Multiple Antecedents
Opened this issue · 4 comments
An anaphora could have multiple antecedents - also called "Split Anaphora" by Mitkov.
Incorporating this will be tricky but it could be explored.
Say anaphora is plural, and in the context you find "NounX and NounY" then this should be a candidate for being an antecedent.
As mentioned on IRC, I think that the thing to do here is to have the antecedent be compound, what we care about for translation is more the features than the lemma, so in the NounX and NounY case, what I would probably do is have a way to just specify what the antecedent should be, e.g. they<prn><tn><p3><mf><pl>
@ftyers The actual target word can be added in transfer (as it happens for changing "his" to "their"), so yeah maybe we can add a dummy lemma and put the features and then it can be converted into what's needed in the target in transfer.
We could also just specify the lemma we want. Would that be hard?
specifying the target lemma would make the module target language dependent, whereas right now it's basically a monolingual module and can be shared across pairs if needed.