambiguisés entre obj et iobj, je propose qu’on fasse obj|iobj avec un pipe « | ». On pourrait calculer la différence entre je me rase et il me rase, mais j’ai un peu la flemme.

De manière générale, je pense que c’est bien de se donner la possibilité de sous-spécifier une distinction. Et comme le signalait Bruno pour obl, de ne pas utiliser la même étiquette pour un obl où la distinction arg vs mod est sous-sécifiée et un obl qui n’est pas obl:arg.

Le 16 oct. 2017 à 12:10, Marie Candito notifications@github.com a écrit :

(discussion Pise MCDM, SK, MC)

NB:We need to consult other Romance people

Suppose we consider the following cases, we need to decide whether to distinguish cases at the non-enhanced level, or whether to postpone this at the enhanced level

true reflexives, inalienable parts (should be obj/iobj according to UD)

Paul se soigne => obj Paul se parle => iobj Paul se gratte le bras => iobj inherent reflexive (se suicider) ==> expl (MC why not fixed ?)

mediopassive (ces livres se vendent bien) and neuter (la branche s'est cassée sous le poids de l'enfant) ==> expl:refl or aux:refl (Sylvain points that aux is coherent, cf. can be used to mark voice markers)? Note: expl:pass in Italian UD

Problem: frontiers are very very difficult to draw, and classification depends on context (for same lemma, can be true reflexive or other case)

MC: some authors distinguish "autocausatives" : e.g. se promener : not really a true reflexive, so would be better to treat like inherent reflexives

NB: in Sequoia and FTB: we do have a fine-frained classification For French UD: I just had a look at the first examples of "se", the current expl / obj distinction is really really noisy

I would be in favor of using a unique label expl for now

What shall we do for version 2.1 ??

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/bguil/UD-French-discussion/issues/9, or mute the thread https://github.com/notifications/unsubscribe-auth/AV--FB1vmy3SfBOz52DhPHZE8pwzFKcVks5ssyuwgaJpZM4P6Yt1.

perrier54 commented 6 years ago

Sur la désambiguisation OBJ / EXPL ou IOBJ / EXPL, je pense que quand on peut le faire, il faut le faire. Quand c'est problématique, mettons EXPL par défaut. Nous nous trouvons confrontés à un problème de frontière que nous avons déjà retrouvé quand il s'agissait de distinguer compléments régis et modificateurs. A la frontière, nous avions choisi plutôt modificateur par défaut. La proposition de Sylvain n'est pas compatible avec le format UD donc nous ne pouvons pas la retenir.