Log-linear Combinations of Monolingual and Bilingual Neural Machine Translation Models for Automatic Post-Editing

TL;DR

This paper presents a neural model-based approach for automatic post-editing in machine translation, combining monolingual and bilingual models within a log-linear framework, and demonstrates significant improvements over baselines.

Contribution

It introduces a novel log-linear combination of neural translation models for post-editing, incorporating multiple inputs and a string-matching penalty to enhance translation quality.

Findings

Achieved -3.2% TER reduction on test set

Improved BLEU score by 5.5% over baseline

Outperformed all other systems in the shared task

Abstract

This paper describes the submission of the AMU (Adam Mickiewicz University) team to the Automatic Post-Editing (APE) task of WMT 2016. We explore the application of neural translation models to the APE problem and achieve good results by treating different models as components in a log-linear model, allowing for multiple inputs (the MT-output and the source) that are decoded to the same target language (post-edited translations). A simple string-matching penalty integrated within the log-linear model is used to control for higher faithfulness with regard to the raw machine translation output. To overcome the problem of too little training data, we generate large amounts of artificial data. Our submission improves over the uncorrected baseline on the unseen test set by -3.2\% TER and +5.5\% BLEU and outperforms any other system submitted to the shared-task by a large margin.