Learning Coreference Resolution for Portuguese Texts

José Guilherme Camargo de Souza, Patrícia Nunes Gonçalves, Renata Vieira

Abstract: This work presents the implementation of an automatic coreference resolution system based on supervised machine learning that is capable of processing any type of noun phrases for Portuguese. The system was trained and tested in a journalistic corpus formed by 50 texts with a total of 5047 markables. Both the induced classifier and the anaphoric clustering algorithm were evaluated using appropriate metrics. The clustering evalution was performed using the MUC and B 3 scorers.