Machine learning from user corrections is key to the industrial deployment of machine translation (MT). We introduce the first on-line approach to automatic post-editing (APE), i.e. the task of automatically correcting MT errors. We present experimental results of APE onEnglish-Italian MT by simulating human post-edits with human reference translations, and by applying online APE on MToutputs of increasing quality. By evaluating APE on generic vs. specialised and static vs. adaptive neural MT, we address the question: At what cost on the MT side will APE become useless?

Online Neural Automatic Post-editing for Neural Machine Translation

Matteo Negri;Marco Turchi;Nicola Bertoldi;Marcello Federico
2018

Abstract

Machine learning from user corrections is key to the industrial deployment of machine translation (MT). We introduce the first on-line approach to automatic post-editing (APE), i.e. the task of automatically correcting MT errors. We present experimental results of APE onEnglish-Italian MT by simulating human post-edits with human reference translations, and by applying online APE on MToutputs of increasing quality. By evaluating APE on generic vs. specialised and static vs. adaptive neural MT, we address the question: At what cost on the MT side will APE become useless?
File in questo prodotto:
File Dimensione Formato  
paper63 (1).pdf

accesso aperto

Licenza: Dominio pubblico
272.39 kB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11582/317665
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
social impact