For languages where the MWT words exactly make up the text of the token, build the pieces of the MWT using the text from the original token we are splitting. Should fix a bunch of the errors observed in https://github.com/stanfordnlp/stanza/issues/1371
For languages where the MWT words exactly make up the text of the token, build the pieces of the MWT using the text from the original token we are splitting. Should fix a bunch of the errors observed in https://github.com/stanfordnlp/stanza/issues/1371