Build a database of MWE (using Wikidata, ConceptNet...)
Concatenate all words of the question that are part of a same MWE (same algorithm than with quotations)
Idea 2
Improve the NER
Idea 3
Collapse the final normal form using MWE:
collapse ((A, B, ?), C, ?) to (A, B+C, ?) if there exists b in B and c in C such that b c is an existing MWE // only if all b in B or all c in C do not exist in our database.
See: https://github.com/ProjetPP/PPP-QuestionParsing-Grammatical/issues/122, https://github.com/ProjetPP/PPP-QuestionParsing-Grammatical/issues/64, https://github.com/ProjetPP/PPP-QuestionParsing-Grammatical/issues/85
Ex:
Idea 1
Idea 2
Improve the NER
Idea 3
Collapse the final normal form using MWE:
Ex:
((Barack Obama, birth,?),date,?)
>(Barack Obama, birth ,date,?)