Online Learning For Automatic Quality Estimation Of Machine Translation Output

Abstract

The automatic estimation of Machine Translation output quality is a hard task, where the selection of the appropriate algorithm and the most predictive features often plays a crucial role. When moving from controlled lab evaluations to real-life scenarios the task becomes even harder. For current Machine Translation Quality Estimation systems, additional complexity comes from the difficulty to model user and domain changes. Systems' instability with respect to data coming from different distributions, in fact, calls for adaptive solutions that quickly react to new operating conditions. To tackle this issue we propose an online framework for adaptive Quality Estimation, targeting reactivity and robustness to user and domain changes.We experiment with different online machine learning techniques like Online Support Vector Regression, Passive Aggressive Algorithms and Online Gaussian Processes. We also perform contrastive experiments with two language pairs, English-Spanish and English-Italian, in different testing conditions. The outcome of the experiments demonstrates the effectiveness of this approach.