SUD dependency relations

Structure of dependency relations

In SUD, each dependency relation is made of up to three parts:

• a required main relation
• an optional subrelation (prefixed by :)
• an optional deep_feature (prefixed by @)

Some examples of the four possible cases:

• Only main relation: cc
• with a subrelation only: comp:obj
• with a deep feature only: subj@expl
• with both: comp:aux@tense

about mSUD

In mSUD, there is a fourth part called type and prefixed by /

SUD specific main relations

In SUD there are three specific (i.e. not used in UD) main relation: subj, comp and mod.

The schema below shows these relations and their standard extensions (possible combination of subrelations and deep features.

The schema also introduce two underspecified relation specificaly used in SUD:

• udep with stands for a relation which is either a comp or a mod. It is typically used for noun complements for which distinction is not done in most of the cases. See for example in SUD_French-GSD.
• unk with stands for a relation which is unkonwn: no specific dependency could be identified. It is mainly used in annotation of idioms.

Other common relations and extensions

Here is a list of commonly used main relations in SUD, with their common extensions.

• cc
• compound
• conj
  • conj:appos
  • conj:coord
  • conj:dicto
  • conj:reform
• det
• discourse
• dislocated
  • dislocated:mod
  • dislocated:obj
  • dislocated:obl
  • dislocated:subj
  • dislocated:unk
• flat
  • flat@foreign
  • flat@name
• goeswith
• orphan
• parataxis
  • parataxis:insert
  • parataxis:parenth
• punct
• repair
• root
• vocative

Note

In treebanks converted from UD to SUD, other deep extensions may appear. See examples in English treebanks: @desc, @npmod, @poss, @tmodand @unmarked are used to keep track of UD subrelations.