Recent advances in automatic post-editing (APE) have shown that it is possible to automatically correct system-atic errors made by machine translation systems. However, most of the current APE techniques have only been tested in controlled batch environments, where training and test data are sampled from the same distribution and the training set is fully available. In this paper, we propose an online APE system based on an instance selection mechanism that is able to efficiently work with a stream of data points belonging to different domains. Our results on a mix of two datasets show that our system is able to: i)outperform state-of-the-art online APE solutions and ii) significantly improve the quality of rough MToutput.

Online Automatic Post-Editing across Domains

Rajen Chatterjee;Gebremedhen Gebremelak;Matteo Negri;MarcoTurchi
2016-01-01

Abstract

Recent advances in automatic post-editing (APE) have shown that it is possible to automatically correct system-atic errors made by machine translation systems. However, most of the current APE techniques have only been tested in controlled batch environments, where training and test data are sampled from the same distribution and the training set is fully available. In this paper, we propose an online APE system based on an instance selection mechanism that is able to efficiently work with a stream of data points belonging to different domains. Our results on a mix of two datasets show that our system is able to: i)outperform state-of-the-art online APE solutions and ii) significantly improve the quality of rough MToutput.
2016
9788899982546
File in questo prodotto:
File Dimensione Formato  
document (6).pdf

accesso aperto

Tipologia: Documento in Post-print
Licenza: DRM non definito
Dimensione 141.33 kB
Formato Adobe PDF
141.33 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: https://hdl.handle.net/11582/313147
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
social impact