relatio-nlp/relatio

Deal with multiple named entities in a semantic role?

Opened this issue · 0 comments

Currently, the pipeline only allows for one named entity per role. The entity chosen is the first one to be a subsequence of the semantic role. This does lead to problems.

For instance, "Republicans and Democrats" would likely be assigned either as "republican" or as "democrat". Arguably, in this case, both are relevant named entities.

Several options to "fix" this:

  • Label semantic roles by the most frequent named entity in the set of matched named entities (easy fix)
  • Create multiple narrative tuples. For instance, "Republicans and Democrats increase taxes." would become [("republican","increase","tax"), ("democrat","increase","tax")].