This relation is used to indicate difluencies when a speaker corrects his speech and to analyse reformulations. The relation
conj:coord is used to link two different objects, two referants (ex: Mary and John are two different referents). The relation
conj:dicto on the other hand is used to link two denotations of the same referent (ex: the desert in Kenya, the Kenya desert is denoting the same referant).
The ability of
conj:dicto to anlyse both disfluencies and reformulations is the reason why we don’t use the the
reparandum relationship as used in UD. For the moment, it is only used in SUD for analysis of written texts.
conj:dicto (SUD) considers the first element as the head of the next one, while
reparandum (UD) considers these constructions as reparations and considers the last element as head.
This is because UD relies more on the semantics - it is the last element that holds the meaning. SUD on the other hand relies on the syntactic aspect and considers that it is the first element that occupies the syntactic position.
This may result in some incoherencies. Note the following example. In this sentence we have two options of annotation. We can annotate the é~ with an unknown POS (
X). In that case
X is the head of a
subj relation which is undesirable (only
AUX should be head of a
subj relation). The other option would be to guess the POS and lemma of the unfinished word with the risk of guessing it wrong. In our case the token é~ would be annotated with the lemma être and POS
AUX. In this scenario, é~ becomes part of the paradigm of the conjugated verb être which is also undesirable.
In SUD we decided to adopt the following analysis.
For more examples on disfluencies, you can refer to this page.