Core Concepts
Translation techniques can be predicted to optimize machine translation processes.
Abstract
Abstract:
Machine translation aims to enhance accuracy.
Human-generated translations rely on diverse techniques.
Introduction:
Neural machine translation systems have made progress but fall short of human parity.
Issues like word-for-word translation persist, leading to low-quality translations.
Related Work:
Different approaches improve translation accuracy in NMT systems.
Post-editing involves correcting errors in MT output.
Data:
English-Chinese aligned pairs labeled with specific translation techniques are used.
Experiments:
Four experiments focus on from-scratch translation and post-editing tasks using different architectures and models.
Results:
Models achieve high accuracy in predicting suitable translation techniques for bad translations.
Stats
The predictive accuracy for from-scratch translation is 82%.
Post-editing process shows an accuracy rate of 93%.