no code implementations • NAACL (AmericasNLP) 2021 • Delfino Zacarías Márquez, Ivan Vladimir Meza Ruiz
The proposed system is based on the Transformer neural architecture and it uses sub-word level tokenization as the input.